What physicians need before they trust an AI tool

Most physicians already use some AI at work. In the AMA's 2026 physician survey, about seven in ten said they use at least one AI tool in their practice.

The more useful part is what the survey's 1,692 physicians said they need before adopting a tool. Their answers make good questions for any vendor.

What physicians said they need

The AMA tested 21 factors that might help physicians adopt AI. Two led the list:

• 88% rated it important that a tool's safety and efficacy be validated by a trusted entity and monitored over time.
• 86% rated it important that their own practice or hospital and their EHR vendor assure data privacy.

Other top requirements included a designated feedback channel when issues arise, coverage under standard malpractice insurance, and seamless EHR integration.

Liability came up again when the AMA asked about regulation. Physicians ranked clear liability frameworks as the step that would most increase their trust. And 85% want to be consulted on, or responsible for, bringing AI into their practice.

Why a colleague's word is not enough

"Other physicians I know are using it" ranked near the bottom. Only 44% called it important. It was one of just two factors out of 21 that fewer than half rated important.

A colleague can tell you a tool is easy to use. They usually can't tell you where your patients' data goes.

Who supplies the tool matters too. In the survey, 71% were concerned about patient privacy with AI tools their institution did not provide. For institutional tools, that figure was 42%.

In an independent practice, there may be no IT team or compliance office doing that vetting. You are the institution. Only 55% of physicians in solo practice had any AI training, compared with 76% in hospital settings.

These six questions cover what the survey says matters most.

Six questions to ask any AI vendor

1. How was accuracy measured, and by whom? Ask what was tested, on how many charts, and who checked the answers. Ask whether they keep measuring after launch. A single percentage with no method is not proof. In 2024, Texas settled with a health AI company that advertised a "severe hallucination rate" below 1 per 100,000. The state said those metrics were likely inaccurate.

2. What leaves my computer, and where does it go? Ask which parts of the chart are sent, where they are processed, and how long they are kept. Ask whether names and record numbers are removed first. Get the answers in writing.

3. Will you sign a business associate agreement? HIPAA lets a vendor handle protected health information on your behalf only if you get satisfactory assurance that it will safeguard the data. That assurance must be documented in a written agreement. If a vendor will handle patient information and won't sign one, stop there.

4. Can it change my chart or submit a claim? Ask what the tool can write, not just what it can read. Physicians want clear rules on who is liable when AI gets it wrong. Until you have that answer, a tool that only points is safer than one that acts. Also ask your malpractice carrier how your policy treats AI-assisted care.

5. Who makes the final call, and does it flag both ways? A person should decide every change. The tool should also tell you when a note doesn't support a code, not only when something could be added. One-way review has been costly. In March 2026, Aetna agreed to pay $117.7 million to resolve allegations about inaccurate diagnosis codes. The government alleged that Aetna used chart reviews to add codes but did not withdraw codes the reviews failed to support.

6. How do I report a wrong answer, and what happens next? Ask who reads the report, how fast they reply, and whether the fix reaches every user. Physicians listed a feedback channel among their top needs. They also ranked ongoing monitoring, including adverse event reporting, second among regulatory steps.

Put the answers in writing

Here is the one concrete step. Before you install anything, email the vendor these six questions and keep the replies. Plain written answers to all six earn a trial. Vague answers about your data or the BAA are an answer too.

We try to answer these in public for Cortex Lens, our Chrome extension that works inside Practice Fusion. It is read-only: it never edits the chart or submits a claim, and a person makes every change. The patient's name and record number are removed in the browser before a chart is checked. Flags go both ways. Our accuracy benchmarks are public, and our data handling is described on our security page. Ask us the same six questions.

Sources: AMA, 2026 Physician Survey on Augmented Intelligence, March 2026, pages 2, 4-5 and 10-13 (fielded January 15 to February 2, 2026; n=1,692). Texas Attorney General, Pieces Technologies settlement, September 18, 2024. 45 CFR 164.502(e), disclosures to business associates. U.S. Department of Justice, Aetna settlement, March 11, 2026.