Report preview · Cyberfuturists Independent Research

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.

Ryan Marohn · 7 November 2026 · Evidence current to 3 November 2026

01 · Adoption

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

The share of US clinicians who use AI for documentation rose from 21% in 2024 to 28% in 2026, but use within each deployment is much lower than the headline numbers suggest.

Clinicians who use AI for documentation, percentage

The open circle shows the first measure and the filled circle the latest.

Level of use in each deployment

02 · Evidence

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

At least five RCTs, a large multisite observational study and a meta-analysis now exist. They agree on the direction of the effects, but the effects are smaller than the earlier studies reported.

Documentation time saved, minutes for each clinician each day

Hover over or select a bar for the source detail.

03 · Regulation

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

The US, Great Britain and the EU now apply the same test: the product's intended purpose.

Where the MHRA says a scribe becomes a medical device

The nine worked examples in the MHRA guidance of 29 July 2026. Select an example. single secondary source

Not a medical device

Medical device

Select an example to see what applies.
04 · Full report

Contents of the full report

Each heading states the section's finding. The full report supports them with 116 labelled claims from 123 sources.

The report page has the full report, the PDF, the full explorer with the claims ledger, and the full Markdown file for agents.

Full report and PDF
05 · Watch

What would change our view

06 · Method

Method and evidence labels

We used 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.