Physician sentiment is the leading indicator for everything else in healthcare AI: tools get piloted where clinicians are curious and abandoned where they are wary. This page tracks the survey record — anchored on the American Medical Association's three Augmented Intelligence waves, with an independent cross-check — and keeps every figure tied to the wave, sample, and question wording that produced it. As of July 2026.
What counts as "use"?
The single most quoted number here — physician AI use more than doubling from 38% to 81% in three years — spans three different denominators. The AMA's 2026 headline counts awareness or use: it includes the 9% of respondents who are unsure which AI tools their practice offers 1. The stricter measure, physicians incorporating one or more named use cases, runs 38% (2023) → 48% (2024) → 72% (2026) 1. Both series are real, and both point the same direction — but they answer different questions, and mixing them overstates the like-for-like jump. The AMA also changed its instrument between waves: the use-case list grew from 15 to 17 items, and the 2026 analysis added qualified partial responses where earlier waves counted full completes only 1.
The AMA series, wave by wave
The 2026 wave was fielded 15 January–2 February 2026 with 1,692 US physicians, following waves of 1,081 (2023) and 1,183 (2024) 1.
| Measure | 2023 | 2024 | 2026 | Source |
|---|---|---|---|---|
| Awareness or use of AI (headline) | 38% | 66% | 81% | 1 |
| Incorporates ≥1 named use case | 38% | 48% | 72% | 1 |
| Unsure what their practice offers | — | 18% | 9% | 1 |
| Average number of use cases | 1.1 | — | 2.3 | 1 |
| See an advantage to patient care | 65% | 69% | 76% | 1 |
| More excited than concerned | 30% | 35% | 37% | 1 |
Two readings stand out. Use is compounding — the average physician user now runs 2.3 use cases, up from 1.1 in 2023 1. And sentiment is climbing far more slowly than adoption: the excited-outweighs-concerned share gained seven points across three years while like-for-like use gained thirty-four. Physicians are using tools faster than they are coming to trust them. The 2024 press release drew the same picture mid-series: 66% use, and enthusiasm edging past concern for 35% 3.
Where does physician use concentrate?
Documentation and summarization dominate. Summaries of research and standards of care became the top use case in 2026 at 39%, up from 13% in 2024 — the fastest riser in the survey 1. Creation of discharge instructions and care plans reached 30%, billing and visit-note documentation 28%, chart summaries 28%, and drafted patient-portal responses 19%; assistive diagnosis sits at 17% 1. Enthusiasm runs ahead of deployment: 57–58% of physicians are enthusiastic about documentation tools that only 28% currently incorporate 1 — the gap our AI scribe adoption tracker watches from the deployment side, and the reason the ambient AI scribe is the category to watch.
What are physicians concerned about?
The 2026 wave reads as conditional confidence rather than conversion. 40% of physicians remain equally excited and concerned 1. Patient privacy is the only dimension where more physicians expect AI to cause harm than help (41% versus 13%) 1. Skill loss is the other concentrated worry: 88% report at least mild concern, 70% are very or somewhat concerned about skill loss in current students and residents, and early-career physicians worry about their own skills at nearly double the rate of late-career colleagues (35% versus 21%) 1. On patients using AI themselves, physicians draw a clear line between routine questions (broadly welcomed) and diagnostic interpretation — nearly half would never or rarely want patients using AI to interpret pathology (49%) or radiology (46%) results 1, a boundary that keeps the human in the loop where judgement is highest.
What would move adoption further is specific: validated safety and efficacy (88% rate it important), data privacy assurances (86%), and clear liability frameworks — ranked the single most important regulatory action, placed first by 31% 1. 85% want to be consulted on, or responsible for, AI adoption decisions in their practice 2.
How large is the training gap?
The most actionable 2026 finding sits in education. 27% of physicians have received no AI training from any source; among those with any exposure, only 11% describe it as extensive; and 92% want more training delivered 1. Exposure varies by setting more than by age — 76% of hospital-based physicians report some training against 55% in solo practice, while physicians under 35 (79%) and over 55 (74%) are nearly level 1. Preferred formats are embedded ones: as part of EHR training (49%) or while doing the job (46%), with training delivered by AI companies themselves the least preferred option at 14% 1.
An independent check
One survey family, however careful, is one lens. Doximity's two-wave survey of 3,151 US physicians offers a lower-running series: AI use in clinical practice rose from 47% (March–April 2025) to 63% (November 2025–January 2026), with literature search (35%) and voice documentation (29%) the top uses, and accuracy and reliability named the greatest barrier by 71% 4. The direction matches the AMA exactly; the level sits roughly ten points lower against the AMA's like-for-like series — a spread explained by different question wording, different panels, and different field dates rather than by one survey being wrong.
How to read these numbers
Four cautions travel with this page. All of it is self-report — surveys measure what physicians say, and adoption claims are never independently verified against system logs 1. Denominators shift between headline and detail, so quote the 72% like-for-like figure when comparing across years and the 81% only with its definition attached. Instruments changed mid-series — a longer use-case list and added partial completes both nudge 2026 upward. And panel composition differs between survey families, which is why the AMA and Doximity can both be right while sitting ten points apart.
We revisit this page on a ninety-day cycle and whenever a new survey wave lands. For the wider dashboard these attitudes feed into — devices, evidence, rules, and money — start at our AI in healthcare statistics hub, and for what the enthusiasm is actually buying, the AI scribe adoption tracker follows the deployment record.