Are doctors actually using AI today?

What counts as a doctor actually using AI?

Count purposeful use in a real medical workflow, whether the doctor opens the tool directly or reviews work it produces.

That includes asking for a research summary, generating a visit note, reviewing a chart summary, drafting patient instructions, checking a differential diagnosis or acting on an imaging alert. Merely working at a hospital that owns AI software says little about personal use.

Invisible AI makes the boundary harder to draw. A radiologist may receive an automatically prioritized scan, while a primary-care doctor may read a risk flag placed inside the electronic record. Both are using an AI-assisted workflow even if neither opens a chatbot.

For this article, we count repeated or deliberate involvement in professional work. We separate administrative help from clinical judgment throughout, because drafting a note and changing a treatment plan carry very different weight.

Are most doctors using AI today?

Yes. Most US doctors now use at least one AI application professionally, with the best estimate sitting around 60% to 70%.

The American Medical Association surveyed 1,692 physicians and reported that 72% selected at least one professional AI use case. Its larger 81% headline also included 9% who were unsure which AI tools their practice offered. Counting all 81% as active users would stretch the result.

Doximity surveyed 3,151 US physicians across two periods and found 54% current use across the full sample. In its later cohort, that reached 63%, up from 47% in the earlier one. Different samples and wording explain part of the gap, but both studies point the same way: physician AI use has moved well beyond a small group of enthusiasts.

Measure	Share	What it really captures
AMA, at least one use case	72%	Explicit professional use
AMA, awareness or use	81%	Includes 9% unsure what their practice offers
Doximity, full sample	54%	Current use in clinical practice
Doximity, later cohort	63%	Current use among the more recent respondents

If you want more recent data on this point, please see our latest healthcare AI market report.

Are doctors using AI every day or just trying it?

AI is already a daily tool for many doctors, and the frequency data rule out the idea that adoption consists mainly of one-off experiments.

Doximity found that 37% of all surveyed physicians used AI at least daily. Among doctors who had adopted it, 69% used it daily and 36% used it several times a day. Daily use among adopters rose from 64% in the earlier survey period to 74% in the later one.

Family medicine makes the shift easy to see. These doctors face high patient volume, heavy documentation and broad clinical questions. Among those who had adopted AI, 88% used it daily and half used it multiple times a day. That is routine workflow use, not casual curiosity.

The AMA also found that physicians now report an average of 2.3 professional AI use cases, compared with 1.1 in its first survey. Adoption is widening and becoming more intensive at the same time.

What are doctors actually using AI for?

Doctors currently use AI far more for reading, writing and organizing information than for making diagnoses.

In the AMA survey, 39% used AI for summaries of medical research or standards of care. Thirty percent used it to create discharge instructions, care plans or progress notes. Billing and visit documentation reached 28%, as did chart summaries. Assistive diagnosis stood at 17%.

Doximity found almost the same order. Literature search led at 35% in its later cohort, up from 22% in the earlier period. Ambient or voice documentation followed at 29%, up from 20%. Two independent surveys put research retrieval and documentation at the center of physician adoption.

Doctors can check a summary, edit a note and reject a weak draft in seconds. Diagnostic recommendations demand much more trust, stronger evidence and clearer liability rules, so they are spreading more slowly.

AI use case	AMA physician use	Earlier AMA result
Research and standards-of-care summaries	39%	13%
Discharge instructions, care plans or progress notes	30%	20%
Billing, charting or visit documentation	28%	21%
Chart summaries	28%	12%
Patient-portal drafts	19%	9%
Translation	18%	14%
Assistive diagnosis	17%	12%

Have ambient AI scribes become medicine's first killer app?

Yes, ambient scribes are currently the first broadly adopted generative AI tool for doctors.

They solve a daily problem without asking physicians to hand over medical authority. The software listens during a visit, drafts the note and sometimes prepares orders. The doctor reviews the output, fixes it and signs the record.

Scribes are spreading through the systems doctors already use. Doximity found ambient or voice documentation use at 29% in its later physician cohort. Epic has released AI charting directly inside its electronic record, while Microsoft's DAX has been deployed across large health systems. Doctors avoid a separate login, and hospitals avoid rebuilding the whole visit around a new product.

Scribes have also moved from optional add-ons toward a wider clinical platform. Epic's product can draft notes and queue suggested orders from the same conversation. That expansion increases the value of the tool, though it also raises the stakes for review.

If you want more recent data on this point, please see our latest healthcare AI market report.

Do AI scribes save doctors meaningful time?

AI scribes save real time; the best studies show a modest operational gain rather than a dramatic rewrite of the medical workday.

A JAMA Network Open study compared 125 scribe users with 478 balanced nonusers. AI use was associated with 2.4 fewer minutes in the electronic record and 1.8 fewer minutes writing notes per appointment. The study found no significant change in after-hours documentation, appointment length or monthly visit volume.

A newer Providence study covered 1,547 active users, including 1,073 physicians. Median note time fell from 7.1 to 6.1 minutes per appointment, and after-hours documentation declined gradually. Appointments per day barely moved, from 12.4 to 12.7, with no clear association between adoption and visit volume. Only about 8% of eligible clinicians met the study's active-use threshold across the full observation period, although monthly uptake reached 28.8% near the end.

The gain is smaller than the marketing often implies. One or two minutes can add up across thousands of visits, especially for overloaded clinicians, but hospitals have rarely turned those savings into more appointments or the end of evening work.

Study	Active users	Main measured change	Unchanged or unclear
Academic health-system cohort	125	2.4 fewer EHR minutes and 1.8 fewer note minutes per visit	After-hours work, appointment length, visit volume
Providence deployment	1,547	Note time fell from 7.1 to 6.1 minutes	Appointments per day showed no clear gain

Are doctors using AI to diagnose patients?

Doctors are using AI for diagnosis today, mostly as a secondary aid whose value depends heavily on training and workflow design.

The AMA found assistive diagnosis in 17% of physician workflows. Imaging specialists also receive algorithmic findings and triage alerts inside their normal systems. Diagnostic AI has crossed into practice, although it trails search and documentation by a wide margin.

Randomized studies explain the slower uptake. In one trial involving 50 physicians, access to a large language model produced no significant improvement over conventional resources on clinical reasoning cases. Another trial with 92 physicians found a 6.5 percentage-point gain on management reasoning, with slightly longer case-review time.

A more recent study in Pakistan reached a much larger result after physicians completed a 20-hour AI-literacy course. The AI-assisted group scored 71.4%, compared with 42.6% for doctors using conventional resources. That 27.5-point gap suggests training may be a major part of the product. It came from six simulated cases and 58 completing physicians, so projecting the result across everyday medicine would be a mistake.

Study	Setting	Result
50-physician diagnostic reasoning trial	Simulated cases	No significant physician improvement
92-physician management trial	Five complex vignettes	6.5-point improvement with GPT-4
Pakistan randomized trial	Six vignettes after 20 hours of training	27.5-point improvement

Has doctor-facing AI improved real patient outcomes?

We still lack convincing evidence that general-purpose AI used by clinicians reliably improves patient outcomes.

The strongest recent test involved 9,691 patients, 103 clinical officers and 16 primary-care facilities in Kenya. The AI-assisted group used a ChatGPT-4o decision-support system inside the medical record. Treatment failure within 14 days occurred in 2.2% of AI-supported consultations and 2.0% of control consultations. The two groups were statistically indistinguishable, and the investigators concluded that any benefit was probably modest.

Researchers found no intervention-related serious safety problem. The main outcome still failed to improve.

A recent systematic evidence map reached a broader version of the same conclusion. Real-world large-language-model interventions are increasing, but reporting quality and patient-centered outcome evidence remain limited. Another review of more than 1,000 medical LLM studies found that most still lacked realistic clinical settings, with many using tiny samples. Adoption has outrun outcome proof.

If you want more recent data on this point, please see our latest healthcare AI market report.

Is radiology already using AI at scale?

Radiology has the deepest clinical AI infrastructure today, although regulatory clearance still tells us more about product supply than everyday use.

Radiology dominates the FDA's public list of authorized AI-enabled devices. Medical images are already digital, many tasks can be narrowly defined, and an algorithm can work inside the radiologist's existing viewer.

The field has now moved into post-deployment governance. The American College of Radiology recently approved its first practice parameter for imaging AI and launched an Assess-AI framework for monitoring performance after installation. The registry covers use cases such as intracranial hemorrhage, pulmonary embolism, pneumothorax, large-vessel occlusion, breast density and cervical-spine fracture.

Professional bodies are writing rules for local testing, drift monitoring and stop procedures because these tools are already entering care. Clearance counts still tell us little about how many hospitals bought each product, how often radiologists act on its output or whether performance holds across patient populations.

Are doctors replacing medical search engines with AI?

Doctors increasingly start medical searches with AI, then return to original studies, guidelines and trusted references for consequential decisions.

The change is happening quickly. AMA data show research and standards-of-care summaries rising from 13% of physicians to 39%. Doximity recorded literature-search use increasing from 22% to 35% between its two survey periods. Few other use cases gained that much share so quickly.

AI fits this task because doctors already search during clinical work. A conversational tool can compress several papers, explain an unfamiliar guideline or surface a dosage question in seconds. The friction is far lower than installing an imaging platform or rebuilding a hospital workflow.

The typical habit looks closer to AI-first search than AI-only search. Doctors can use the generated answer to find the relevant evidence, then verify the point that affects care. Products built specifically for clinicians are also adding citations, audit controls and links to source material, which makes them easier to approve than a consumer chatbot.

Is medical AI limited to a few tech-heavy specialties?

No, physician AI use now spans primary care and many specialties, although the tool changes with the job.

In Doximity's survey, use reached 64% in neurology, 61% in gastroenterology and 60% in internal medicine. Family medicine and cardiology were both at 58%, while oncology reached 57%. Several other specialties also cleared 50%. The striking part is how close those numbers are, even across very different specialties.

Primary-care doctors have strong reasons to adopt search, chart summaries, portal drafting and ambient notes. Oncologists and neurologists face dense records and fast-changing evidence. Radiologists use computer vision and worklist prioritization. Adoption is broad because AI can attach itself to different bottlenecks.

The tools are also reaching beyond flagship academic hospitals. Doximity recently began integrating its clinical search and scribe tools into Aledade's software for value-based primary-care practices. In another recent update, Doximity said its clinical AI suite had been deployed across more than 150 health systems. These are vendor-reported figures, but they show that large systems and community practices can now buy the same type of tools.

Are hospitals scaling AI or still running pilots?

Leading hospitals are scaling selected AI workflows now, while the typical health system still has a patchwork of mature tools, small rollouts and unfinished experiments.

Epic says more than 85% of its customers use at least one Epic AI capability. Its chart-summary feature is being used more than 16 million times a month, nearly three times the level reported a few months earlier. More than 200 organizations use its AI for professional billing coding. Those numbers cover broad categories and come from the vendor, but they are far beyond pilot volume.

The Providence study gives a useful reality check. A system can offer ambient AI widely and still see only a minority become active users. Local templates, specialty fit, training, consent rules and confidence in output quality determine whether a license turns into routine use.

Hospitals are scaling tasks with clear review points: notes, chart summaries, coding, patient messages, scheduling and prior authorization. Diagnostic and treatment systems face slower expansion because an error is harder to catch and carries greater consequences.

If you want more recent data on this point, please see our latest healthcare AI market report.

Do doctors trust AI enough to follow it?

Doctors trust AI enough to use its work, while keeping a firm grip on the final decision.

Doximity found that 71% of physicians worried about accuracy and reliability. Forty-seven percent said their institution's AI decision process was still evolving, and only 8% described the rules as clear. The AMA found that 88% wanted trusted safety and efficacy validation, 86% prioritized privacy assurances and 85% wanted physicians involved in adoption decisions.

Governance is trailing actual use. An earlier UK survey of 1,006 general practitioners found that one in five had already used generative AI in clinical practice, including for documentation and differential diagnosis, even while policies were unclear. Today's larger adoption figures make that gap more urgent.

You can see the caution in what doctors choose to delegate. They readily edit a note or check a research summary, then slow down when software interprets pathology, recommends treatment or changes a diagnosis. Hospitals can accelerate adoption by placing the tool inside approved systems, logging its actions and making verification easy.

Could doctors lose skills by relying on AI?

Yes, skill loss is a credible risk, especially for trainees who may learn medicine with AI support from the beginning.

The AMA found that 88% of physicians had at least some concern about skill loss. Seventy percent were very or somewhat concerned about medical students and residents. Strong concern about personal skill loss was lower at 28%, rising to 35% among early-career doctors.

Experienced doctors can compare an AI answer with years of clinical pattern recognition. Trainees may meet the answer before they have built enough knowledge to challenge it.

The diagnostic trials add an important clue. Simply giving doctors access to a strong model produced mixed results, while the Pakistan study paired access with 20 hours of AI-literacy training and found a much larger gain. The studies differ in several ways, but the contrast is hard to ignore. The human workflow may matter as much as the model.

Training remains weak. The AMA reported that one-quarter of physicians had received no AI training from any source, while 92% wanted more. Medicine is adopting the tool faster than it is teaching people how to use it.

Is AI replacing doctors today?

AI is changing doctors' jobs today; handling the whole job remains far beyond available systems.

Available systems handle pieces of work: searching evidence, drafting notes, summarizing charts, translating instructions, prioritizing scans, suggesting codes and proposing diagnostic possibilities. Doctors still combine the history, examination, patient preferences, uncertainty and accountability into a final plan.

The deployment pattern confirms it. The fastest-growing tools produce drafts or alerts that a clinician can review. Even radiology AI usually highlights findings or changes worklist priority while the radiologist signs the report. Clinical deployments still keep doctors responsible for the decision and the consequences.

Automation can still reshape staffing. Hospitals may need fewer hours for transcription, coding, chart abstraction and routine correspondence. Doctors may also be pushed to see more patients once managers expect AI to save time. The evidence so far supports modest efficiency gains, so aggressive productivity targets would run ahead of what the studies have shown.

So, are doctors actually using AI today?

Yes, doctors are genuinely using AI today, and the claim is mostly true rather than hype.

The strongest surveys place active professional use somewhere around 60% to 70%, with variation by sample and definition. More than one-third of physicians in Doximity's study used AI daily, and adopters increasingly use several applications. Search, summarization and documentation are already normal parts of work for a substantial share of doctors.

Everyday use still clusters around search and documentation, while diagnosis trails well behind. Large health systems are recording millions of AI interactions, yet the newest pragmatic trial found no significant improvement in treatment failure.

Our judgment is direct: AI has become a real workplace technology for doctors. It helps them prepare, read, write, organize and review far more often than it decides. Adoption is established; autonomous medical practice and broad outcome gains remain unproven.

If you want more recent data on this point, please see our latest healthcare AI market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.