What will AI do to the healthcare industry?

Where AI is already changing healthcare in 2026, what the evidence shows about scribes, intake and chatbots, and what AI will not replace.

Family physician listening to an older patient in a clinic exam room

It is one of the most searched questions in health right now, and most answers swing between two extremes: AI will replace doctors, or AI is all hype. The real picture in late 2026 is less dramatic and a lot more useful. AI is already changing healthcare, mostly in the places patients never see.

The short answer

Over the next few years, AI will do four things to healthcare:

Shrink paperwork. Documentation, forms and inbox work are the first jobs AI is taking on at scale.
Change the front door. Phones, booking and the history a patient gives before a visit are moving to AI.
Put a second opinion in every pocket. Patients and clinicians both now have AI tools that answer medical questions in seconds.
Force new rules. Regulators, privacy commissioners and medical colleges are writing guardrails as fast as the tools arrive.

What it will not do anytime soon is replace the clinician. The evidence below shows why.

AI is already in the exam room

This is not a future scenario. In the American Medical Association's 2026 survey of nearly 1,700 physicians, 81% said they use AI professionally, more than double the 2023 rate. The most common uses were summarizing research, drafting discharge instructions and notes, and documentation.

Canada is further behind but moving quickly. A C.D. Howe Institute analysis published in April 2026 found that 46% of Canadian physicians use AI in some form, and 68% of those with access to AI scribes use them often.

1. Paperwork is the first thing to shrink

The biggest, fastest impact is on administrative work, because that is where the waste is largest. A 2026 report from the Canadian Federation of Independent Business and the Canadian Medical Association estimated that unnecessary administrative work costs Canadian physicians 20 million hours a year, the equivalent of 9,000 full-time doctors.

AI scribes, which listen to a visit and draft the clinical note, are the clearest example of AI taking some of that back:

At Kaiser Permanente, 7,260 physicians used AI scribes across more than 2.5 million encounters, saving an estimated 1,794 working days in one year.
Canada Health Infoway's national AI scribe program enrolled more than 12,000 primary care clinicians. Users saved about an hour a week during work hours and nearly another hour after hours, and the share of notes finished the same day rose from about 56% to about 75%.
A randomized trial at UCLA, published in NEJM AI, found more modest gains: about a 10% cut in note-writing time with one of the two tools tested, plus a small improvement in burnout.

The gains are real, but so are the errors. When Ontario's Auditor General reviewed testing of the 20 AI scribe vendors approved for the province, 9 had invented information in simulated visits, such as referrals that were never discussed. That is why every serious guideline insists a clinician reviews every AI-written note.

2. The front door of healthcare is changing

For many patients, the hardest part of healthcare is getting in: reaching the clinic, booking a visit and explaining the problem. AI is starting to handle that first contact.

AI phone agents can answer every call, book appointments and route urgent concerns. In a 90-day pilot at Mount Sinai covering 701 patients, 86.4% of calls were completed and 94.9% of patients were satisfied, though the study was small and had no comparison group.

AI is also starting to take the patient's history before the visit. In a study published in The Lancet in October 2026, 98 patients at Beth Israel Deaconess chatted with Google's research system AMIE before seeing a doctor. Its list of possible diagnoses included the final diagnosis 90% of the time, and clinicians said the summaries helped them prepare in 75% of cases. It is still a research tool, and the authors call for larger trials, but it points clearly to where intake is heading. This is the space AI co-clinicians like Athel work in: a structured interview before the visit, reviewed by the physician, so the appointment starts with understanding rather than paperwork.

3. Patients are bringing AI with them

Patients are not waiting for their doctors to adopt AI. OpenAI says more than 300 million people a week ask ChatGPT health questions, and in July 2026 it opened ChatGPT Health, which can connect to medical records and wellness apps, to US adults on its free and paid plans. In a KFF poll in early 2026, 32% of US adults said they had used an AI chatbot for health information in the past year.

Canadians are doing the same. In a CMA survey of 5,000 adults, about half said they use AI tools to help diagnose or treat health issues, but only 27% trust AI to give accurate health information, and people who followed AI health advice were five times more likely to report harm.

The gap between what AI knows and how people use it is the big lesson so far. In a randomized study from the University of Oxford, published in Nature Medicine in February 2026, AI models identified the right condition about 95% of the time when tested on their own. But when ordinary people used those same models, they named a relevant condition in fewer than 35% of cases, no better than people who did not use AI at all.

4. Clinicians get an always-on research assistant

Tools built for clinicians are spreading fast. OpenEvidence, which answers questions from medical literature, is free for verified clinicians and reported about 18 million consultations a month by late 2025. OpenAI launched ChatGPT for Clinicians in April 2026, free for verified US physicians, nurse practitioners, physician assistants and pharmacists.

Research systems go further. Microsoft reported in 2025 that its MAI-DxO system solved 85.5% of 304 difficult published cases, compared with 20% for doctors working without books or colleagues. Microsoft itself describes it as a research demonstration that is not approved for clinical use, which is the right way to read every headline like it.

5. The rules are catching up

Regulators are drawing lines around what counts as a medical device. The US FDA updated its guidance on clinical decision support software in January 2026, and Health Canada has published guidance for machine learning-enabled medical devices. In Ontario, the Information and Privacy Commissioner issued guidance on AI scribes, and the College of Physicians and Surgeons of Ontario has made clear that the physician remains ultimately accountable for any AI-generated note.

Safety groups are paying attention too. ECRI, an independent patient safety organization, ranked misuse of AI chatbots as the number one health technology hazard for 2026.

What AI will not do

AI will not replace the judgment, accountability or relationship at the centre of care. The Oxford study shows that even accurate models do not automatically lead to better decisions, and the scribe audits show that AI still makes mistakes a human has to catch. The most realistic future is one where AI takes the paperwork, the phone queue and the information hunting, and gives clinicians more time for the part only they can do.

The money suggests this shift is only starting. One market research firm, Grand View Research, estimates the AI in healthcare market at about US$36.7 billion in 2025 and projects it could reach more than US$500 billion by 2033.

What this means for you

If you are a patient, use AI to prepare questions and understand information, then confirm anything important with a clinician.
If you are a clinician, start where the evidence is strongest: documentation and administrative work, with a human reviewing every output.
If you run a clinic, ask every AI vendor where patient data is stored, who can see it, and what evidence they have from real use, not just benchmarks.