Eight of twelve digital scribes may be medical devices. None of the twelve said so.

What the TGA found about the digital scribe#

The fox reads the fine print on a sign before it crosses the field, and it never trusts a trail that a disclaimer has swept clean. The Therapeutic Goods Administration published its digital scribe review outcomes on 16 October. Of the 12 scribes it reviewed, none of the suppliers identified their products as medical devices, and eight included functions likely to meet the definition of a medical device. Health Services Daily, reporting the review, adds that the TGA drew the 12 from about 100 suppliers, with 40 answering a voluntary survey.

Read the sentence in its place. The TGA says plainly that it did not test any digital scribe products, check safety and performance, or assess compliance with detailed medical device requirements as part of this review. Health Services Daily is just as careful: the finding does not establish that those products are unsafe or that individual suppliers have breached the law. It is a finding about definition, not about harm.

The regulator has not said which eight. Health Services Daily notes that the TGA has not publicly identified them, so no product should be assumed to be among the eight, and we name none. Several well-known scribes were reviewed; being reviewed is not the same as being flagged.

The simplest way to hold the result is the ratio: two thirds, a pack running through the hedgerow while the rest of the den assumes it is empty. That is an alarm for suppliers who thought a medical scribe product could sit outside the rules because it only transcribes, and a prompt for buyers who assumed the label on the box was the regulator's view.

Where a medical scribe stops being administrative#

The line is drawn on the TGA's own digital scribes page. Digital scribes intended only to perform administrative tasks are not a medical device. A digital scribe will meet the definition of a medical device if it makes clinical recommendations. A medical scribe tool that turns a consultation into a note sits on the safe side. One that suggests a diagnosis or a treatment sits on the other.

The page adds a point that marketers should read twice. A product may contain both device and non-device functions, and its intended purpose is set by product labelling, instructions for use, advertising and promotional materials, product websites and other marketing materials. In other words, the sales page is evidence. A line of copy about spotting problems in the record can change what the regulator thinks the software is for.

That is also why the disclaimer fails. The TGA's review says statements or disclaimers are not sufficient in determining if a product is, or is not, a medical device, and its scribes page says general disclaimers, for example this product is not for diagnosis, are not acceptable to demonstrate a digital scribe is not a medical device. A paragraph at the foot of the page cannot undo a feature on the front of it.

The TGA's overview of software and AI is explicit that it is the starting point if you are planning to supply a software medical device in Australia, including those using artificial intelligence. Its separate guidance on software-based medical devices, dated 24 February 2026, and on clinical decision support, dated 16 October 2026, are listed there; we did not open them and do not summarise them.

What other regulators say about AI scribes#

Australia is the first regulator we know of to have reviewed a named group of scribes this way, but the logic travels. The UK's MHRA states in a 2024 policy paper that where AI is used for a medical purpose, it is very likely to come within the definition of a general medical device. That is a position paper from April 2024 rather than current device guidance, and MHRA's current software guidance was not opened.

In the United States, the FDA keeps a final guidance on Clinical Decision Support Software, shown as January 2026, which says its existing digital health policies continue to apply to software functions that meet the definition of a device, including those intended for use by patients or caregivers. We did not read the four criteria for non-device decision support from the primary text, and we do not state them.

Under the EU AI Act, the independent site artificialintelligenceact.eu reproduces Article 6 and says AI systems used solely for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control do not qualify as safety components. That site is a third-party mirror, not the Official Journal, and it marks some paragraphs as new, so check EUR-Lex before relying on it. Whether a given AI scribes product is high-risk depends on its function and sector.

Put the four together and a pattern appears, like fresh prints in the undergrowth. Every regulator ties the answer to what the software does, and every one leaves the supplier to work it out first. An AI scribes vendor selling across borders needs one claims map, not four separate hopes.

What the evidence says about the medical scribe software itself#

A regulatory finding invites a lazy reply: the tools help doctors, so relax. The honest reading is narrower. A preprint of a randomised trial, Lukac and colleagues, compared two ambient AI scribes with usual care across 238 physicians in 14 specialties at one large academic health system in California. It is a preprint and has not been peer-reviewed.

Its results differ by product. One tool's users experienced a 9.5% decrease in time-in-note versus the control group (95% CI -17.2% to -1.8%, p=.02), and the other showed no significant change versus control (-1.7%, p=.66). The authors report that any scribe led to modest improvements in physician burnout, work exhaustion and task load, and that clinically significant inaccuracies were noted occasionally on five-point Likert questions.

A university press release on separate studies reports that self-reported clinician burnout dropped from roughly 52% to 39%, and that clinicians using an ambient documentation tool spent 8.5% less total time in the EHR than matched controls. That comes from the UChicago Medicine release, published November 2025, not from the papers themselves, which we did not open, and burnout there is self-reported.

Neither source measures an audited error rate. Both findings can be true: the tools help with documentation time and they may also do something clinical that a regulator has noticed. The first is a reason to buy. The second is a reason to read the feature list.

A claims map for the best AI medical scribe pitch#

Every vendor wants to be named the best ai medical scribe, and no regulator will say it. What a vendor can do instead is make its claims safe to repeat. The US FTC's health products compliance guidance is blunt about the principle: before disseminating an ad, advertisers must have adequate substantiation for all objective product claims conveyed, expressly or by implication. It is US law and non-binding guidance, but the discipline is universal.

That map is also a marketing asset, a scent trail buyers can follow. A vendor who can show the buyer which functions are administrative, which are clinical, and what evidence sits behind each is selling confidence, which is scarcer than features. Our earlier writing on the ambient scribe claim ladder and on the claim a regulator can quote back at you lays out the same logic for other regulators.

The practical context for a healthcare team is in what a health brand can say while state boards write the rules and in why briefs need a licensing trail. If you sell a digital scribe or any AI product into care, folkfox healthcare marketing writes the claims map, the audience pages and the evidence trail, with content marketing and SEO and GEO carrying them. Talk to us at contact us.

Frequently asked questions#