---
title: "Ambient scribes win clinicians before proving outcomes"
description: "In randomized trials, ambient AI scribes reduce clinician burnout and save approximately 20 minutes a day or less. No trial measured a patient outcome, and insurers contest the coding gains that vendors sell."
type: research-report
publisher: Cyberfuturists
attribution: Cyberfuturists Independent Research
author: Ryan Marohn
date_published: 2026-11-07
evidence_current_to: 2026-11-03
language: en-GB
writing_standard: ASD-STE100 Simplified Technical English (writing rules)
canonical_url: https://cyberfuturists.com/research/ambient-scribes
markdown_url: https://cyberfuturists.com/research/ambient-scribes/full.md
summary_url: https://cyberfuturists.com/research/ambient-scribes.md
access: full report (public)
data_sources:
  - Aunoo news corpus, approximately March to October 2026
  - Web search of trade press, peer-reviewed abstracts, regulators and law-firm summaries, 3 November 2026
topics: [ambient scribe, ambient AI, clinical documentation, physician burnout, Epic AI Charting, Abridge, Microsoft Dragon Copilot, Heidi Health, coding intensity, MHRA, FDA, EU AI Act, CIPA, ambient sensing]
evidence_labels:
  vendor claim: A figure that a vendor reports. No independent check.
  single secondary source: Data from only one secondary source, for example a vendor blog or one summary of a primary document.
  analyst estimate: An estimate from an analyst firm.
---

# Ambient scribes win clinicians before proving outcomes

> In randomized trials, ambient AI scribes reduce clinician burnout and save approximately 20 minutes a day or less. No trial measured a patient outcome, and insurers contest the coding gains that vendors sell.

Cyberfuturists Independent Research. Analyst: Ryan Marohn. Published 7 November 2026. Evidence current to 3 November 2026.

## Notes for agents

- This file holds the full text of the report, written under the ASD-STE100 Simplified Technical English writing rules.
- The evidence is current to 3 November 2026, so events after that date are not in this file.
- Labels in square brackets show the quality of the evidence: [vendor claim], [single secondary source] and [analyst estimate]. When you repeat a labelled figure, keep its label.
- Do not present a vendor claim as an independent finding.
- Each claim has an inline link to its source. The "Claims ledger" section lists the main claims in a table, with the label and the source for each one.

## Key findings

- Scribes are already routine. About 28% of US physicians, and about 40% of GPs in the UK and Australia, use AI to write their notes.
- Clinicians who have a scribe use it less than those numbers suggest. In the largest trial, at UCLA, doctors used their scribe in only 30% to 34% of visits.
- Scribes help most with burnout. Randomized trials consistently show less burnout and exhaustion, but scribes save only about 20 minutes of note-writing a day, or less.
- The bigger time savings that health services and vendors quote, 59 minutes and "up to 60 minutes per day", come from comparisons without a control group.
- No trial has measured whether patients do better or worse. No country with reliable incident reporting has measured whether scribes are safe in everyday use. England's safety investigator, HSSIB, expects to report in summer 2027.
- When a VA study ran the same visits through 11 AI scribes, blinded reviewers rated the doctors' own notes higher in every case.
- Basic scribes are getting cheaper, and Epic now ships its own: AI Charting became generally available on 4 February 2026.
- Vendors now sell scribes on the promise of higher billing, which insurers see as extra cost. Blue Health Intelligence estimated about $2.3bn of added US healthcare costs linked to AI-enabled coding.
- In Great Britain, the US and the EU, a scribe that only drafts notes for a clinician to check is generally not a regulated medical device. Suggesting diagnoses, placing orders or saving notes without review moves it toward device rules.
- Most scribes record on the clinician's phone. An independent preprint found that, in a quiet room, the speech recognition software mattered more than the microphone. Microsoft says Teams-certified conference microphones also work with its Dragon Copilot scribe.
- In the US, patients are suing hospitals, not scribe vendors, over recording without consent (Sharp HealthCare, Sutter Health and MemorialCare). By early November 2026 we had found no report of an early ruling in these cases, such as on a motion to dismiss or on class certification.

## Summary

As of November 2026, almost all of the value that studies show for ambient AI in healthcare comes from one product: the ambient scribe. A scribe listens to a consultation and drafts the clinical note, which the clinician then checks. Clinicians have taken scribes up faster than the evidence has matured.

KLAS calls scribes the fastest-adopted health technology since the electronic health record (EHR). About 28% of US physicians and about 40% of UK and Australian GPs use AI for documentation. Epic released its own scribe in February 2026. Health services in England, New Zealand, Queensland and Spain buy scribes at scale, and Norway's largest regional health authority has opened a tender. Having a scribe is not the same as using it, though: in the UCLA trial, doctors used theirs in only 30% to 34% of visits.

The trials support the case on burnout, not the case on time. Randomized controlled trials (RCTs) consistently show less burnout and exhaustion, but scribes save only about 20 minutes of documentation a day, or less. The bigger figures that health services and vendors quote, 59 minutes and "up to 60 minutes per day", come from comparisons without a control group.

Nobody has shown whether scribes are good or bad for patients. No trial measured a patient outcome, and no country with reliable incident reporting has measured safety in everyday use. That gap is why England's safety investigator, HSSIB, opened a national investigation in August 2026.

Basic scribes are becoming cheap. Epic and other EHR vendors now build scribes into their products, and Microsoft cut its price in May 2026. A better note no longer wins the deal on its own, so independent vendors are adding coding, orders and automated tasks. Those extras earn the most money, and they also carry the most risk from regulators, insurers and the courts. The notes that move a hospital to higher billing codes are the same notes that insurers count as extra cost: an estimated $2.3bn in the US.

Hospitals are also signing contracts for other ambient tools, such as cameras that detect falls and smart rooms, but almost all of the evidence that these tools work comes from the vendors. Most scribes record on the clinician's phone. Dedicated microphones now compete in noisy, mobile and multi-speaker settings, but there is little independent evidence that they produce better notes.

We are watching how insurers respond to the higher billing, what HSSIB finds in summer 2027, and how US courts rule on consent.

## Method and evidence labels

We drew on two sources of data. The first is the Aunoo news corpus, which we searched for the period from approximately March to October 2026. The second is a web search on 3 November 2026 of trade press, peer-reviewed abstracts, regulators and law-firm summaries. Where the two sources disagree, we use the newer data.

The label **[vendor claim]** marks figures that vendors report. The label **[single secondary source]** marks data from only one secondary source, such as a vendor blog or one summary of a primary document that we could not open. The label **[analyst estimate]** marks estimates from analyst firms.

## Many clinicians have a scribe, but most use it in half their visits or fewer

Most health systems now have scribes, but most clinicians who have one use it in only some of their visits. In the UCLA trial, doctors used their scribe in only 30% to 34% of visits ([PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/)). In a JAMA study of five health systems, only 32% of clinicians who adopted a scribe used it in more than half of their visits ([EurekAlert](https://www.eurekalert.org/news-releases/1122192)). So the size of a deployment says little about how much its clinicians rely on the scribe.

KLAS calls ambient speech the fastest-adopted technology in healthcare since the EHR, and the numbers bear that out ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/the-fastest-adopted-tech-in-healthcare-report/)). By June 2025, **62% of Epic customers had installed an ambient or AI scribe**. In early 2025, more than 90% of health systems were using a scribe or evaluating one ([DH Insights](https://dhinsights.org/news/epic-shakes-up-the-ai-ambient-listening-market-with-integrated-option)). AMA data cited by the GAO show the share of US clinicians who use AI for documentation rising from 21% in 2024 to **28% in 2026** ([CarePilot summary of GAO-26-109116](https://carepilot.com/blog/gao-ai-medical-notes-coding); [GAO](https://www.gao.gov/products/gao-26-109116)).

Some deployments now cover a whole health system. Cleveland Clinic expanded a pilot of 250 physicians to 4,000 clinicians ([Business Insider](https://www.businessinsider.com/cleveland-clinic-ambient-ai-scribe-reducing-doctor-workload-2026-06)), and Mass General Brigham has approximately 3,000 regular users ([WBUR](https://www.wbur.org/upnext/2026/05/05/artificial-intelligence-primary-care-privacy)). Abridge reports that more than 300 health systems and more than 75 VA medical centres are its customers **[vendor claim; single secondary source]** ([New Market Pitch](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft)).

Even where every clinician has access, use varies widely from one clinician to the next. In the UW Health trial, ambient AI was used for 38% of notes ([PubMed 41625485](https://pubmed.ncbi.nlm.nih.gov/41625485/)).

### Outside the US, national and regional health services are the big buyers

| Market | Position as of November 2026 | Sources |
| :---- | :---- | :---- |
| United States | Approximately 28% of physicians use AI for documentation. Epic AI Charting became generally available on 4 Feb 2026. Abridge for Nurses is available in 250+ health systems. | [GAO summary](https://carepilot.com/blog/gao-ai-medical-notes-coding), [TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live), [Newsweek](https://www.newsweek.com/ai-nurses-health-care-abridge-access-health-11921461) |
| England | Approximately 40% of GPs use scribes. NHS England guidance (version 3, 29 Jul 2026) governs deployment. National adoption support starts with outpatients. Heidi is the sole supplier for NHS England Midlands **[vendor claim]**. | [BBC](https://www.bbc.co.uk/news/articles/cy0zylpdveqo), [MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide), [IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare) |
| Australia | GP use rose from 22% in August 2024 to 40% in November 2025 (RACGP). Metro South Health in Queensland deploys Heidi. | [The Guardian](https://www.theguardian.com/australia-news/2026/mar/29/doctors-using-ai-for-notes-australia), [MobiHealthNews](https://www.mobihealthnews.com/news/anz/queensland-health-services-ai-scribe-saves-doctors-59-minutes-daily) |
| New Zealand | Health NZ is deploying Heidi to all hospital emergency departments (EDs), covering 39 EDs and mental health services. | [RNZ](https://www.rnz.co.nz/news/health/589774/emergency-doctors-estimate-ai-scribe-heidi-saving-up-to-10-minutes-per-patient), [NZ Herald](https://www.nzherald.co.nz/business/health-new-zealands-ai-scribe-partner-heidi-raises-capital-at-154-billion-valuation/premium/C4DDY7UP6RE5XPMS65KGHED4WU/) |
| Canada | Alberta is testing its locally built Jenkins scribe in 81 EDs with 373 trained physicians. | [University of Alberta](https://www.ualberta.ca/en/folio/2026/03/ai-scribe-now-being-tested-in-eds-across-alberta.html) |
| Spain | An emergency network of 48 hospitals used a scribe in 1,032,558 of approximately 2.27m consultations (45.3%) between February 2025 and January 2026. Monthly use rose from 7.7% to 57.8%. | [PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/) |
| France, Nordics, DACH | Tandem reports that almost 200 French establishments are clients, and that it operates in 14 European countries **[vendor claim]**. Norway's largest regional health authority started a tender for an ambient scribe in January 2026 but has not published an award. | [FrenchWeb](https://www.frenchweb.fr/avec-865-millions-deuros-tandem-health-veut-passer-du-scribe-medical-au-systeme-dexploitation-clinique/463406), [Mercell](https://www.mercell.com/permalink/267602147.aspx) |
| Asia | Heidi has a company in Japan with in-country hosting, and a deployment in 27 hospitals with KPJ Healthcare in Malaysia **[vendor claim]**. India has a domestic scribe (EkaScribe), but no named hospital deployments. | [IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare), [eCorpIT](https://ecorpit.com/ambient-ai-medical-scribes-india-build-vs-buy-dpdp-2026/) |

## Trials show scribes reduce burnout but save 20 minutes a day or less, not the hour vendors claim

Scribes make clinicians less burned out, but they save less time than vendors claim. Every controlled study found less burnout or work exhaustion. None found much more than 20 minutes a day saved on notes, while uncontrolled reports claim up to an hour. Buyers who justify a scribe on time saved should use the trial figures.

Until late 2025, the evidence came mostly from quality-improvement projects and health systems' own reports. In one study of six health systems, burnout fell from 51.9% to 38.8% after 30 days ([HackerNoon](https://hackernoon.com/the-real-ambient-scribe-question-is-trust-not-note-quality)). Across approximately 1,400 physicians at Mass General Brigham and Emory, burnout fell by 21% ([Medical Daily](https://www.medicaldaily.com/ai-ambient-scribe-physician-burnout-mass-general-brigham-ucla-study-2026-475610)).

Since then, journals have published at least five RCTs, a large multisite observational study and a meta-analysis. They point the same way as the earlier studies, but the effects are smaller.

| Study | Design and size | Documentation time | Burnout and wellbeing |
| :---- | :---- | :---- | :---- |
| UCLA, NEJM AI, Dec 2025 | Pragmatic RCT with three arms: 238 physicians in 14 specialties. DAX, Nabla and usual care. | Nabla: −9.5% time in note (significant), approximately 23 seconds less per note than usual care. DAX: −1.7% (not significant). | Burnout improved by approximately 7% with both scribes. Work exhaustion fell. |
| UW Health, NEJM AI, Dec 2025 | Stepped-wedge RCT: 66 practitioners, 71,487 notes. | Approximately 22 minutes less each day on notes. | Work exhaustion fell by 0.44 on a 5-point scale (P<0.001). |
| JAMIA, May 2026 | Randomized crossover of two anonymised products: 160 clinicians. | Product B saved 3.19 minutes a day more than product A. | Both products reduced burnout, with no difference between them. After-hours time did not change. |
| Pediatric subspecialists, Appl Clin Inform, Aug 2025 | Small randomized QI trial: 23 clinicians. | No significant change. | Burnout among users fell from 89% to 56%. |
| JAMA, Apr 2026 | Observational: five health systems, 1,800+ adopters and 6,770 controls, over two years. | Approximately 16 minutes less documentation each day (approximately 10%). Approximately 13 minutes less EHR time (approximately 3%). | The authors say that the time saved probably does not explain the fall in burnout. |

Sources: [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/), [UCLA Health](https://www.uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time), [PubMed 41625485](https://pubmed.ncbi.nlm.nih.gov/41625485/), [PubMed 41729180](https://pubmed.ncbi.nlm.nih.gov/41729180/), [PubMed 40675605](https://pubmed.ncbi.nlm.nih.gov/40675605/), [EurekAlert](https://www.eurekalert.org/news-releases/1122192).

A JMIR meta-analysis of 21 studies found **an odds ratio of 0.47 for burnout prevalence**, which means scribes roughly halved the odds of burnout, and a **−0.35 reduction in work exhaustion**. But only two or three studies supplied data for each outcome. The authors rated their certainty as moderate for workload and low for burnout, and noted that all of the evidence came from early adopters who volunteered ([PubMed 42550089](https://pubmed.ncbi.nlm.nih.gov/42550089/)). STAT summed up the first trials this way: "doctors like the tech, and it saves a little time, but not a lot" ([STAT](https://www.statnews.com/2025/12/03/ambient-scribe-trial-doctor-burnout-ai-prognosis-newsletter/)).

### Trials agree more on burnout than on time saved

The JAMA authors found that the time saved does not explain the fall in burnout, so something else is helping, perhaps less mental strain or better contact with patients. None of the trials was blinded, though, so clinicians who expected to feel better may simply have reported feeling better.

### Results from one scribe do not apply to another

At UCLA, Nabla beat usual care by a significant margin but DAX did not, and the JAMIA crossover found one product better than the other. A trial result for one scribe says little about another.

### Health services report much bigger time savings than trials do

Metro South Health reports 59 minutes saved per doctor per day ([MobiHealthNews](https://www.mobihealthnews.com/news/anz/queensland-health-services-ai-scribe-saves-doctors-59-minutes-daily)), and Epic reports "up to 60 minutes per day" **[vendor claim]** ([TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live)). These figures come from heavy users and from comparisons without a control group.

### There is plenty of data from emergency departments and almost none from nursing

A US study of 275 attending physicians in 20 EDs found that notes written with a scribe took 2.4 fewer minutes of active work and were about 14% longer ([PubMed 42949694](https://pubmed.ncbi.nlm.nih.gov/42949694/)). In the Spanish network, consultations were 21.8% shorter, transcription accuracy was 93.9% and patients were more likely to recommend the service (net promoter score). The abstract does not name the vendor or declare conflicts of interest ([PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/)).

For nursing, the only peer-reviewed study looked at safety-incident reports in Singapore and found that the reports were more complete ([PubMed 42574744](https://pubmed.ncbi.nlm.nih.gov/42574744/)). A nursing RCT is registered, but it has no results yet ([NCT07456241](https://clinicaltrials.gov/study/NCT07456241)).

### Patients who have tried a scribe mostly accept it, and those who have not are wary

At Stanford, 73.6% of the 2,202 patients who had a scribe at a visit wanted one again. In Canada, 61.8% of 12,153 adults who had never had one were reluctant to try it ([KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/)). In the UCLA trial, fewer than 10% of patients refused the scribe ([UCLA Health](https://www.uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time)).

## Nobody tracks whether scribes harm patients, and their errors are hard for doctors to spot

Nobody is counting harm from scribes in everyday use, so the lack of reported harm tells us nothing. The kind of error matters too. In a study funded by Heidi, most AI errors were invented details, such as medications or examination findings that never came up, while most clinicians' errors were things left out. A busy doctor checking a note is less likely to spot something added than something missing, so clinician review is a weaker safeguard than it looks.

### No harm has been reported, because nobody is looking

No trial has measured patient outcomes such as diagnostic accuracy, medication errors, readmissions or deaths, and none was large enough to detect safety problems. The largest trial, at UCLA, recorded **one mild adverse event** in more than 48,000 visits with a scribe. Physicians said that clinically significant inaccuracies occurred "occasionally", and most of these were omissions and wrong pronouns ([PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/)).

The best evidence from everyday use is the 20-ED study, which found **no increase in 72-hour ED returns** for visits with a scribe ([PubMed 42949694](https://pubmed.ncbi.nlm.nih.gov/42949694/)). But a patient coming back to the ED within three days is only a rough sign of harm.

On 6 August 2026, England's Health Services Safety Investigations Body (HSSIB) started a national investigation into ambient voice technology in acute adult care. HSSIB said that adoption is fast, that hospitals are rolling scribes out with more attention to efficiency than to safety, and that incident reporting is "not yet mature" enough to pick up new risks. HSSIB expects to publish its findings in summer 2027, and its announcement does not cite a specific incident of harm ([Bird & Bird](https://biotalk.twobirds.com/post/102nh3r/ai-scribes-in-the-nhs-from-compliance-questions-to-formal-investigation)).

We found no published analysis of malpractice claims involving scribe notes. Most scribes are sold as software that is not a medical device, so problems with them fall outside the FDA's MAUDE database of device incidents. The ECRI list of the top health technology hazards for 2026 puts chatbot misuse first and does not mention scribes ([Healthcare Dive](https://www.healthcaredive.com/news/ecri-health-tech-hazards-2026/810223/)). So when no harm is reported, we read that as nobody measuring, not as proof of safety.

### Independent reviewers rated doctors' notes above every AI scribe's notes

A VA study in Annals of Internal Medicine (April 2026) ran recordings of five scripted primary care visits through **11 AI scribe tools** and compared the results with notes written by 18 clinicians. Thirty blinded raters gave **higher scores to the human notes in all cases**. The AI notes were weakest on thoroughness, organisation and usefulness. The study did not name the vendors ([Newswise/ACP](https://www.newswise.com/articles/ai-scribe-tools-produce-lower-quality-medical-notes-compared-to-human-clinicians); [Annals](https://www.acpjournals.org/doi/10.7326/ANNALS-25-02772)).

A preprint that Heidi funded, written by its own employees, found the opposite. It used 385 paired consultations with actors in five European countries. It found at least one critical or high-severity error in 6.2% of AI notes against 21.8% of notes by junior and middle-grade clinicians **[vendor claim]** ([medRxiv](https://www.medrxiv.org/content/10.64898/2026.08.18.26360701v1.full)). We think the gap between the two studies comes mostly from who Heidi compared against (less senior doctors), which product it tested (its own) and who paid for the study (Heidi).

### Scribes sometimes invent details, which doctors are less likely to notice than something left out

The Heidi preprint is still useful for what it shows about the kind of error. Most AI errors were invented or unsupported content, such as medications or examination findings that never came up, while most clinician errors were things left out. Scribes leave things out too: across studies, omissions made up 71% to 83% of scribe errors ([Forbes](https://www.forbes.com/sites/jessepines/2026/03/06/medical-ai-scribes-are-everywhere-research-shows-benefits--risks/)).

For a busy doctor, invented content is the harder error to catch, because they have to notice that something in the note never happened. At HIMSS, Medicomp's CEO warned that unchecked output can put unsupported diagnoses into the permanent medical record ([Healthcare IT News](https://www.healthcareitnews.com/news/health-it-leaders-must-stay-top-ai-delves-clinical-workflows)).

### Scribes miss context and make more mistakes with non-native speakers, dialects and complex specialties

A University of Edinburgh review of 27 papers found that scribes miss gestures and emotional state, and give more weight to clinical facts than to the patient's own account. The review also found "few empirical studies examining the risks" in the literature ([BBC](https://www.bbc.co.uk/news/articles/cy0zylpdveqo); [PubMed 42877726](https://pubmed.ncbi.nlm.nih.gov/42877726/)).

Whisper, a widely used speech recognition model, made significantly more errors on clinical English spoken by non-native speakers. Passing the transcript through a large language model recovered much of the lost accuracy ([PubMed 41772044](https://pubmed.ncbi.nlm.nih.gov/41772044/)). In a Dutch comparison of three commercial systems, conversations in dialect scored lowest of the test scenarios (3.77 of 5). Across specialties, surgical oncology scored much lower than ENT (3.68 against 4.36), and completeness was the weakest area ([PubMed 42887000](https://pubmed.ncbi.nlm.nih.gov/42887000/)).

We found no real-world study of patients with limited English, or of visits with an interpreter. Patients who know that AI is listening may hold back information about substance use, abuse or mental health ([News-Medical](https://www.news-medical.net/news/20260903/Ambient-AI-scribes-may-risk-missing-vital-patient-information.aspx)). One study found that with a scribe, neuropsychiatric symptoms were recorded more often, yet were linked to fewer psychiatric interventions ([MedPage Today](https://www.medpagetoday.com/opinion/second-opinions/122622)).
## Epic now includes a scribe in its EHR, which pushes down prices for vendors that sell only a scribe

Epic now builds a scribe into its EHR, so a hospital that runs Epic can use one without buying a separate product. A better note alone no longer wins the deal: being inside the hospital's EHR and costing less now count for more. Independent vendors have kept their Epic customers so far, but those customers expect Epic's scribe to cost less, and that pressure falls hardest on vendors that sell nothing but a scribe. Microsoft gains either way, because Epic's scribe runs on Microsoft's speech technology.

### Epic is big enough to extend its scribe into orders and diagnoses

Epic released AI Charting, its built-in scribe, to all customers on **4 February 2026**, after announcing it at its user group meeting in August 2025. It drafts notes and suggests orders during the visit. Microsoft's Dragon technology turns the speech into text and works out who is speaking, and Epic writes the final note ([Digital Health](https://www.digitalhealth.net/?p=238603); [Healthcare Dive](https://www.healthcaredive.com/news/epic-rolls-out-ai-charting-art-notetaking-documentation-scribe/811462/); [STAT](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/)).

At least seven organisations started using it at launch, and early adopters reported approximately 26% less documentation after hours **[vendor claim]** ([TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live)). By August 2026, FMOL Health, one of the first five outpatient users, was also using it in its EDs ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/inside-health-systems-early-use-of-epics-newest-ai-tools/)).

Epic's chief medical officer said the tool should extend to orders and diagnoses. Epic is well placed to do that, because it holds 42% of the US acute hospital EHR market and 55% of beds ([STAT](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/)). Epic has not published a price or licence terms.

### Independent vendors are keeping their Epic customers so far, but face lower prices

So far, independent vendors are not losing many Epic customers. KLAS interviewed 35 Epic customers in September 2025, and two-thirds reacted positively to Epic's scribe. Most expect to adopt it within two years, but many plan to keep other scribes for particular specialties alongside it. They also expect Epic's product to cost less ([HIT Consultant](https://hitconsultant.net/2025/09/29/klas-report-epics-native-ambient-speech-tool-reshapes-customer-ai-strategies/)), so the risk to independent vendors is lower prices more than lost customers.

We found no dated report of a named health system that replaced Abridge, DAX or Ambience with Epic. Abridge won Best in KLAS 2026 for ambient speech with a score of 94.7 ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/best-in-klas-2026-whos-winning-in-ambient-ai-ehrs-revenue-cycle-and-more/)). A secondhand summary of the October 2026 KLAS report says that Abridge leads purchase decisions, while Microsoft's performance, as customers report it, "fell sharply" **[single secondary source]** ([New Market Pitch](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft)). An undated analyst note says that Epic has equity in Abridge and gets a share of revenue **[single secondary source]** ([Sacra](https://sacra.com/c/abridge/)).

### Microsoft makes money from Epic's scribe, and its rivals do not

Microsoft sells its scribe technology as the speech engine inside Epic's scribe, as Dragon Copilot sold directly to hospitals, and through athenahealth ([Business Wire](https://secure.businesswire.com/news/home/20251209452332/en/athenahealth-Collaborates-with-Microsoft-to-Offer-Microsoft-Dragon-Copilot-as-Ambient-Notes-Option-Enhancing-Choice-and-Flexibility-for-Ambulatory-Clinicians)). So Microsoft earns money from Epic's move, while Abridge, Ambience and Suki do not.

The last widely cited share estimate is out of date. Menlo Ventures published it in October 2025 and put the ambient scribe market at **$600m**. It gave Microsoft 33%, Abridge 30%, Ambience 13%, Suki 10%, Nabla 4% and Freed 4% ([New Market Pitch summary of Menlo](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft)). That estimate predates both the Epic release and Microsoft's price cut, and we found no share data for 2026.

### Investors pay more for vendors that sell more than a scribe

Investors put large sums into scribe companies in 2026, and they value companies that do more than scribing far more highly than those that do not.

| Company | 2026 position | Source |
| :---- | :---- | :---- |
| Commure | $70m in May 2026 at a $7bn post-money valuation. Its ambient documentation comes from Augmedix, inside a broader operations platform. | [HLTH](https://hlth.com/insights/news/commure-raises-70m-to-expand-ai-healthcare-operations-platform) |
| Abridge | Valuation of $5.3bn in June 2025. Only an analyst firm, and a blog that cites it, report a $316m extension in April 2026 **[single secondary source]**. Estimated ARR of approximately $100m in May 2025 **[analyst estimate]**. | [KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/), [Sacra](https://sacra.com/c/abridge/) |
| Heidi Health | US$340m in September 2026: a US$100m Series C and a US$240m growth investment from the General Catalyst Customer Value Fund. Reported valuation of US$900m. The NZ Herald figure of $1.54bn is most probably the same valuation in New Zealand dollars. ARR of US$50m in April 2026 **[vendor claim]**. | [HLTH](https://hlth.com/insights/news/heidi-raises-340m-to-expand-clinical-ai-into-supervised-agentic-workflows), [Business News Australia](https://www.businessnewsaustralia.com/amp-html/articles/heidi-edges-closer-to-unicorn-status-with-us340m-raise-valuing-ai-health-company-at-us900m.html), [NZ Herald](https://www.nzherald.co.nz/business/health-new-zealands-ai-scribe-partner-heidi-raises-capital-at-154-billion-valuation/premium/C4DDY7UP6RE5XPMS65KGHED4WU/) |
| Suki | $90m Series E in October 2026. The valuation is not public. 450+ organisations. An unnamed source gives projected 2026 revenue of $60m to $80m. | [Healthcare Dive](https://www.healthcaredive.com/news/suki-raises-90m-in-series-e-round-to-create-ai-agents/832862/), [AI Weekly](https://aiweekly.co/alerts/suki-raises-90m-series-e-led-by-healthier-capital-at-450-customer-scale) |
| Tandem Health | $100m Series B in September 2026, led by the EQT Scaleup Europe Fund. $160m raised in total. | [HLTH](https://hlth.com/insights/news/tandem-health-raises-100m-to-expand-clinical-ai-across-europe) |
| Ambience | We found no 2026 round. $243m Series C in July 2025 at $1.25bn. | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/ambience-healthcare-reaches-1-25b-valuation/) |

Investors pay more for companies that go beyond drafting notes into revenue cycle (billing and claims) and hospital operations. Commure, which sells a broad operations platform, raised money at $7bn. Heidi is valued at approximately 18 times its reported annual recurring revenue (ARR). Suki and Heidi are both spending their new money on automated tasks after the visit, such as referrals and scheduling. That fits a market where the basic scribe is becoming a standard EHR feature. A vendor that sells nothing but a scribe is selling the part of the market that is losing value.

## Scribes are getting cheaper, so vendors pitch them on higher billing, and insurers are pushing back

Scribes are getting cheaper, and the extra revenue they bring does not clearly pay for an expensive licence. So vendors now sell them on a different promise: better notes that support higher billing codes, and so more revenue. Insurers see that as inflated costs, and they already have software that cuts claims back to lower codes. A hospital that buys a scribe for the extra billing is betting on revenue that insurers can take back.

### Prices vary widely, and every recent move has been down

Prices vary widely. Freed sells a self-serve product at $99 a month, and Nabla sells at $119 a month. The Ambience scribe costs an estimated $2,800 to $3,200 per clinician a year, or $4,000 to $5,000 with its add-on for improving clinical documentation. DAX Copilot was estimated at $600 a month **[analyst estimates]** ([Sacra on Abridge](https://sacra.com/c/abridge/); [Sacra on Ambience](https://sacra.com/c/ambience/)).

Every recent price move has been downward. From 1 May 2026, Microsoft cut the per-user price of Dragon Copilot in the US, Canada, the UK, Ireland, Germany, Austria, Switzerland, France and the Benelux countries. Microsoft did not state the new price, and it also dropped its separate offer for small practices. All of these details come from one source **[single secondary source, a Microsoft reseller]** ([Schneider IT Management](https://www.schneider.im/microsoft-dragon-copilot-price-decrease-licensing-changes/)). WellSky gives its scribe to current customers free of charge ([KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/)), and KLAS found that Epic customers expect Epic's scribe to cost less.

### The extra visits pay for a cheap scribe, but not clearly for an expensive one

There is little evidence that scribes pay for themselves. The JAMA multisite study found approximately **0.5 more visits each week and approximately $167 more revenue for each clinician each month** ([EurekAlert](https://www.eurekalert.org/news-releases/1122192)). That covers a cheap product, but not clearly a licence of $600 a month. In March 2025, the Peterson Health Technology Institute (PHTI) reached a similar view: scribes reduce burnout but do not yet improve health systems' finances ([Axios](https://www.axios.com/2025/03/27/ai-scribes-reduce-burnout-financial-improvement)).

### Vendors pitch higher billing instead, and one study found less accurate coding

So vendors sell a different benefit: notes that support higher billing codes. Ambience prices its documentation pilots on a verified **$4.50 revenue lift per visit** **[analyst estimate]** ([Sacra](https://sacra.com/c/ambience/)). In the UW Health trial, the diagnosis codes used for billing improved ([PubMed 41625485](https://pubmed.ncbi.nlm.nih.gov/41625485/)). An orthopaedic study of 7,236 visits found that notes from the scribe were billed at higher levels than dictated notes (3.24 against 2.96 on the E/M scale) but coded less accurately (68% against 74%). It modelled approximately $232,000 a year of extra revenue ([PubMed 42671176](https://pubmed.ncbi.nlm.nih.gov/42671176/)).

### Insurers see the same higher billing as extra cost

Trilliant Health studied six health systems that had announced scribe roll-outs, and at all six, more visits were billed at the two highest levels. For new patients, the share of visits billed at the highest levels rose by 12 to 20 percentage points ([Trilliant](https://www.trillianthealth.com/market-research/studies/outpatient-coding-intensity-increases-as-hospitals-adopt-ai-enabled-scribing)). Blue Health Intelligence estimated approximately **$2.3bn of added US healthcare costs** linked to AI-enabled coding. Its estimate includes births coded for anaemia after heavy bleeding, where many of the patients never received a transfusion ([TechTarget](https://www.techtarget.com/revcyclemanagement/feature/How-AI-scribes-are-shifting-coding-intensity-reimbursement)).

Caroline Pearson of PHTI agreed with Trilliant's main point but criticised its method ([STAT](https://www.statnews.com/2026/04/08/are-scribes-raising-health-care-costs-ai-prognosis/)). Trilliant says part of the rise corrects years of under-billing. Both can be true, and insurers and CMS, which runs Medicare, will settle the argument.

### US federal agencies have raised concerns but have not changed payment rules

So far, the US government has studied the issue and issued warnings, but has not changed how it pays. CMS released its proposed 2027 payment rule for physicians (the CY2027 Physician Fee Schedule) on 14 July 2026. It asks for comments on "the appropriate role of ambient documentation tools" in physician payment, but proposes no change ([Holland & Knight](https://www.hklaw.com/en/insights/publications/2026/07/cms-issues-cy-2027-medicare-physician-fee-schedule-proposed-rule); [CMS](https://www.cms.gov/newsroom/fact-sheets/calendar-year-cy-2027-medicare-physician-fee-schedule-proposed-rule)). CMS released the final rule on 4 November 2026 ([AHA](https://www.aha.org/news/headline/2026-11-04-cms-issues-cy-2027-physician-fee-schedule-final-rule)). On 7 November, we checked two summaries of the final rule, and neither mentions a payment policy for ambient documentation tools ([AHA](https://www.aha.org/news/headline/2026-11-04-cms-issues-cy-2027-physician-fee-schedule-final-rule); [Medical Economics](https://www.medicaleconomics.com/view/medicare-s-final-2027-fee-schedule-cuts-physician-pay-more-than-proposed-but-softens-the-same-day-visit-penalty)). We did not review the full text of the final rule.

The HHS inspector general (HHS-OIG) lists risk adjustment, the system that pays Medicare Advantage plans more for sicker patients, as a top enforcement concern. Its compliance guidance, published in February 2026, does not prohibit AI. It does warn against software that prompts clinicians to add payment-raising diagnoses that patients do not have ([Parker Poe](https://www.parkerpoe.com/news/2026/06/oig-guidance-warns-physician-practices-of-compliance-risks)).

In a technology review published in July 2026, the GAO found few independent studies of scribe accuracy. It said that some products do not keep the recordings needed to check a note, and it warned that errors could lead to overpayment or underpayment. It made no recommendations **[single secondary source for its content]** ([CarePilot summary](https://carepilot.com/blog/gao-ai-medical-notes-coding)).

### Private insurers already cut claims back to lower billing codes

Cigna, Aetna and several Blue Cross plans already use software that automatically lowers visit billing codes, although none of these policies targets AI ([JUCM](https://www.jucm.com/documents/2026/03/what-to-know-about-payers-downcoding-procedures.pdf/)). A hospital that justifies a scribe on higher billing is therefore counting on revenue that insurers have both the tools and the stated intention to cut.
## Scribes that only draft notes for a doctor to check are generally not regulated as medical devices, but suggesting diagnoses or placing orders changes that

Regulators in the US, Great Britain and the EU all ask the same question: what is the product for? A scribe that transcribes, summarises or drafts notes for a clinician to check is usually not a medical device. One that suggests diagnoses, places orders or saves notes without review usually is. The features vendors are adding to earn more money are exactly the ones that cross that line.

### Great Britain: the regulator draws the line at diagnosis and at notes saved without review

The MHRA, Britain's medicines and medical devices regulator, gave the clearest statement on **29 July 2026**, in guidance on ambient voice products that sets out nine worked examples. Our UK details in this section come from one secondary summary **[single secondary source]**.

These five examples are not devices:

* Transcripts.
* Summaries for clinician review.
* Structured data for the clinician to confirm.
* Codes matched to terms the clinician actually said.
* Draft letters.

These four examples are devices:

* Generated "insights."
* Summaries that a vendor markets as a guide to diagnosis or treatment.
* Records saved without proper review, and orders placed automatically.
* Suggested diagnoses meant to be used without checking the reasoning.

A scribe that counts as a device and supports diagnosis directly falls under Rule 10 (Class IIa), which needs assessment by a UK Approved Body. Other device scribes fall under Rule 12 (Class I). A "not for diagnosis" disclaimer does not help if the vendor makes medical claims elsewhere ([MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide)).

NHS buyers have lost a shortcut they relied on. Two days after the guidance, the Royal College of Physicians (RCP) objected that it goes against NHS practice, which expected suppliers to have MHRA Class I registration before the NHS listed them. That shows buyers were using Class I registration as an entry test, and the guidance removes any regulatory basis for it.

For a scribe that only writes drafts, Class I registration says nothing about whether the product is better. Commure got Class I registration with the MHRA in September 2026 ([MENAFN](https://menafn.com/1111615428/Commure-Ambient-AI-Achieves-Class-I-Certification-And-MHRA-Registration)).

Whatever a scribe's device status, NHS England puts duties on the organisation that deploys it. Version 3 of its guidance, published the same day as the MHRA's, requires a Clinical Safety Officer, DCB0160 clinical safety documentation and a data protection impact assessment. The AVT Self-Certified Supplier Registry remains only a buying aid. We did not open the primary gov.uk and NHS England pages **[single secondary source]** ([MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide); [NHS England guidance](https://www.england.nhs.uk/long-read/guidance-on-the-use-of-ai-enabled-ambient-scribing-products-in-health-and-care-settings/)).

### US: the FDA steps in when software drives diagnosis or clinical decisions

The FDA updated its guidance on clinical decision support software on 6 January 2026. It still oversees software that "effectively drives diagnosis or immediate clinical action", and it treats software whose reasoning is harder to see as more likely to be a device. The summaries we reviewed do not mention scribes, so our view that the FDA does not oversee scribes that only draft notes is our own inference ([Faegre Drinker](https://www.faegredrinker.com/en/insights/publications/2026/1/key-updates-in-fdas-2026-general-wellness-and-clinical-decision-support-software-guidance)).

### EU: stricter rules for AI in medical devices start in 2028, but people must already be told when they deal with AI

The Digital Omnibus on AI, Regulation (EU) 2026/1744, has been in force since 27 July 2026. It delayed the AI Act's high-risk obligations: those in Annex III now start on **2 December 2027**, and those for AI in medical devices (under the MDR and IVDR) on **2 August 2028**. The Article 50 transparency duties, which require people to be told when they are dealing with AI, started on 2 August 2026 ([COSM](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices); [Morgan Lewis](https://www.morganlewis.com/pubs/2026/06/eu-approves-delays-and-other-amendments-to-certain-eu-ai-act-obligations-what-businesses-should-know)).

Older dates are still circulating. A Corti press release in May 2026 gave 2 August 2026 as the start date for high-risk requirements for medical AI ([TNGlobal](https://technode.global/prnasia/corti-steps-up-its-support-for-the-next-wave-of-healthcare-startups-as-openai-pushes-into-the-space-and-others-retreat-from-europe/)). That date was right before the EU adopted the omnibus, but it is wrong now. We took the regulation number and the vote counts from one source and have not checked them against EUR-Lex, the EU's official law database.

So a scribe that only drafts notes, and is not a medical device, probably falls outside the high-risk rules but inside the Article 50 disclosure duty.

### US states: the consent risk comes from wiretapping laws, not from laws about scribes

Most US state laws on AI disclosure cover either AI in patient communications (California AB 3030) or AI for diagnosis and treatment. Texas TRAIGA, an example of the second kind, has been in force since 1 January 2026, with penalties of $10,000 to $200,000 for each violation ([Akerman](https://www.akerman.com/en/perspectives/hrx-new-year-new-ai-rules-healthcare-ai-laws-now-in-effect.html)). It is not clear whether a scribe that only drafts notes counts as AI for diagnosis or treatment under Texas law. No state has a consent law written for scribes. The legal risk on consent comes instead from wiretap laws that require everyone in a conversation to agree to a recording, such as the California Invasion of Privacy Act (CIPA).

### The features vendors are adding bring scribes closer to device regulation

The current rules favour a scribe that only writes drafts, but vendors are investing in suggested diagnoses, order entry and automated follow-up tasks. Each of these moves a product toward device classification in Great Britain, toward FDA oversight in the US, and toward the EU's high-risk AI rules from 2028. Heidi has already kept its new "supervised task" features out of the UK and EU, without saying why ([IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare); [iatroX](https://www.iatrox.com/blog/heidis-coming-ai-capabilities-exclude-the-uk-and-eu-an-opening-for-tortus)).

## Patients are suing hospitals, not scribe vendors, over recording without consent, and the only ruling so far favoured a clinic

In the US, the lawsuits over scribes target the hospitals that use them, not the vendors. Neither California case has a ruling yet. The one ruling so far, in Washington, kept a recording away from a patient, but it says nothing about getting consent before recording. So a hospital's legal risk depends on how it asks patients for consent.

### California: two proposed class actions over recordings sent to Abridge

The first case came in **November 2025**, when Jose Saucedo filed a proposed class action against Sharp HealthCare in San Diego Superior Court over the use of Abridge without consent ([WTL Governance](https://wtlgovernance.com/insights/updates/sutter-memorialcare-abridge-ai-transcription-consent-lawsuit/); [Medscape](https://www.medscape.com/viewarticle/health-system-sued-over-ai-scribe-technology-patient-consent-2026a10001k7)). The complaint alleges that more than 100,000 patients were recorded, and that consent statements went into charts automatically **[single secondary source, a vendor blog, for these details]** ([Basil AI](https://basilai.app/articles/2026-06-05-ambient-ai-scribes-abridge-sutter-memorialcare-sharp-lawsuits-cipa-false-consent.html)).

A second proposed class action followed on **9 April 2026**, against **Sutter Health and MemorialCare** in the Northern District of California. It alleges violations of CIPA sections 631 and 632 and, according to reports, of the federal wiretap law and the California medical confidentiality law. The complaint alleges that the health systems had no standard way of getting consent from everyone in the room before recordings went to Abridge's servers ([WTL Governance](https://wtlgovernance.com/insights/updates/sutter-memorialcare-abridge-ai-transcription-consent-lawsuit/); [Ars Technica](https://arstechnica.com/tech-policy/2026/04/californians-sue-over-ai-tool-that-records-doctor-visits/)).

The complaint names Abridge as the scribe, but the sources we reviewed do not list Abridge as a defendant. One source refers to a Sutter complaint and a separate MemorialCare complaint, so these may be two filings. By early November 2026, we had found no report of any ruling in either case, whether on dismissing it, combining the cases or letting one go ahead as a class action.

### Washington: a court refused a patient access to the scribe's recording

In *Raphael v. Mantei* (Clark County Superior Court, Washington, August 2026), a patient of The Vancouver Clinic asked for the Nuance DAX recording of her telehealth visit. The judge held that the recording "is exempt from disclosure under the Uniform Healthcare Information Act." The judge accepted that the recording is a working aid for drafting the note, like dictation. The AMA Litigation Center and the state medical and hospital associations filed in support of the clinic ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/patients-cant-access-ai-scribe-recordings-washington-judge-rules/)).

The ruling helps providers that keep recordings only briefly and limit patients' access to them. It does nothing about a failure to get consent before recording, which is what the California cases are about. It also sits uneasily with the GAO's concern that products which delete recordings leave nothing to check a note against.

### Outside the US: some clinicians refuse patients who will not accept a scribe

Digital Rights Watch reported that some Australian clinicians refused to treat patients who did not accept a scribe. In one case, a psychiatrist in Melbourne turned away new patients who would not agree to one ([The Guardian](https://www.theguardian.com/australia-news/2026/may/19/melbourne-psychiatrist-ai-note-taking-new-patients); [AAP](https://aapnews.aap.com.au/news/doctors-ai-usage-raising-privacy-and-safety-concerns)). Digital Rights Watch campaigns for a legal right to refuse a scribe ([New Atlas](https://newatlas.com/medical-ai/ai-medical-scribe-consent-raises-patient-rights-concerns/)). Some Australian practices treat a poster in the waiting room as consent ([The Guardian](https://www.theguardian.com/australia-news/2026/mar/29/doctors-using-ai-for-notes-australia)). In the UK, a response published in the BMJ argued that vendors who reuse audio to train their models may become joint controllers of the data under UK GDPR, sharing legal responsibility for it ([BMJ](https://www.bmj.com/content/390/bmj-2025-085754/rr)).
## European vendors report large user numbers, but little published data backs them up

European scribe vendors have raised large sums, but almost every usage figure comes from the vendors themselves. The largest European deployment with published data is a Spanish emergency network of 48 hospitals. We found no national programme like NHS England's in Germany, France, the Netherlands, Spain or Italy, and no verified figures on use in German hospitals. We also found no independent European study of outcomes, so a buyer on the Continent has only vendor numbers to go on.

### The big user numbers come from the vendors themselves

Tandem (Stockholm) raised a $100m Series B in September 2026 and connects to more than 100 other systems, including Doctolib, Epic and Cerner. Tandem says its users cut administrative time by 29% **[vendor claim]** ([FrenchWeb](https://www.frenchweb.fr/avec-865-millions-deuros-tandem-health-veut-passer-du-scribe-medical-au-systeme-dexploitation-clinique/463406); [HLTH](https://hlth.com/insights/news/tandem-health-raises-100m-to-expand-clinical-ai-across-europe)).

Nabla (Paris) reports more than 85,000 clinicians and 20m consultations a year **[vendor claim]**, and it was one of the two products in the UCLA trial. Corti (Copenhagen) describes itself as the technology underneath other scribe companies and says it serves more than 100m patients a year **[vendor claim]** ([TNGlobal](https://technode.global/prnasia/corti-steps-up-its-support-for-the-next-wave-of-healthcare-startups-as-openai-pushes-into-the-space-and-others-retreat-from-europe/)).

Doctolib reaches more than 400,000 health professionals in France and Germany, yet it publishes no usage figures for the consultation assistant it launched in 2024. It announced a €20m clinical AI lab in February 2026 ([Caducée.net](https://www.caducee.net/actualite-medicale/16814/doctolib-accelere-dans-l-ia-clinique-avec-un-laboratoire-dedie.html)). Heidi has deployments with Swiss Medical Network and in NHS England Midlands ([IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare)). In the NHS, TORTUS competes with Heidi and Accurx ([iatroX](https://www.iatrox.com/blog/ambient-scribing-uk-gps-tortus-vs-heidi-vs-accurx-2026-comparison)).

### Norway has a regional tender, not a deployment

In continental Europe, public buying happens region by region. In January 2026, Sykehusinnkjøp opened a tender (a competitive dialogue) for a "speech to summary" ambient scribe on behalf of Helse Sør-Øst, Norway's largest regional health authority. No award has been published, so Norway has an open tender but no deployment yet ([Mercell](https://www.mercell.com/permalink/267602147.aspx)).

### Local languages are hard for scribes, which helps vendors that handle them well

Local languages are hard for scribes, so a vendor that handles them well has something to sell. In January 2026, Speechmatics released a Swedish medical model and named compound words, dialects and pitch accent as particular difficulties. Speechmatics says its model gets 3.91% of medical keywords wrong **[vendor claim]** ([Speechmatics](https://www.speechmatics.com/company/articles-and-news/speechmatics-launches-new-swedish-medical-model-cutting-transcription-errors)). In the Dutch comparison of three vendors, conversations in dialect scored lowest ([PubMed 42887000](https://pubmed.ncbi.nlm.nih.gov/42887000/)).

In India, scribes built for the US are expected to struggle with consultations that switch between Hindi, English, Tamil and Bengali. The Digital Personal Data Protection Act makes the hospital legally responsible for patient data (the "data fiduciary"). It also requires consent notices in several languages, with penalties of up to INR 250 crore for major breaches. All of these points about India come from one guide **[single secondary source, a vendor-adjacent guide]** ([eCorpIT](https://ecorpit.com/ambient-ai-medical-scribes-india-build-vs-buy-dpdp-2026/)).

## Hospitals are signing large contracts for room cameras and sensors, but every reported result comes from vendors and lacks a comparison group

Hospitals are signing system-wide contracts for cameras and room sensors, but every result we found comes from a vendor, and none was measured against a comparison group. Buyers should treat the reported falls reductions of 47% to 78% as the best case, not as what to expect.

Beyond scribes, hospitals use cameras to detect falls and to let staff watch patients remotely, which is called virtual nursing or virtual sitting. Some are also building smart rooms that combine cameras, voice and location tracking. Our news collection had no reports of named hospital deployments, but our web search found several.

### UChicago Medicine and Stryker show room sensing moving into large deals

**On 28 April 2026, UChicago Medicine signed a system-wide agreement with Artisight.** The agreement covers more than 1,800 devices in patient rooms, operating theatres, recovery units and a cancer facility that opens in April 2027. HIT Consultant's headline says 1,800 patient rooms while its article counts devices, so reports do not agree on what was counted. The platform combines cameras, voice, location tracking, virtual nursing, automated EHR documentation and tracking of each stage of an operation ([HIT Consultant](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/); [Fierce Healthcare](https://www.fiercehealthcare.com/hospitals/uchicago-medicine-artisight-partner-smart-hospital-platform-deployment)).

Artisight also runs Epic MyChart Bedside on the televisions in patient rooms ([Rutland Herald/BusinessWire](https://www.rutlandherald.com/news/business/artisight-now-runs-epic-mychart-bedside-tv-powering-a-fully-connected-smart-hospital-experience/article_52e91483-a03f-5d82-9079-f8bb503486d7.html)). Stryker's purchase of care.ai puts ambient monitoring into a large portfolio of beds and devices, because care.ai sells a station for virtual nursing and ambient monitoring ([Medical Device Network](https://www.medicaldevice-network.com/news/stryker-acquires-smart-hospital-tech-provider-care-ai/)).

Artisight cites a **78% reduction in falls** and four times more on-time discharges from unnamed "previous implementations" **[vendor claim]** ([HIT Consultant](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/)). Teton.ai reports a **47% fall reduction** in one Danish neurology department, compared with a one-month baseline **[vendor claim]**. Teton had no comparison group, and its own report says other changes in the department at the same time may have played a part ([Teton](https://www.teton.ai/blog/nhl-noh-report)).

A US trial of cameras for preventing falls is registered, but we could not find its status or results ([NCT06339125](https://clinicaltrials.gov/study/NCT06339125)).

### Nurses say virtual nursing adds to bedside workload

Nurses at Providence Alaska said virtual nursing will put more work on bedside staff and worsen nurse-to-patient ratios. National Nurses United won contract terms on AI, and in October 2024 nurses at 17 HCA hospitals ratified AI safeguards ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/nurses-push-back-on-healthcare-tech.html)). Camera privacy in operating theatres was an issue in strike negotiations ([OR Manager](https://ormanager.com/briefs/or-camera-privacy-concerns-influence-nursing-strike-negotiations)).

### The questions hospitals face with scribes will also apply to cameras

For buyers, the bigger change is that scribes and room sensing are merging. Artisight already runs documentation and monitoring on the same room hardware, so the questions scribes raise about consent, how long recordings are kept and medical device rules will also apply to cameras. And a patient has fewer ways to opt out of a monitored room than to opt out of a recorded consultation.
## Dedicated microphones have to beat the phone doctors already carry, and no study shows they do

A company selling dedicated microphones for scribes has two hurdles. The scribe vendors have to approve its device, and the device has to produce more accurate notes than the clinician's phone, which costs nothing extra. Microsoft recommends the phone, and the only independent study found that the speech recognition software mattered more than the microphone. The best openings are where phones struggle: emergency departments, nursing, and rooms with several people talking.

### Microphone makers need approval from scribe vendors and proof that their devices improve notes

Scribe and EHR vendors decide which microphones work with their products, so getting onto their approved lists is the main way in. Microsoft points to Teams conference microphones for Dragon Copilot. Philips lists Dragon Copilot, Dragon Medical One and Tandem Health, while Heidi sells its own device for its own software.

A dedicated device has to show that it makes notes measurably more accurate, because the phone costs nothing extra and is already in the clinician's pocket. The only independent study found that the microphone made little difference in a quiet room, which is not enough to justify buying one. Scribe prices are also falling, which leaves less budget per clinician for extra hardware.

Recording is also moving away from the individual clinician. In hospital rooms, built-in cameras and microphones can replace wearable devices, and video-visit platforms record through their own equipment.

We are watching for:

- independent tests of each device in real clinics that measure the quality of the final note;
- published hardware lists from scribe and EHR vendors, including Epic;
- adoption data for Heidi Remote and the Philips SpeechMike Ambient;
- any regulator or NHS guidance on recording hardware.

### Microsoft recommends the phone and sets strict requirements for other microphones

Most scribes record the consultation on the clinician's phone. Microsoft says the Dragon Copilot mobile app on a phone "provides optimal audio quality" for ambient recording ([Microsoft Support](https://support.microsoft.com/en-us/dragon-copilot/physicians/select-and-configure-your-microphone)). It adds that the mobile app "offers the best ambient recording experience" and can record offline ([Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/)). Research does the same: a registered Abridge trial records visits on the EHR's mobile app ([NCT07598721](https://clinicaltrials.gov/study/NCT07598721)).

Microsoft also allows other microphones, but sets a high bar. Dictation microphones such as PowerMic and SpeechMike pick up sound from one direction, while recording a conversation needs a microphone that picks up sound from all directions. Its technical requirements include a condenser microphone, a USB connection, a signal-to-noise ratio above 70 dB and 24-bit/96 kHz recording, and it does not recommend Bluetooth. It says the microphones it recommends for Microsoft Teams conference calls "are also suitable for ambient recording in Dragon Copilot" ([Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/)).

### In the only independent study, the speech recognition software mattered more than the microphone

The only independent test found that the microphone made little difference. A preprint from the University of California, Irvine (May 2026) recorded 35 re-enacted primary care visits on five devices at once. The devices were a smartphone, a laptop microphone, a portable recorder, a clip-on microphone and a desktop microphone. It then ran the recordings through six speech recognition engines ([medRxiv](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full)). The microphone made a statistically significant difference with every engine, but only a small one with AWS, Azure and Whisper, and the desktop microphone made the fewest word errors. The authors wrote: "Engine choice produced larger variation in transcription performance than recording device" ([medRxiv](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full)).

That result has limits. The study recorded in a quiet room, did not measure the quality of the notes, and has not been peer reviewed. An opinion piece in npj Digital Medicine says current scribes cannot reliably tell several speakers apart, even in a quiet room, and that some settings may need portable, high-quality microphones ([npj Digital Medicine](https://www.nature.com/articles/s41746-026-02554-0)).

Hardware vendors report larger effects. In a January 2026 Philips benchmark, the SpeechMike Ambient had the highest speech recognition accuracy in both quiet and noisy conditions, while smartphones were reliable only in quiet conditions **[vendor claim]** ([Pacific Transcription](https://shop.pacifictranscription.com.au/blogs/dictation-solutions-2026/check-out-the-new-philips-speechmike-ambient-wearable-ai-assistant)). We found no published study that compares a dedicated microphone with a phone on the quality of the final clinical note.

### Wearable microphones from Philips and Heidi compete with microphones built into rooms and video visits

| Recording option | Examples | Strength | Limit | Sources |
| :---- | :---- | :---- | :---- | :---- |
| Clinician's phone | Dragon Copilot mobile app, EHR mobile apps | No extra hardware. Microsoft recommends it for ambient recording. | A personal device, and its position changes from visit to visit. | [Microsoft Support](https://support.microsoft.com/en-us/dragon-copilot/physicians/select-and-configure-your-microphone), [eDist](https://www.edist.com/become-a-reseller/philips-reseller/philips-speechmike-ambient-for-dragon-copilot/) |
| Wearable microphone | Philips SpeechMike Ambient, Heidi Remote | Stays near the speaker and moves with the clinician. | Extra cost for each clinician. Little independent evidence. | [Philips](https://www.dictation.philips.com/us/philips-speechmike-ambient-wearable-ai-assistant-psm5000-series/), [Heidi](https://www.heidihealth.com/blog/heidi-launches-hardware) |
| Room or conference microphone | Teams-certified conference microphones, Artisight All-In-One Device | Picks up everyone in the room. Microsoft publishes requirements. | Fixed position. Room platforms can add cameras. | [Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/), [Medical Design and Outsourcing](https://www.medicaldesignandoutsourcing.com/medtronic-recor-oculogica-and-more-make-times-best-inventions-list/) |
| Video-visit platform | Teladoc SoloScribe | Recording is built into the video-visit platform. | Quality depends on the patient's and clinician's equipment. | [Fierce Healthcare](https://fiercehealthcare.com/ai-and-machine-learning/teladoc-health-unveils-2-new-ai-capabilities-within-solo-platform) |

The Philips SpeechMike Ambient has four directional (beamforming) microphones, and Philips lists support for Dragon Copilot (Flex) and Dragon Medical One ([Philips](https://www.dictation.philips.com/us/philips-speechmike-ambient-wearable-ai-assistant-psm5000-series/)). A Philips comparison page also lists Tandem Health ([Philips SpeechLive](https://www.speechlive.com/ca/content-hub/blog/philips-speechmike-ambient-vs-heidi-remote/)). A US reseller lists the device at $497 ([StartStop](https://www.startstop.com/philips-psm5000-00-speechmike-ambient-wearable-ai-assistant/)), and resellers market it as a device that IT teams can "standardize, charge, assign, and support" ([eDist](https://www.edist.com/become-a-reseller/philips-reseller/philips-speechmike-ambient-for-dragon-copilot/)).

Heidi built its own device to take the phone out of the consultation. It launched Heidi Remote, a 21-gram clip-on device with one button, in March 2026 ([Forbes Australia](https://www.forbes.com.au/news/entrepreneurs/australian-ai-startup-heidi-health-launches-its-first-hardware-device/); [Medical Republic](https://www.medicalrepublic.com.au/heidis-hardware-play-takes-aim-at-apple-sized-problem-in-clinical-ai/124330)). Heidi Remote stores encrypted audio on the device and needs neither a phone nor a connection during the visit ([iatroX](https://www.iatrox.com/blog/heidi-remote-dedicated-ai-scribe-hardware-device-clinical-documentation-2026)). The Medical Republic, an Australian publication, reports a price of $290 ([Medical Republic](https://www.medicalrepublic.com.au/heidis-hardware-play-takes-aim-at-apple-sized-problem-in-clinical-ai/124330)). Heidi says that "the weakest link has always been the device in the room" **[vendor claim]** ([Heidi](https://www.heidihealth.com/blog/heidi-launches-hardware)). Availability varies from country to country.

Room and video-visit platforms build recording into their own hardware. The Artisight All-In-One Device, on the TIME list of Best Inventions of 2026, uses sensors and cameras to update medical records automatically ([Medical Design and Outsourcing](https://www.medicaldesignandoutsourcing.com/medtronic-recor-oculogica-and-more-make-times-best-inventions-list/)). Teladoc added the SoloScribe ambient scribe to its Solo virtual care platform in October 2026. Teladoc said other products ask hospitals to manage "one more piece of technology," and that its clients want the opposite ([Fierce Healthcare](https://fiercehealthcare.com/ai-and-machine-learning/teladoc-health-unveils-2-new-ai-capabilities-within-solo-platform)).

### Demand is growing in emergency departments and nursing, where the phone is weakest

Scribes are now moving into settings with noise, movement and many speakers. Emergency departments in New Zealand, Alberta and Spain use scribes at national or regional scale. In the Spanish network, transcription accuracy was 93.9% ([PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/)), which means approximately one word in sixteen was wrong.

Scribes for nurses reached the market in 2026. A nurse moves between patients, so a phone left in one place cannot follow the conversation. Abridge made its nursing product available to all of its health system clients in May 2026 ([Newsweek](https://www.newsweek.com/ai-nurses-health-care-abridge-access-health-11921461)). Oracle Health made its Clinical AI Agent for nurses commercially available in the US in September 2026 ([HIT Consultant](https://hitconsultant.net/2026/09/14/oracle-health-launches-clinical-ai-agent-for-nurses-inpatient-ehr-documentation/)). In an April 2026 KLAS report, eight of nine leaders were "highly satisfied" with the Abridge nursing product ([Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/hows-ambient-ai-for-nurses-an-early-review/)).

Transcription errors end up in the patient's record, so better recording has a direct payoff. On 31 August 2026, Healthwatch England warned that AI scribes get the names of drugs and diagnoses wrong ([The Guardian](https://www.theguardian.com/society/2026/aug/31/doctors-ai-scribes-get-names-of-drugs-and-diagnoses-wrong-nhs-watchdog-warns)). Healthwatch found 27 different scribe products in use in England, and patients spotted some of the errors before clinicians did ([The Next Web](https://thenextweb.com/news/nhs-ai-scribes-errors-healthwatch-mhra)). Accuracy also falls for non-native speakers and for dialects ([PubMed 41772044](https://pubmed.ncbi.nlm.nih.gov/41772044/); [PubMed 42887000](https://pubmed.ncbi.nlm.nih.gov/42887000/)).

## Conclusion

Ambient scribes have won clinicians over, but they have not yet earned premium prices. Since late 2025, randomised trials have given a clearer picture, and the effects are smaller than earlier studies suggested. Scribes reliably make clinicians feel less exhausted, and they save a little time. That explains why adoption is so fast: clinicians resist having the scribe taken away. But the time saved is too small to justify premium prices on its own.

Epic, Microsoft and WellSky are cutting the price of basic scribes, so independent vendors now make their case on higher billing and automated follow-up tasks. Insurers dispute the extra billing, and in Great Britain and the US those extra features also bring a product under medical device rules. Heidi has already kept them out of the UK and EU. For anything beyond drafting notes, scribes will be worth paying for only if vendors can give regulators and insurers auditable proof that what the note records actually happened.

Our view of where the money is would change if insurers or Medicare act against the higher billing. That could come in later CMS rules, although the summaries of the 2027 final rule mention none. It could also come from private insurers using their existing tools to lower billing codes at hospitals that use scribes heavily.

Our view of the risk would change with the first safety data from everyday use, because today's evidence cannot tell a safe technology apart from one that nobody is monitoring. That data could come from HSSIB's findings in 2027, the Nuffield Trust's Phase 2 evaluation for the NHS, or an independent benchmark of each vendor.

And the first ruling in the California consent cases will show whether the hundreds of hospitals that use scribes must change how they ask for consent. Because platforms such as Artisight combine scribes with room sensing, every one of these answers will soon apply to cameras as well as microphones.
## Claims ledger

Each row is one claim from the report. "Peer-reviewed" marks a study published in a peer-reviewed journal, and only this table and the explorer page use that marker.

| # | Section | Claim | Label | Source |
| ---: | :--- | :--- | :--- | :--- |
| 1 | Adoption | KLAS called ambient speech the fastest-adopted technology in healthcare since the EHR. |  | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/the-fastest-adopted-tech-in-healthcare-report/) |
| 2 | Adoption | By June 2025, 62% of Epic customers had installed an ambient or AI scribe. |  | [DH Insights](https://dhinsights.org/news/epic-shakes-up-the-ai-ambient-listening-market-with-integrated-option) |
| 3 | Adoption | In early 2025, more than 90% of health systems were using a scribe or evaluating one. |  | [DH Insights](https://dhinsights.org/news/epic-shakes-up-the-ai-ambient-listening-market-with-integrated-option) |
| 4 | Adoption | AMA data cited by the GAO show the share of US clinicians who use AI for documentation rising from 21% in 2024 to 28% in 2026. |  | [CarePilot summary of GAO-26-109116](https://carepilot.com/blog/gao-ai-medical-notes-coding) |
| 5 | Adoption | Cleveland Clinic grew a 250-physician pilot to 4,000 clinicians. |  | [Business Insider](https://www.businessinsider.com/cleveland-clinic-ambient-ai-scribe-reducing-doctor-workload-2026-06) |
| 6 | Adoption | Mass General Brigham has approximately 3,000 regular users. |  | [WBUR](https://www.wbur.org/upnext/2026/05/05/artificial-intelligence-primary-care-privacy) |
| 7 | Adoption | Abridge reports more than 300 health systems and more than 75 VA medical centres as customers. | vendor claim; single secondary source | [New Market Pitch](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft) |
| 8 | Adoption | In the UCLA trial, doctors with a scribe used it in only 30% to 34% of visits. | peer-reviewed | [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/) |
| 9 | Adoption | In the UW Health trial, ambient AI was used for 38% of notes. | peer-reviewed | [PubMed 41625485](https://pubmed.ncbi.nlm.nih.gov/41625485/) |
| 10 | Adoption | In a JAMA study of five health systems, only 32% of clinicians who adopted a scribe used it in more than half of their visits. | peer-reviewed | [EurekAlert (JAMA)](https://www.eurekalert.org/news-releases/1122192) |
| 11 | Adoption | Approximately 40% of GPs in England use scribes. |  | [BBC](https://www.bbc.co.uk/news/articles/cy0zylpdveqo) |
| 12 | Adoption | Australian GP use rose from 22% in August 2024 to 40% in November 2025 (RACGP). |  | [The Guardian](https://www.theguardian.com/australia-news/2026/mar/29/doctors-using-ai-for-notes-australia) |
| 13 | Adoption | Health NZ is rolling Heidi out to all hospital EDs, covering 39 EDs and mental health services. |  | [RNZ](https://www.rnz.co.nz/news/health/589774/emergency-doctors-estimate-ai-scribe-heidi-saving-up-to-10-minutes-per-patient) |
| 14 | Adoption | Alberta is testing its locally built Jenkins scribe in 81 EDs with 373 trained physicians. |  | [University of Alberta](https://www.ualberta.ca/en/folio/2026/03/ai-scribe-now-being-tested-in-eds-across-alberta.html) |
| 15 | Adoption | A Spanish emergency network of 48 hospitals used a scribe in 1,032,558 of approximately 2.27m consultations (45.3%), February 2025 to January 2026. | peer-reviewed | [PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/) |
| 16 | Adoption | In the Spanish network, monthly use rose from 7.7% to 57.8%. | peer-reviewed | [PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/) |
| 17 | Adoption | Heidi is the sole supplier for NHS England Midlands. | vendor claim | [IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare) |
| 18 | Adoption | Tandem reports almost 200 French establishments as clients and operations in 14 European countries. | vendor claim | [FrenchWeb](https://www.frenchweb.fr/avec-865-millions-deuros-tandem-health-veut-passer-du-scribe-medical-au-systeme-dexploitation-clinique/463406) |
| 19 | Adoption | Heidi reports a deployment in 27 hospitals with KPJ Healthcare in Malaysia. | vendor claim | [IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare) |
| 20 | Evidence of benefit | UCLA trial (238 physicians): Nabla cut the time spent writing each note by 9.5% (significant), while DAX cut it by 1.7% (not significant). | peer-reviewed | [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/) |
| 21 | Evidence of benefit | UCLA trial: burnout improved by approximately 7% with both scribes. | peer-reviewed | [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/) |
| 22 | Evidence of benefit | UW Health trial: approximately 22 minutes a day less on notes, and work exhaustion fell by 0.44 points on a 5-point scale. | peer-reviewed | [PubMed 41625485](https://pubmed.ncbi.nlm.nih.gov/41625485/) |
| 23 | Evidence of benefit | JAMIA trial (160 clinicians, each trying both products): both reduced burnout, and product B saved 3.19 minutes a day more than product A. | peer-reviewed | [PubMed 41729180](https://pubmed.ncbi.nlm.nih.gov/41729180/) |
| 24 | Evidence of benefit | Trial in paediatric specialists (23 clinicians): no significant change in documentation time, but burnout among users fell from 89% to 56%. | peer-reviewed | [PubMed 40675605](https://pubmed.ncbi.nlm.nih.gov/40675605/) |
| 25 | Evidence of benefit | JAMA study of five health systems: approximately 16 minutes a day less on documentation. The authors say the time saved probably does not explain the fall in burnout. | peer-reviewed | [EurekAlert (JAMA)](https://www.eurekalert.org/news-releases/1122192) |
| 26 | Evidence of benefit | JMIR meta-analysis of 21 studies: scribes roughly halved the odds of burnout (odds ratio 0.47 for burnout prevalence). The authors rate their certainty on burnout as low, and all of the evidence came from early adopters who volunteered. | peer-reviewed | [PubMed 42550089](https://pubmed.ncbi.nlm.nih.gov/42550089/) |
| 27 | Evidence of benefit | Metro South Health reports 59 minutes saved per doctor each day, from a comparison without a control group. |  | [MobiHealthNews](https://www.mobihealthnews.com/news/anz/queensland-health-services-ai-scribe-saves-doctors-59-minutes-daily) |
| 28 | Evidence of benefit | Epic reports "up to 60 minutes per day" saved with AI Charting. | vendor claim | [TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live) |
| 29 | Evidence of benefit | A US study of 275 attending physicians in 20 EDs: notes written with a scribe took 2.4 fewer minutes of active work and were approximately 14% longer. | peer-reviewed | [PubMed 42949694](https://pubmed.ncbi.nlm.nih.gov/42949694/) |
| 30 | Evidence of benefit | Spanish emergency network: consultations were 21.8% shorter and transcription accuracy was 93.9%. | peer-reviewed | [PubMed 42456601](https://pubmed.ncbi.nlm.nih.gov/42456601/) |
| 31 | Evidence of benefit | The only peer-reviewed nursing study, on safety-incident reports in Singapore, found the reports were more complete. | peer-reviewed | [PubMed 42574744](https://pubmed.ncbi.nlm.nih.gov/42574744/) |
| 32 | Evidence of benefit | At Stanford, 73.6% of 2,202 patients who had a scribe at a visit wanted one again. |  | [KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/) |
| 33 | Evidence of benefit | In Canada, 61.8% of 12,153 adults who had never had a scribe were reluctant to try one. |  | [KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/) |
| 34 | Evidence of benefit | In the UCLA trial, fewer than 10% of patients refused the scribe. |  | [UCLA Health](https://www.uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time) |
| 35 | Safety and accuracy | No trial has measured a patient outcome, and none was large enough to detect safety problems. | peer-reviewed | [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/) |
| 36 | Safety and accuracy | The UCLA trial recorded one mild adverse event in more than 48,000 visits with a scribe. | peer-reviewed | [PubMed 41497288](https://pubmed.ncbi.nlm.nih.gov/41497288/) |
| 37 | Safety and accuracy | The 20-ED study found no increase in patients returning to the ED within 72 hours after a visit with a scribe. | peer-reviewed | [PubMed 42949694](https://pubmed.ncbi.nlm.nih.gov/42949694/) |
| 38 | Safety and accuracy | HSSIB, England's safety investigator, started a national investigation into ambient voice technology in acute adult care on 6 August 2026 and expects to publish findings in summer 2027. |  | [Bird & Bird](https://biotalk.twobirds.com/post/102nh3r/ai-scribes-in-the-nhs-from-compliance-questions-to-formal-investigation) |
| 39 | Safety and accuracy | The ECRI list of the top health technology hazards for 2026 does not mention scribes. |  | [Healthcare Dive](https://www.healthcaredive.com/news/ecri-health-tech-hazards-2026/810223/) |
| 40 | Safety and accuracy | A VA study ran five scripted visits through 11 AI scribe tools, and 30 blinded raters scored the clinicians' notes higher in all cases. | peer-reviewed | [Annals of Internal Medicine](https://www.acpjournals.org/doi/10.7326/ANNALS-25-02772) |
| 41 | Safety and accuracy | A preprint funded by Heidi found a critical or high-severity error in 6.2% of AI notes and 21.8% of notes by junior and middle-grade clinicians. | vendor claim | [medRxiv](https://www.medrxiv.org/content/10.64898/2026.08.18.26360701v1.full) |
| 42 | Safety and accuracy | In the Heidi preprint, most AI errors were invented or unsupported content, while most clinician errors were things left out. | vendor claim | [medRxiv](https://www.medrxiv.org/content/10.64898/2026.08.18.26360701v1.full) |
| 43 | Safety and accuracy | Across studies, omissions made up 71% to 83% of scribe errors. |  | [Forbes](https://www.forbes.com/sites/jessepines/2026/03/06/medical-ai-scribes-are-everywhere-research-shows-benefits--risks/) |
| 44 | Safety and accuracy | An Edinburgh review of 27 papers found that scribes miss gestures and emotional state. | peer-reviewed | [PubMed 42877726](https://pubmed.ncbi.nlm.nih.gov/42877726/) |
| 45 | Safety and accuracy | Whisper, a speech recognition model, made significantly more errors on clinical English spoken by non-native speakers. | peer-reviewed | [PubMed 41772044](https://pubmed.ncbi.nlm.nih.gov/41772044/) |
| 46 | Safety and accuracy | In a Dutch comparison of three systems, conversations in dialect scored lowest of the test scenarios (3.77 of 5). Across specialties, surgical oncology scored 3.68 and ENT 4.36. | peer-reviewed | [PubMed 42887000](https://pubmed.ncbi.nlm.nih.gov/42887000/) |
| 47 | Market and Epic | Epic released AI Charting to all customers on 4 February 2026. It runs on Microsoft's Dragon speech technology. |  | [STAT](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/) |
| 48 | Market and Epic | Early adopters of Epic AI Charting reported approximately 26% less documentation after hours. | vendor claim | [TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live) |
| 49 | Market and Epic | Epic has 42% of the US acute hospital EHR market and 55% of beds. |  | [STAT](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/) |
| 50 | Market and Epic | KLAS interviewed 35 Epic customers in September 2025: two-thirds reacted positively, and most expect to adopt Epic's scribe within two years. |  | [HIT Consultant](https://hitconsultant.net/2025/09/29/klas-report-epics-native-ambient-speech-tool-reshapes-customer-ai-strategies/) |
| 51 | Market and Epic | Abridge won Best in KLAS 2026 for ambient speech with a score of 94.7. |  | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/best-in-klas-2026-whos-winning-in-ambient-ai-ehrs-revenue-cycle-and-more/) |
| 52 | Market and Epic | A secondhand summary of KLAS's October 2026 report says Abridge leads in purchase decisions. | single secondary source | [New Market Pitch](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft) |
| 53 | Market and Epic | An undated analyst note says Epic has equity in Abridge and gets a share of revenue. | single secondary source | [Sacra](https://sacra.com/c/abridge/) |
| 54 | Market and Epic | Menlo Ventures (October 2025) put the ambient scribe market at $600m, with Microsoft at 33% and Abridge at 30%. |  | [New Market Pitch summary of Menlo](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft) |
| 55 | Market and Epic | Commure raised $70m in May 2026 at a $7bn post-money valuation. |  | [HLTH](https://hlth.com/insights/news/commure-raises-70m-to-expand-ai-healthcare-operations-platform) |
| 56 | Market and Epic | Only an analyst firm and one blog report a $316m Abridge extension in April 2026. | single secondary source | [Sacra](https://sacra.com/c/abridge/) |
| 57 | Market and Epic | Abridge's annual recurring revenue (ARR) was approximately $100m in May 2025. | analyst estimate | [Sacra](https://sacra.com/c/abridge/) |
| 58 | Market and Epic | Heidi raised US$340m in September 2026 at a reported valuation of US$900m. |  | [HLTH](https://hlth.com/insights/news/heidi-raises-340m-to-expand-clinical-ai-into-supervised-agentic-workflows) |
| 59 | Market and Epic | Heidi's annual recurring revenue (ARR) was US$50m in April 2026. | vendor claim | [HLTH](https://hlth.com/insights/news/heidi-raises-340m-to-expand-clinical-ai-into-supervised-agentic-workflows) |
| 60 | Market and Epic | Suki raised a $90m Series E in October 2026 and has more than 450 customer organisations. |  | [Healthcare Dive](https://www.healthcaredive.com/news/suki-raises-90m-in-series-e-round-to-create-ai-agents/832862/) |
| 61 | Market and Epic | Tandem raised a $100m Series B in September 2026, in a round led by the EQT Scaleup Europe Fund. |  | [HLTH](https://hlth.com/insights/news/tandem-health-raises-100m-to-expand-clinical-ai-across-europe) |
| 62 | Market and Epic | Ambience raised a $243m Series C in July 2025 at $1.25bn. We found no 2026 round. |  | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/ambience-healthcare-reaches-1-25b-valuation/) |
| 63 | Prices and insurers | Freed sells a self-serve product at $99 a month, and Nabla at $119 a month. |  | [Sacra](https://sacra.com/c/abridge/) |
| 64 | Prices and insurers | The Ambience scribe has an estimated price of $2,800 to $3,200 per clinician per year. | analyst estimate | [Sacra on Ambience](https://sacra.com/c/ambience/) |
| 65 | Prices and insurers | The estimate for DAX Copilot was $600 a month. | analyst estimate | [Sacra](https://sacra.com/c/abridge/) |
| 66 | Prices and insurers | Microsoft cut the Dragon Copilot per-user price from 1 May 2026 but did not state the new price. | single secondary source | [Schneider IT Management](https://www.schneider.im/microsoft-dragon-copilot-price-decrease-licensing-changes/) |
| 67 | Prices and insurers | WellSky offers its scribe free of charge to current customers. |  | [KEYT/Stacker](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/) |
| 68 | Prices and insurers | PHTI (March 2025) found that scribes reduce burnout but do not yet improve health systems' finances. |  | [Axios](https://www.axios.com/2025/03/27/ai-scribes-reduce-burnout-financial-improvement) |
| 69 | Prices and insurers | The JAMA multisite study found approximately $167 more revenue per clinician per month. | peer-reviewed | [EurekAlert (JAMA)](https://www.eurekalert.org/news-releases/1122192) |
| 70 | Prices and insurers | Ambience prices its documentation pilots on a verified $4.50 revenue lift per visit. | analyst estimate | [Sacra](https://sacra.com/c/ambience/) |
| 71 | Prices and insurers | An orthopaedic study of 7,236 visits found that scribe notes were billed at higher levels than dictated notes (3.24 against 2.96 on the E/M scale) but coded less accurately (68% against 74%). | peer-reviewed | [PubMed 42671176](https://pubmed.ncbi.nlm.nih.gov/42671176/) |
| 72 | Prices and insurers | At all six health systems it studied, Trilliant Health found more visits billed at the two highest levels. |  | [Trilliant](https://www.trillianthealth.com/market-research/studies/outpatient-coding-intensity-increases-as-hospitals-adopt-ai-enabled-scribing) |
| 73 | Prices and insurers | Blue Health Intelligence estimated approximately $2.3bn of added US healthcare costs linked to AI-enabled coding. |  | [TechTarget](https://www.techtarget.com/revcyclemanagement/feature/How-AI-scribes-are-shifting-coding-intensity-reimbursement) |
| 74 | Prices and insurers | CMS's proposed 2027 payment rule for physicians (14 July 2026) asks for comments on ambient documentation tools but proposes no change. |  | [Holland & Knight](https://www.hklaw.com/en/insights/publications/2026/07/cms-issues-cy-2027-medicare-physician-fee-schedule-proposed-rule) |
| 75 | Prices and insurers | The AHA and Medical Economics summaries of the 2027 final rule (4 November 2026) mention no payment policy for ambient documentation tools. |  | [AHA](https://www.aha.org/news/headline/2026-11-04-cms-issues-cy-2027-physician-fee-schedule-final-rule) |
| 76 | Prices and insurers | HHS-OIG guidance (February 2026) warns against software that prompts clinicians to add payment-raising diagnoses. |  | [Parker Poe](https://www.parkerpoe.com/news/2026/06/oig-guidance-warns-physician-practices-of-compliance-risks) |
| 77 | Prices and insurers | The GAO (July 2026) found few independent studies of scribe accuracy and made no recommendations. | single secondary source | [CarePilot summary](https://carepilot.com/blog/gao-ai-medical-notes-coding) |
| 78 | Prices and insurers | Cigna, Aetna and several Blue Cross plans already use software that automatically lowers visit billing codes, but none of these policies targets AI. |  | [JUCM](https://www.jucm.com/documents/2026/03/what-to-know-about-payers-downcoding-procedures.pdf/) |
| 79 | Regulation | The MHRA's guidance of 29 July 2026 gives nine worked examples and says that a scribe that only drafts notes for a clinician to check is not a medical device. | single secondary source | [MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide) |
| 80 | Regulation | In Great Britain, a scribe that counts as a device and supports diagnosis directly is Class IIa (Rule 10) and needs a UK Approved Body. Other device scribes are Class I (Rule 12). | single secondary source | [MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide) |
| 81 | Regulation | NHS England guidance (version 3) requires a Clinical Safety Officer, DCB0160 clinical safety documentation and a data protection impact assessment. | single secondary source | [MedDeviceGuide](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide) |
| 82 | Regulation | The FDA's clinical decision support guidance (6 January 2026) keeps oversight of software that "effectively drives diagnosis or immediate clinical action." |  | [Faegre Drinker](https://www.faegredrinker.com/en/insights/publications/2026/1/key-updates-in-fdas-2026-general-wellness-and-clinical-decision-support-software-guidance) |
| 83 | Regulation | The EU Digital Omnibus delayed the rules for AI in medical devices (under the MDR and IVDR) to 2 August 2028. |  | [COSM](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices) |
| 84 | Regulation | Since 2 August 2026, Article 50 of the EU AI Act has required that people be told when they are dealing with AI. |  | [Morgan Lewis](https://www.morganlewis.com/pubs/2026/06/eu-approves-delays-and-other-amendments-to-certain-eu-ai-act-obligations-what-businesses-should-know) |
| 85 | Regulation | Texas TRAIGA has been in force since 1 January 2026, with penalties of $10,000 to $200,000 for each violation. |  | [Akerman](https://www.akerman.com/en/perspectives/hrx-new-year-new-ai-rules-healthcare-ai-laws-now-in-effect.html) |
| 86 | Regulation | Heidi kept its new "supervised task" features out of the UK and EU without saying why. |  | [IT Brief UK](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare) |
| 87 | Litigation and consent | Jose Saucedo filed a proposed class action against Sharp HealthCare in November 2025 over the use of Abridge without consent. |  | [WTL Governance](https://wtlgovernance.com/insights/updates/sutter-memorialcare-abridge-ai-transcription-consent-lawsuit/) |
| 88 | Litigation and consent | The Sharp complaint alleges more than 100,000 patients were recorded. | single secondary source | [Basil AI](https://basilai.app/articles/2026-06-05-ambient-ai-scribes-abridge-sutter-memorialcare-sharp-lawsuits-cipa-false-consent.html) |
| 89 | Litigation and consent | A proposed class action filed against Sutter Health and MemorialCare on 9 April 2026 alleges breaches of sections 631 and 632 of the California Invasion of Privacy Act (CIPA). |  | [Ars Technica](https://arstechnica.com/tech-policy/2026/04/californians-sue-over-ai-tool-that-records-doctor-visits/) |
| 90 | Litigation and consent | In Raphael v. Mantei (Washington, August 2026), a judge ruled that a patient's Nuance DAX recording is exempt from disclosure. |  | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/patients-cant-access-ai-scribe-recordings-washington-judge-rules/) |
| 91 | Litigation and consent | Digital Rights Watch reported that some Australian clinicians refused to treat patients who did not accept a scribe. |  | [New Atlas](https://newatlas.com/medical-ai/ai-medical-scribe-consent-raises-patient-rights-concerns/) |
| 92 | Litigation and consent | A response published in the BMJ argued that vendors who reuse audio to train their models may become joint controllers of the data under UK GDPR. |  | [BMJ](https://www.bmj.com/content/390/bmj-2025-085754/rr) |
| 93 | Europe | Tandem connects to more than 100 other systems and says its users cut administrative time by 29%. | vendor claim | [FrenchWeb](https://www.frenchweb.fr/avec-865-millions-deuros-tandem-health-veut-passer-du-scribe-medical-au-systeme-dexploitation-clinique/463406) |
| 94 | Europe | Corti says it serves more than 100m patients a year. | vendor claim | [TNGlobal](https://technode.global/prnasia/corti-steps-up-its-support-for-the-next-wave-of-healthcare-startups-as-openai-pushes-into-the-space-and-others-retreat-from-europe/) |
| 95 | Europe | Doctolib announced a €20m clinical AI lab in February 2026 but does not publish usage figures for its assistant. |  | [Caducée.net](https://www.caducee.net/actualite-medicale/16814/doctolib-accelere-dans-l-ia-clinique-avec-un-laboratoire-dedie.html) |
| 96 | Europe | Sykehusinnkjøp opened a tender for Helse Sør-Øst in January 2026, and no award has been published. |  | [Mercell](https://www.mercell.com/permalink/267602147.aspx) |
| 97 | Europe | Speechmatics says its Swedish medical model gets 3.91% of medical keywords wrong. | vendor claim | [Speechmatics](https://www.speechmatics.com/company/articles-and-news/speechmatics-launches-new-swedish-medical-model-cutting-transcription-errors) |
| 98 | Europe | Under India's data protection law (the DPDP Act), penalties for major breaches reach INR 250 crore. | single secondary source | [eCorpIT](https://ecorpit.com/ambient-ai-medical-scribes-india-build-vs-buy-dpdp-2026/) |
| 99 | Ambient sensing | UChicago Medicine signed a system-wide agreement with Artisight on 28 April 2026 that covers more than 1,800 devices. |  | [HIT Consultant](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/) |
| 100 | Ambient sensing | Artisight cites a 78% reduction in falls from unnamed previous implementations. | vendor claim | [HIT Consultant](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/) |
| 101 | Ambient sensing | Teton.ai reports a 47% fall reduction in one Danish neurology department, with no control group. | vendor claim | [Teton](https://www.teton.ai/blog/nhl-noh-report) |
| 102 | Ambient sensing | Stryker bought care.ai, which sells a station for virtual nursing and ambient monitoring. |  | [Medical Device Network](https://www.medicaldevice-network.com/news/stryker-acquires-smart-hospital-tech-provider-care-ai/) |
| 103 | Ambient sensing | In October 2024, nurses at 17 HCA hospitals ratified AI safeguards. |  | [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/nurses-push-back-on-healthcare-tech.html) |
| 104 | Recording hardware | Microsoft says the Dragon Copilot mobile app on a phone "provides optimal audio quality" for ambient recording. |  | [Microsoft Support](https://support.microsoft.com/en-us/dragon-copilot/physicians/select-and-configure-your-microphone) |
| 105 | Recording hardware | Microsoft says the microphones it recommends for conference calls in Microsoft Teams are also suitable for ambient recording in Dragon Copilot. |  | [Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/) |
| 106 | Recording hardware | For recording conversations, Microsoft requires a microphone that picks up sound from all directions (an omnidirectional condenser microphone), a USB connection, a signal-to-noise ratio above 70 dB and 24-bit/96 kHz recording. |  | [Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/) |
| 107 | Recording hardware | A UC Irvine preprint recorded 35 re-enacted visits on five devices and found that the choice of speech recognition software mattered more than the microphone. |  | [medRxiv](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full) |
| 108 | Recording hardware | In the same preprint, the desktop microphone made the fewest word errors, in recordings made in a quiet room. |  | [medRxiv](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full) |
| 109 | Recording hardware | In a January 2026 Philips benchmark, the SpeechMike Ambient had the highest accuracy in quiet and noisy conditions. | vendor claim | [Pacific Transcription](https://shop.pacifictranscription.com.au/blogs/dictation-solutions-2026/check-out-the-new-philips-speechmike-ambient-wearable-ai-assistant) |
| 110 | Recording hardware | The Philips SpeechMike Ambient has four directional (beamforming) microphones, and Philips lists support for Dragon Copilot (Flex) and Dragon Medical One. |  | [Philips](https://www.dictation.philips.com/us/philips-speechmike-ambient-wearable-ai-assistant-psm5000-series/) |
| 111 | Recording hardware | Heidi launched Heidi Remote, a 21-gram clip-on device that stores encrypted audio on the device, in March 2026. |  | [iatroX](https://www.iatrox.com/blog/heidi-remote-dedicated-ai-scribe-hardware-device-clinical-documentation-2026) |
| 112 | Recording hardware | The Artisight All-In-One Device, which uses sensors and cameras, is on the TIME list of Best Inventions of 2026. |  | [Medical Design and Outsourcing](https://www.medicaldesignandoutsourcing.com/medtronic-recor-oculogica-and-more-make-times-best-inventions-list/) |
| 113 | Recording hardware | Teladoc added the SoloScribe ambient scribe to its Solo virtual care platform in October 2026. |  | [Fierce Healthcare](https://fiercehealthcare.com/ai-and-machine-learning/teladoc-health-unveils-2-new-ai-capabilities-within-solo-platform) |
| 114 | Recording hardware | Oracle Health made its Clinical AI Agent for nurses commercially available in the US in September 2026. |  | [HIT Consultant](https://hitconsultant.net/2026/09/14/oracle-health-launches-clinical-ai-agent-for-nurses-inpatient-ehr-documentation/) |
| 115 | Recording hardware | On 31 August 2026, Healthwatch England warned that AI scribes get drug and diagnosis names wrong, and it found 27 scribe products in use in England. |  | [The Next Web](https://thenextweb.com/news/nhs-ai-scribes-errors-healthwatch-mhra) |
| 116 | Litigation and consent | A psychiatrist in Melbourne turned away new patients who would not agree to a scribe. |  | [The Guardian](https://www.theguardian.com/australia-news/2026/may/19/melbourne-psychiatrist-ai-note-taking-new-patients) |

## Timeline

| Date | Category | Event | Label |
| :--- | :--- | :--- | :--- |
| Mar 2025 | Evidence | [PHTI finds that scribes reduce burnout but do not yet improve health systems' finances.](https://www.axios.com/2025/03/27/ai-scribes-reduce-burnout-financial-improvement) |  |
| Jun 2025 | Product and market | [Abridge is valued at $5.3bn, and 62% of Epic customers have now installed a scribe.](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/) |  |
| Jul 2025 | Investment | [Ambience raises a $243m Series C at $1.25bn.](https://www.beckershospitalreview.com/healthcare-information-technology/ai/ambience-healthcare-reaches-1-25b-valuation/) |  |
| Aug 2025 | Product and market | [Epic announces its own scribe, AI Charting, at its user group meeting.](https://www.healthcaredive.com/news/epic-rolls-out-ai-charting-art-notetaking-documentation-scribe/811462/) |  |
| Sep 2025 | Product and market | [KLAS interviews 35 Epic customers about Epic's own scribe.](https://hitconsultant.net/2025/09/29/klas-report-epics-native-ambient-speech-tool-reshapes-customer-ai-strategies/) |  |
| Oct 2025 | Product and market | [Menlo Ventures puts the ambient scribe market at $600m.](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft) |  |
| Nov 2025 | Litigation | [A patient files a proposed class action against Sharp HealthCare over the use of Abridge without consent.](https://wtlgovernance.com/insights/updates/sutter-memorialcare-abridge-ai-transcription-consent-lawsuit/) |  |
| Dec 2025 | Evidence | [NEJM AI publishes the UCLA and UW Health RCTs.](https://pubmed.ncbi.nlm.nih.gov/41497288/) |  |
| 1 Jan 2026 | Regulation | [Texas TRAIGA comes into force.](https://www.akerman.com/en/perspectives/hrx-new-year-new-ai-rules-healthcare-ai-laws-now-in-effect.html) |  |
| 6 Jan 2026 | Regulation | [FDA updates its Clinical Decision Support guidance.](https://www.faegredrinker.com/en/insights/publications/2026/1/key-updates-in-fdas-2026-general-wellness-and-clinical-decision-support-software-guidance) |  |
| Jan 2026 | Product and market | [Sykehusinnkjøp opens a scribe tender for Helse Sør-Øst, and Speechmatics releases a Swedish medical model.](https://www.mercell.com/permalink/267602147.aspx) |  |
| 4 Feb 2026 | Product and market | [Epic releases AI Charting to all customers.](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/) |  |
| Feb 2026 | Regulation | [HHS-OIG publishes Medicare Advantage guidance that warns against software prompting payment-raising diagnoses.](https://www.parkerpoe.com/news/2026/06/oig-guidance-warns-physician-practices-of-compliance-risks) |  |
| Mar 2026 | Prices and insurers | [Trilliant and Blue Health Intelligence link scribe adoption to higher billing.](https://www.techtarget.com/revcyclemanagement/feature/How-AI-scribes-are-shifting-coding-intensity-reimbursement) |  |
| 23 Mar 2026 | Product and market | [Heidi launches Heidi Remote, a wearable recording device.](https://www.iatrox.com/blog/heidi-remote-dedicated-ai-scribe-hardware-device-clinical-documentation-2026) |  |
| 9 Apr 2026 | Litigation | [A proposed class action is filed against Sutter Health and MemorialCare.](https://arstechnica.com/tech-policy/2026/04/californians-sue-over-ai-tool-that-records-doctor-visits/) |  |
| Apr 2026 | Evidence | [JAMA publishes the five-system study, and Annals of Internal Medicine the VA test of 11 scribes.](https://www.eurekalert.org/news-releases/1122192) |  |
| Apr 2026 | Investment | [An analyst firm reports a $316m Abridge extension; no other source confirms it.](https://sacra.com/c/abridge/) | single secondary source |
| 28 Apr 2026 | Product and market | [UChicago Medicine signs a system-wide agreement with Artisight.](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/) |  |
| 1 May 2026 | Prices and insurers | [Microsoft cuts the Dragon Copilot per-user price.](https://www.schneider.im/microsoft-dragon-copilot-price-decrease-licensing-changes/) | single secondary source |
| May 2026 | Investment | [Commure raises $70m at a $7bn post-money valuation.](https://hlth.com/insights/news/commure-raises-70m-to-expand-ai-healthcare-operations-platform) |  |
| May 2026 | Evidence | [JAMIA publishes a trial in which clinicians tried two unnamed products in turn.](https://pubmed.ncbi.nlm.nih.gov/41729180/) |  |
| May 2026 | Litigation | [A Melbourne psychiatrist refuses new patients who do not agree to a scribe.](https://www.theguardian.com/australia-news/2026/may/19/melbourne-psychiatrist-ai-note-taking-new-patients) |  |
| May 2026 | Evidence | [A UC Irvine preprint compares five recording devices and six speech engines.](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full) |  |
| 14 Jul 2026 | Regulation | [CMS's proposed 2027 physician payment rule asks for comments on ambient documentation tools.](https://www.hklaw.com/en/insights/publications/2026/07/cms-issues-cy-2027-medicare-physician-fee-schedule-proposed-rule) |  |
| Jul 2026 | Regulation | [The GAO publishes a short review of AI for medical notes and coding.](https://carepilot.com/blog/gao-ai-medical-notes-coding) | single secondary source |
| 27 Jul 2026 | Regulation | [The EU Digital Omnibus on AI comes into force.](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices) |  |
| 29 Jul 2026 | Regulation | [The MHRA publishes guidance on ambient voice technology, and NHS England publishes version 3 of its guidance.](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide) | single secondary source |
| 2 Aug 2026 | Regulation | [EU AI Act Article 50 starts: people must be told when they are dealing with AI.](https://www.morganlewis.com/pubs/2026/06/eu-approves-delays-and-other-amendments-to-certain-eu-ai-act-obligations-what-businesses-should-know) |  |
| 6 Aug 2026 | Evidence | [HSSIB starts a national investigation into ambient voice technology.](https://biotalk.twobirds.com/post/102nh3r/ai-scribes-in-the-nhs-from-compliance-questions-to-formal-investigation) |  |
| Aug 2026 | Litigation | [Raphael v. Mantei: a Washington judge rules that a patient's DAX recording is exempt from disclosure.](https://www.beckershospitalreview.com/healthcare-information-technology/ai/patients-cant-access-ai-scribe-recordings-washington-judge-rules/) |  |
| Aug 2026 | Evidence | [JMIR publishes a meta-analysis of 21 studies, and a Heidi-funded preprint reports on note errors.](https://pubmed.ncbi.nlm.nih.gov/42550089/) |  |
| 31 Aug 2026 | Evidence | [Healthwatch England warns that scribes get drug and diagnosis names wrong.](https://www.theguardian.com/society/2026/aug/31/doctors-ai-scribes-get-names-of-drugs-and-diagnoses-wrong-nhs-watchdog-warns) |  |
| 14 Sep 2026 | Product and market | [Oracle Health makes its Clinical AI Agent for nurses available in the US.](https://hitconsultant.net/2026/09/14/oracle-health-launches-clinical-ai-agent-for-nurses-inpatient-ehr-documentation/) |  |
| Sep 2026 | Investment | [Heidi raises US$340m at a reported US$900m. Tandem raises a $100m Series B.](https://hlth.com/insights/news/heidi-raises-340m-to-expand-clinical-ai-into-supervised-agentic-workflows) |  |
| 7 Oct 2026 | Product and market | [Teladoc adds the SoloScribe ambient scribe to its Solo platform.](https://fiercehealthcare.com/ai-and-machine-learning/teladoc-health-unveils-2-new-ai-capabilities-within-solo-platform) |  |
| Oct 2026 | Investment | [Suki raises a $90m Series E.](https://www.healthcaredive.com/news/suki-raises-90m-in-series-e-round-to-create-ai-agents/832862/) |  |
| Oct 2026 | Product and market | [The Artisight All-In-One Device is on the TIME list of Best Inventions of 2026.](https://www.medicaldesignandoutsourcing.com/medtronic-recor-oculogica-and-more-make-times-best-inventions-list/) |  |
| 4 Nov 2026 | Regulation | [CMS releases the 2027 final rule; summaries mention no payment policy for ambient documentation tools.](https://www.aha.org/news/headline/2026-11-04-cms-issues-cy-2027-physician-fee-schedule-final-rule) |  |
| Summer 2027 (expected) | Evidence | [HSSIB expects to publish its findings.](https://biotalk.twobirds.com/post/102nh3r/ai-scribes-in-the-nhs-from-compliance-questions-to-formal-investigation) |  |
| 2 Dec 2027 (expected) | Regulation | [The EU AI Act's Annex III high-risk obligations start.](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices) |  |
| 2 Aug 2028 (expected) | Regulation | [The EU rules for AI in medical devices (under the MDR and IVDR) start.](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices) |  |

## Sources

1. [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/the-fastest-adopted-tech-in-healthcare-report/)
2. [DH Insights](https://dhinsights.org/news/epic-shakes-up-the-ai-ambient-listening-market-with-integrated-option)
3. [CarePilot summary of GAO-26-109116](https://carepilot.com/blog/gao-ai-medical-notes-coding)
4. [AI for medical notes and coding, GAO-26-109116, GAO, Jul 2026](https://www.gao.gov/products/gao-26-109116)
5. [Cleveland Clinic's AI scribe is giving some doctors joy again, Business Insider, 2 Jun 2026](https://www.businessinsider.com/cleveland-clinic-ambient-ai-scribe-reducing-doctor-workload-2026-06)
6. [Is it OK to record your visit?, WBUR, 5 May 2026](https://www.wbur.org/upnext/2026/05/05/artificial-intelligence-primary-care-privacy)
7. [New Market Pitch](https://newmarketpitch.com/blogs/news/digital-health-abridge-vs-microsoft)
8. [Randomized trial of DAX and Nabla at UCLA, NEJM AI, Dec 2025](https://pubmed.ncbi.nlm.nih.gov/41497288/)
9. [Stepped-wedge trial at UW Health, NEJM AI, Dec 2025](https://pubmed.ncbi.nlm.nih.gov/41625485/)
10. [Five-system study of ambient AI, JAMA via EurekAlert, Apr 2026](https://www.eurekalert.org/news-releases/1122192)
11. [TechTarget](https://www.techtarget.com/searchHealthIT/news/366638795/Epics-AI-Charting-goes-live)
12. [Abridge expands nursing AI platform to 250+ health systems, Newsweek, 6 May 2026](https://www.newsweek.com/ai-nurses-health-care-abridge-access-health-11921461)
13. [AI tools that take notes for GPs could miss vital information, BBC, 3 Sep 2026](https://www.bbc.co.uk/news/articles/cy0zylpdveqo)
14. [MHRA guidance on ambient voice technology, MedDeviceGuide, 2026](https://meddeviceguide.com/blog/mhra-ambient-voice-technology-avt-ai-scribe-medical-device-qualification-guide)
15. [Heidi excludes UK and EU from new features, IT Brief UK, Sep 2026](https://itbrief.co.uk/story/heidi-raises-us-340m-to-expand-ai-agents-in-healthcare)
16. [Two in five Australian GPs use AI scribes, The Guardian, 28 Mar 2026](https://www.theguardian.com/australia-news/2026/mar/29/doctors-using-ai-for-notes-australia)
17. [Queensland health service's AI scribe saves doctors 59 minutes daily, MobiHealthNews, 9 Sep 2026](https://www.mobihealthnews.com/news/anz/queensland-health-services-ai-scribe-saves-doctors-59-minutes-daily)
18. [Emergency doctors estimate Heidi saving up to 10 minutes per patient, RNZ, 16 Mar 2026](https://www.rnz.co.nz/news/health/589774/emergency-doctors-estimate-ai-scribe-heidi-saving-up-to-10-minutes-per-patient)
19. [Heidi raises capital at $1.54 billion valuation, NZ Herald, 23 Sep 2026](https://www.nzherald.co.nz/business/health-new-zealands-ai-scribe-partner-heidi-raises-capital-at-154-billion-valuation/premium/C4DDY7UP6RE5XPMS65KGHED4WU/)
20. [AI scribe now being tested in EDs across Alberta, University of Alberta, 2 Mar 2026](https://www.ualberta.ca/en/folio/2026/03/ai-scribe-now-being-tested-in-eds-across-alberta.html)
21. [Spanish 48-hospital emergency network, PubMed](https://pubmed.ncbi.nlm.nih.gov/42456601/)
22. [FrenchWeb](https://www.frenchweb.fr/avec-865-millions-deuros-tandem-health-veut-passer-du-scribe-medical-au-systeme-dexploitation-clinique/463406)
23. [Helse Sør-Øst ambient scribe tender, Mercell, Jan 2026](https://www.mercell.com/permalink/267602147.aspx)
24. [eCorpIT](https://ecorpit.com/ambient-ai-medical-scribes-india-build-vs-buy-dpdp-2026/)
25. [The real ambient scribe question is trust, HackerNoon, 3 Apr 2026](https://hackernoon.com/the-real-ambient-scribe-question-is-trust-not-note-quality)
26. [AI scribing tools cut physician burnout by over 21 percent, Medical Daily, 12 Jun 2026](https://www.medicaldaily.com/ai-ambient-scribe-physician-burnout-mass-general-brigham-ucla-study-2026-475610)
27. [UCLA Health](https://www.uclahealth.org/news/release/ucla-study-finds-ai-scribes-may-reduce-documentation-time)
28. [Crossover trial of two ambient products, JAMIA, May 2026](https://pubmed.ncbi.nlm.nih.gov/41729180/)
29. [PubMed 40675605](https://pubmed.ncbi.nlm.nih.gov/40675605/)
30. [Meta-analysis of ambient AI and burnout, JMIR, Aug 2026](https://pubmed.ncbi.nlm.nih.gov/42550089/)
31. [STAT](https://www.statnews.com/2025/12/03/ambient-scribe-trial-doctor-burnout-ai-prognosis-newsletter/)
32. [20-ED study of AI-scribed notes, Annals of Emergency Medicine, Oct 2026](https://pubmed.ncbi.nlm.nih.gov/42949694/)
33. [PubMed 42574744](https://pubmed.ncbi.nlm.nih.gov/42574744/)
34. [NCT07456241](https://clinicaltrials.gov/study/NCT07456241)
35. [The doctor's AI will see you now, KEYT/Stacker, 6 Oct 2026](https://keyt.com/stacker-ai/2026/10/06/the-doctors-ai-will-see-you-now/)
36. [AI scribes in the NHS: HSSIB investigation, Bird & Bird, Aug 2026](https://biotalk.twobirds.com/post/102nh3r/ai-scribes-in-the-nhs-from-compliance-questions-to-formal-investigation)
37. [Healthcare Dive](https://www.healthcaredive.com/news/ecri-health-tech-hazards-2026/810223/)
38. [AI scribe tools produce lower-quality notes than clinicians, Newswise/ACP, Apr 2026](https://www.newswise.com/articles/ai-scribe-tools-produce-lower-quality-medical-notes-compared-to-human-clinicians)
39. [Annals](https://www.acpjournals.org/doi/10.7326/ANNALS-25-02772)
40. [Heidi safety preprint, medRxiv, Aug 2026](https://www.medrxiv.org/content/10.64898/2026.08.18.26360701v1.full)
41. [Medical AI Scribes Are Everywhere. Research Shows Benefits (& Risks), Forbes, 6 Mar 2026](https://www.forbes.com/sites/jessepines/2026/03/06/medical-ai-scribes-are-everywhere-research-shows-benefits--risks/)
42. [Health IT leaders must stay on top of AI that delves into clinical workflows, Healthcare IT News, 11 Mar 2026](https://www.healthcareitnews.com/news/health-it-leaders-must-stay-top-ai-delves-clinical-workflows)
43. [PubMed 42877726](https://pubmed.ncbi.nlm.nih.gov/42877726/)
44. [PubMed 41772044](https://pubmed.ncbi.nlm.nih.gov/41772044/)
45. [Dutch comparison of three scribe systems, PubMed](https://pubmed.ncbi.nlm.nih.gov/42887000/)
46. [Ambient AI scribes may risk missing vital patient information, News-Medical, 4 Sep 2026](https://www.news-medical.net/news/20260903/Ambient-AI-scribes-may-risk-missing-vital-patient-information.aspx)
47. [Health AI should be assistive, not autonomous, MedPage Today, 16 Aug 2026](https://www.medpagetoday.com/opinion/second-opinions/122622)
48. [Digital Health](https://www.digitalhealth.net/?p=238603)
49. [Epic rolls out AI Charting, Healthcare Dive, Feb 2026](https://www.healthcaredive.com/news/epic-rolls-out-ai-charting-art-notetaking-documentation-scribe/811462/)
50. [Epic AI Charting launch, STAT, 4 Feb 2026](https://www.statnews.com/2026/02/04/epic-ai-charting-ambient-scribe-abridge-microsoft/)
51. [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/inside-health-systems-early-use-of-epics-newest-ai-tools/)
52. [KLAS report on Epic's native ambient tool, HIT Consultant, Sep 2025](https://hitconsultant.net/2025/09/29/klas-report-epics-native-ambient-speech-tool-reshapes-customer-ai-strategies/)
53. [Best in KLAS 2026, Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/best-in-klas-2026-whos-winning-in-ambient-ai-ehrs-revenue-cycle-and-more/)
54. [Abridge company research, Sacra](https://sacra.com/c/abridge/)
55. [Business Wire](https://secure.businesswire.com/news/home/20251209452332/en/athenahealth-Collaborates-with-Microsoft-to-Offer-Microsoft-Dragon-Copilot-as-Ambient-Notes-Option-Enhancing-Choice-and-Flexibility-for-Ambulatory-Clinicians)
56. [Commure raises $70m, HLTH, May 2026](https://hlth.com/insights/news/commure-raises-70m-to-expand-ai-healthcare-operations-platform)
57. [Heidi raises $340m, HLTH, 23 Sep 2026](https://hlth.com/insights/news/heidi-raises-340m-to-expand-clinical-ai-into-supervised-agentic-workflows)
58. [Heidi valued at US$900m, Business News Australia, Sep 2026](https://www.businessnewsaustralia.com/amp-html/articles/heidi-edges-closer-to-unicorn-status-with-us340m-raise-valuing-ai-health-company-at-us900m.html)
59. [Suki raises $90m Series E, Healthcare Dive, Oct 2026](https://www.healthcaredive.com/news/suki-raises-90m-in-series-e-round-to-create-ai-agents/832862/)
60. [AI Weekly](https://aiweekly.co/alerts/suki-raises-90m-series-e-led-by-healthier-capital-at-450-customer-scale)
61. [Tandem Health raises $100m, HLTH, Sep 2026](https://hlth.com/insights/news/tandem-health-raises-100m-to-expand-clinical-ai-across-europe)
62. [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/ambience-healthcare-reaches-1-25b-valuation/)
63. [Sacra on Ambience](https://sacra.com/c/ambience/)
64. [Dragon Copilot price decrease, Schneider IT Management, 2026](https://www.schneider.im/microsoft-dragon-copilot-price-decrease-licensing-changes/)
65. [Axios](https://www.axios.com/2025/03/27/ai-scribes-reduce-burnout-financial-improvement)
66. [Orthopaedic coding accuracy study, PubMed](https://pubmed.ncbi.nlm.nih.gov/42671176/)
67. [Trilliant](https://www.trillianthealth.com/market-research/studies/outpatient-coding-intensity-increases-as-hospitals-adopt-ai-enabled-scribing)
68. [How AI scribes are shifting coding intensity, reimbursement, TechTarget, 17 Mar 2026](https://www.techtarget.com/revcyclemanagement/feature/How-AI-scribes-are-shifting-coding-intensity-reimbursement)
69. [STAT](https://www.statnews.com/2026/04/08/are-scribes-raising-health-care-costs-ai-prognosis/)
70. [CY2027 physician fee schedule proposed rule, Holland & Knight, Jul 2026](https://www.hklaw.com/en/insights/publications/2026/07/cms-issues-cy-2027-medicare-physician-fee-schedule-proposed-rule)
71. [CMS](https://www.cms.gov/newsroom/fact-sheets/calendar-year-cy-2027-medicare-physician-fee-schedule-proposed-rule)
72. [AHA](https://www.aha.org/news/headline/2026-11-04-cms-issues-cy-2027-physician-fee-schedule-final-rule)
73. [Medical Economics](https://www.medicaleconomics.com/view/medicare-s-final-2027-fee-schedule-cuts-physician-pay-more-than-proposed-but-softens-the-same-day-visit-penalty)
74. [OIG guidance on physician practice compliance, Parker Poe, Jun 2026](https://www.parkerpoe.com/news/2026/06/oig-guidance-warns-physician-practices-of-compliance-risks)
75. [JUCM](https://www.jucm.com/documents/2026/03/what-to-know-about-payers-downcoding-procedures.pdf/)
76. [Commure Ambient AI achieves Class I certification and MHRA registration, MENAFN, 3 Sep 2026](https://menafn.com/1111615428/Commure-Ambient-AI-Achieves-Class-I-Certification-And-MHRA-Registration)
77. [NHS England guidance](https://www.england.nhs.uk/long-read/guidance-on-the-use-of-ai-enabled-ambient-scribing-products-in-health-and-care-settings/)
78. [Faegre Drinker](https://www.faegredrinker.com/en/insights/publications/2026/1/key-updates-in-fdas-2026-general-wellness-and-clinical-decision-support-software-guidance)
79. [COSM](https://www.cosmhq.com/resources-posts/eu-ai-act-digital-omnibus-new-timeline-medical-devices)
80. [EU AI Act delays and amendments, Morgan Lewis, Jun 2026](https://www.morganlewis.com/pubs/2026/06/eu-approves-delays-and-other-amendments-to-certain-eu-ai-act-obligations-what-businesses-should-know)
81. [TNGlobal](https://technode.global/prnasia/corti-steps-up-its-support-for-the-next-wave-of-healthcare-startups-as-openai-pushes-into-the-space-and-others-retreat-from-europe/)
82. [Healthcare AI laws now in effect, Akerman, 2026](https://www.akerman.com/en/perspectives/hrx-new-year-new-ai-rules-healthcare-ai-laws-now-in-effect.html)
83. [iatroX](https://www.iatrox.com/blog/heidis-coming-ai-capabilities-exclude-the-uk-and-eu-an-opening-for-tortus)
84. [WTL Governance](https://wtlgovernance.com/insights/updates/sutter-memorialcare-abridge-ai-transcription-consent-lawsuit/)
85. [Medscape](https://www.medscape.com/viewarticle/health-system-sued-over-ai-scribe-technology-patient-consent-2026a10001k7)
86. [Basil AI](https://basilai.app/articles/2026-06-05-ambient-ai-scribes-abridge-sutter-memorialcare-sharp-lawsuits-cipa-false-consent.html)
87. [Californians sue over AI tool that records doctor visits, Ars Technica, 10 Apr 2026](https://arstechnica.com/tech-policy/2026/04/californians-sue-over-ai-tool-that-records-doctor-visits/)
88. [Patients can't access AI scribe recordings, Becker's, Sep 2026](https://www.beckershospitalreview.com/healthcare-information-technology/ai/patients-cant-access-ai-scribe-recordings-washington-judge-rules/)
89. [The Guardian](https://www.theguardian.com/australia-news/2026/may/19/melbourne-psychiatrist-ai-note-taking-new-patients)
90. [AAP](https://aapnews.aap.com.au/news/doctors-ai-usage-raising-privacy-and-safety-concerns)
91. [AI medical scribe consent raises patient rights concerns, New Atlas, 5 Aug 2026](https://newatlas.com/medical-ai/ai-medical-scribe-consent-raises-patient-rights-concerns/)
92. [Re: Unintended consequences of using ambient scribes in general practice, BMJ, 16 Mar 2026](https://www.bmj.com/content/390/bmj-2025-085754/rr)
93. [Doctolib clinical AI lab, Caducée.net, Feb 2026](https://www.caducee.net/actualite-medicale/16814/doctolib-accelere-dans-l-ia-clinique-avec-un-laboratoire-dedie.html)
94. [iatroX](https://www.iatrox.com/blog/ambient-scribing-uk-gps-tortus-vs-heidi-vs-accurx-2026-comparison)
95. [Speechmatics](https://www.speechmatics.com/company/articles-and-news/speechmatics-launches-new-swedish-medical-model-cutting-transcription-errors)
96. [UChicago Medicine and Artisight, HIT Consultant, 28 Apr 2026](https://hitconsultant.net/2026/04/28/uchicago-medicine-artisight-smart-hospital-deployment/)
97. [Fierce Healthcare](https://www.fiercehealthcare.com/hospitals/uchicago-medicine-artisight-partner-smart-hospital-platform-deployment)
98. [Rutland Herald/BusinessWire](https://www.rutlandherald.com/news/business/artisight-now-runs-epic-mychart-bedside-tv-powering-a-fully-connected-smart-hospital-experience/article_52e91483-a03f-5d82-9079-f8bb503486d7.html)
99. [Stryker acquires care.ai, Medical Device Network](https://www.medicaldevice-network.com/news/stryker-acquires-smart-hospital-tech-provider-care-ai/)
100. [Teton.ai Nordsjællands Hospital report, Teton](https://www.teton.ai/blog/nhl-noh-report)
101. [NCT06339125](https://clinicaltrials.gov/study/NCT06339125)
102. [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/nurses-push-back-on-healthcare-tech.html)
103. [OR Manager](https://ormanager.com/briefs/or-camera-privacy-concerns-influence-nursing-strike-negotiations)
104. [Microsoft Support](https://support.microsoft.com/en-us/dragon-copilot/physicians/select-and-configure-your-microphone)
105. [Microsoft Learn](https://learn.microsoft.com/en-us/industry/healthcare/dragon-copilot/installation/physicians/)
106. [NCT07598721](https://clinicaltrials.gov/study/NCT07598721)
107. [medRxiv](https://www.medrxiv.org/content/10.64898/2026.05.19.26353590v1.full)
108. [npj Digital Medicine](https://www.nature.com/articles/s41746-026-02554-0)
109. [Pacific Transcription](https://shop.pacifictranscription.com.au/blogs/dictation-solutions-2026/check-out-the-new-philips-speechmike-ambient-wearable-ai-assistant)
110. [eDist](https://www.edist.com/become-a-reseller/philips-reseller/philips-speechmike-ambient-for-dragon-copilot/)
111. [Philips](https://www.dictation.philips.com/us/philips-speechmike-ambient-wearable-ai-assistant-psm5000-series/)
112. [Heidi](https://www.heidihealth.com/blog/heidi-launches-hardware)
113. [Medical Design and Outsourcing](https://www.medicaldesignandoutsourcing.com/medtronic-recor-oculogica-and-more-make-times-best-inventions-list/)
114. [Fierce Healthcare](https://fiercehealthcare.com/ai-and-machine-learning/teladoc-health-unveils-2-new-ai-capabilities-within-solo-platform)
115. [Philips SpeechLive](https://www.speechlive.com/ca/content-hub/blog/philips-speechmike-ambient-vs-heidi-remote/)
116. [StartStop](https://www.startstop.com/philips-psm5000-00-speechmike-ambient-wearable-ai-assistant/)
117. [Forbes Australia](https://www.forbes.com.au/news/entrepreneurs/australian-ai-startup-heidi-health-launches-its-first-hardware-device/)
118. [Medical Republic](https://www.medicalrepublic.com.au/heidis-hardware-play-takes-aim-at-apple-sized-problem-in-clinical-ai/124330)
119. [iatroX](https://www.iatrox.com/blog/heidi-remote-dedicated-ai-scribe-hardware-device-clinical-documentation-2026)
120. [HIT Consultant](https://hitconsultant.net/2026/09/14/oracle-health-launches-clinical-ai-agent-for-nurses-inpatient-ehr-documentation/)
121. [Becker's](https://www.beckershospitalreview.com/healthcare-information-technology/ai/hows-ambient-ai-for-nurses-an-early-review/)
122. [The Guardian](https://www.theguardian.com/society/2026/aug/31/doctors-ai-scribes-get-names-of-drugs-and-diagnoses-wrong-nhs-watchdog-warns)
123. [The Next Web](https://thenextweb.com/news/nhs-ai-scribes-errors-healthwatch-mhra)

## About this report

Cyberfuturists Independent Research. Analyst: Ryan Marohn. More research: https://cyberfuturists.com/research

© 2026 Cyber Futurists. Cyberfuturists is owned by Aunoo AI Ltd (Company no. 16402069).
