AI in healthcare: how clinics are using it now
Short answer Most doctors now use AI: 81% in the AMA's 2026 survey, up from 38% in 2023. Clinics use it mainly for notes, chart summaries, research lookups and patient-message drafts. In the US, a HIPAA-covered clinic needs a business associate agreement with any AI vendor that handles patient data, and California and Texas require telling patients in some cases.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callHow clinics use AI today
Mostly for paperwork, not diagnosis. In the AMA's 2026 survey of nearly 1,700 US physicians, 81% said they use AI at work, more than double the 2023 rate (AMA, March 2026). The top uses:
| Use | Share of physicians |
|---|---|
| Summaries of medical research and standards of care | 39% |
| Discharge instructions, care plans or progress notes | 30% |
| Billing codes, medical charts or visit notes | 28% |
| Chart summaries | 28% |
| Draft replies to patient portal messages | 19% |
| Translation | 18% |
| Help with diagnosis | 17% |
One newer tool is the ambient AI scribe: an app that listens to the visit and drafts the clinical note for the clinician to check. On the business side, clinics also use AI to answer phones, book appointments and reply to messages; the privacy rules below apply to those tools too.
Diagnostic AI is real, but it's a regulated product, not a chat app. The FDA has authorized over 1,600 AI-enabled medical devices as of September 2026 (FDA: AI-Enabled Medical Devices).
Does AI actually help a clinic?
For note-taking, the best evidence so far says yes. A 2025 study in JAMA Network Open followed 263 clinicians at six US health systems who used an AI scribe for 30 days. Burnout fell from 51.9% to 38.8%, and after-hours documentation dropped by almost an hour (Olson et al., 2025).
Read it with care: there was no control group, and two authors worked for the scribe vendor. Physicians are cautious too. In the AMA survey, 88% had some concern that relying on AI could erode clinical skills, and 86% called data privacy critical before they would use it more widely.
What AI doesn't change: the clinician who signs the note is responsible for it. Texas writes that into law (below).
The US rules for AI in a clinic
Five rules matter to a typical practice. Most predate the AI boom; they just apply to new tools.
| Rule | Who it covers | What it requires |
|---|---|---|
| HIPAA business associate rule | Clinics covered by HIPAA | Any AI vendor that creates, receives, stores or sends patient data for you signs a business associate agreement (BAA) first |
| Section 1557, 45 CFR 92.210 | Clinics that receive federal financial assistance from HHS | Reasonable efforts to find and reduce bias in decision support tools that use race, color, national origin, sex, age or disability. In force since May 1, 2025 |
| FDA device rules | Makers of diagnostic and treatment software | Tools that diagnose or direct treatment may need FDA authorization; check that yours has it |
| California Health and Safety Code 1339.75 | California clinics and physician offices | A notice on AI-written patient messages about clinical information, unless a licensed provider reviewed them |
| Texas Health and Safety Code 183.005 | Texas practitioners | Review AI-made records, and tell patients when you use AI for diagnosis. In force since September 1, 2025 |
HIPAA. HHS's cloud guidance settles the vendor question:
"When a covered entity engages the services of a CSP to create, receive, maintain, or transmit ePHI (such as to process and/or store ePHI), on its behalf, the CSP is a business associate under HIPAA." (HHS: Guidance on HIPAA and cloud computing)
An AI scribe, chatbot or phone agent that handles patient data is that kind of vendor. Which ChatGPT plans qualify is in is ChatGPT HIPAA compliant.
Section 1557. "Decision support tool" is broad: any tool "used by a covered entity to support clinical decision-making," automated or not (45 CFR 92.4). The duty is ongoing:
"A covered entity has an ongoing duty to make reasonable efforts to identify uses of patient care decision support tools in its health programs or activities that employ input variables or factors that measure race, color, national origin, sex, age, or disability." (45 CFR 92.210(b))
FDA. In January 2026 the FDA issued new final guidance on clinical decision support software: which tools that advise clinicians are not regulated as devices. A scribe that writes notes isn't a diagnostic device; software that flags a stroke on a scan is. Ask the vendor for its FDA authorization if it claims to diagnose.
California. The disclaimer must say the message was "generated by generative artificial intelligence," and come with "clear instructions describing how a patient may contact a human health care provider" (California Health and Safety Code 1339.75). Scheduling and billing messages are exempt; so is any message a licensed provider read and reviewed.
Texas. The 2025 law is short:
"A health care practitioner who uses artificial intelligence for diagnostic purposes as described by Subsection (a) must disclose the practitioner's use of that technology to the practitioner's patients." (Texas Health and Safety Code 183.005(b))
Other states have passed or proposed their own rules; check your state medical board.
Patient privacy: what to check before you switch an AI tool on
- Does it touch patient data? A tool that drafts blog posts doesn't. A scribe, intake chatbot or phone agent does.
- Will the vendor sign a BAA? No BAA, no patient data. In Canada, look for a written agreement that meets your provincial health privacy law.
- Where does the data go, and for how long? Ask whether recordings are deleted after the note is made, and whether your data trains the vendor's model.
- Do patients know? California and Texas require telling patients in the cases above. In Ontario, read the privacy commissioner's AI scribe checklist before you record visits.
- Who reviews the output? Name the person who checks every AI note or message before it reaches a chart or a patient.
Your website has its own privacy trap: tracking pixels on booking pages. That's covered in clinic website tracking and HIPAA.
Canada: AI rules for clinics
Two sets of rules matter to a Canadian clinic.
- Devices: Health Canada's pre-market guidance covers machine learning-enabled medical devices in Classes II to IV. As in the US, a scribe is not a device; diagnostic software can be.
- Privacy: provincial health privacy laws, such as Ontario's PHIPA, apply to any AI tool that handles patient records. On January 28, 2026, Ontario's privacy commissioner issued guidance and a checklist for AI scribes, and British Columbia's commissioner released its own the same day.
"Health providers must put in place the core privacy, governance, and accountability measures needed before introducing new AI tools into their practice." (Patricia Kosseim, Information and Privacy Commissioner of Ontario, January 28, 2026)
What changed since 2024
Older articles on this topic describe rules that are gone or rules that arrived since.
| Older claim | Status in November 2026 |
|---|---|
| A US executive order sets AI rules for healthcare | Executive Order 14110 was revoked on January 20, 2025 (Executive Order 14148) |
| HHS may regulate AI in the future | Section 1557's decision support tool rule has applied since May 1, 2025 |
| No state requires telling patients about AI | California (clinical messages) and Texas (diagnostic use) do |
| Adoption figures from 2023 or 2024 (38% of physicians in 2023) | 81% in 2026, per the AMA |
| "AI will save the US $150 billion a year by 2026" | A forecast, not a measured result; no current source backs it |
Example
Illustrative cases with round numbers, not client results.
A three-doctor family practice in California
It tries an AI scribe at $300 per doctor per month. The vendor signs a BAA and deletes audio after the note is made. Doctors review every note before signing. Portal replies drafted by AI go out only after a doctor reads them, so no AI disclaimer is needed. Unreviewed messages would need one.
A physiotherapy clinic in Ontario (Canadian dollars)
It wants an AI phone agent at C$400 a month to book visits after hours. It confirms the vendor stores data under a written agreement that meets PHIPA, tells callers they're speaking with AI, and keeps clinical questions for staff.
A dermatology practice in Texas
A skin-check app suggests likely diagnoses from photos. The practice checks that the app is FDA-authorized, tells patients it uses AI in diagnosis, and has the dermatologist confirm every finding.
FAQ
Will AI replace doctors?
Not under current rules. Laws like Texas's require a licensed practitioner to review AI-made records, and 85% of physicians in the AMA survey want a say in how AI is adopted.
Is it legal to use ChatGPT in a medical practice?
Yes, for tasks with no patient data, like drafting a blog post. With patient data, only plans that come with a signed BAA qualify.
Do I have to tell patients I use AI?
It depends on where you are. California requires a notice on unreviewed AI messages about clinical information; Texas requires disclosure when AI is used for diagnosis. Elsewhere, telling patients is still the safer default.
What is an ambient AI scribe?
An app that listens to a visit, with consent, and drafts the clinical note. The clinician edits and signs it. It handles patient data, so it needs a BAA in the US.
Different for you?
- You want to use ChatGPT with patient messages: is ChatGPT HIPAA compliant
- You're a dental office looking at an AI receptionist: AI receptionist for a dental office
- You run ads or analytics on your clinic site: clinic website tracking and HIPAA
- You reply to patient reviews: HIPAA review responses
- You want patients to find your practice in AI answers and on Google: SEO & GEO Agent