I'm Lilach Bullock, and my answer on AI for doctors starts with one question: which class of tool is it? Consumer chat is fine for writing that holds no patient data, clinician assistants sold on enterprise health terms can draft notes with review, and software with a medical purpose may be a regulated device. Patient details never go into a personal account.
Most of the questions doctors ask about AI come down to the same worry: can I use this with patient information? The answer depends far less on which brand you pick than on the kind of product you're using and the agreement your practice has with the vendor.
I'm not a clinician and nothing here is medical advice. What I do is help teams adopt AI safely, and the order matters: the rule should be written before the tool gets bought. Below I've sorted the AI products doctors hear about into classes, using the vendors' own pages and the UK regulators' pages as of October 2026, and set out the data rule a practice should agree first.
Three classes of AI tool, plus one for patients
Here's the table I'd put in front of a partners' meeting. The examples come from the vendors' own pages. The point isn't to rank the products. It's to agree which class is approved for which data.
| Tool class | Example from the vendors' own pages | Use it for | Keep it away from |
|---|---|---|---|
| Consumer chat | ChatGPT, Claude (personal accounts) | Drafting non-patient documents, policy wording, patient leaflets from public information | Anything identifying a patient |
| Patient-facing health product | ChatGPT Health (US adults connect their own records) | Patients understanding their own results | Clinician workflow |
| Clinician assistant under an enterprise health agreement | Claude for Healthcare (HIPAA-ready products), Microsoft Dragon Copilot | Notes, coding support, prior authorisation and appeals drafting, with clinician review | Autonomous diagnosis |
| Regulated medical device (UK) | Software or AI with a medical purpose | Only if the maker holds the right classification | Home-built workarounds |
ChatGPT Health is built for patients, not your practice
OpenAI introduced ChatGPT Health on 7 January 2026 as a dedicated health experience for individuals, who can connect their medical records and wellness apps. OpenAI says over 230 million people globally ask health and wellness questions on ChatGPT every week, and it describes the product as designed to support, not replace, care from clinicians. An update dated 23 July 2026 says Health in ChatGPT is launching to US users aged 18 and older on web and iOS.
That matters for you because your patients will arrive having already asked it. It doesn't make ChatGPT Health a clinical tool. It's a consumer product for patients, and the enterprise health offers are sold separately to providers and payers. A doctor typing patient details into a personal chat account is using neither. That's the gap most practices haven't closed.
What the clinician assistants say they do
Anthropic introduced Claude for Healthcare on 11 January 2026: tools for healthcare providers, payers and health tech companies to use Claude through what it calls HIPAA-ready products. Organisations that are HIPAA-compliant can use Claude for Enterprise, and the connectors include CMS coverage data, ICD-10 and the National Provider Identifier Registry. Anthropic edited the page on 7 February 2026 to clarify that the HIPAA-ready products are specific to healthcare providers and payers, while the personal health integrations are for individual users. Read that line twice, because it's the whole distinction in one sentence.
Microsoft describes Dragon Copilot as an AI clinical assistant for physicians, nurses, radiologists and other professionals. On Microsoft's own page it covers:
- Ambient capture of multi-party, multilingual conversations at the point of care, turned into specialty-specific notes.
- Answers drawn from transcripts, notes and third-party medical references, with citations.
- Task automation, and integration with electronic health records.
UK practices: the device question and data protection
In the UK, the MHRA says software, including AI, plays an essential part in health and social care, and that many products are regulated as medical devices. Its Software and AI as a Medical Device Change Programme is reforming how products are qualified and classified and how they're monitored after they reach the market, including transparency, explainability and how AI that's retrained over time is handled. The page was last updated on 3 February 2025. The practical point for a practice: if a tool has a medical purpose, ask the maker how it's classified before anyone builds a workaround around it.
On data, the ICO's AI guidance hub offers Guidance on AI and data protection, Explaining decisions made with AI, and an AI and data protection risk toolkit for assessing risks to individual rights from an organisation's own AI systems. I'd run that toolkit before any AI product touches patient records. My guide to GDPR-compliant AI tools for UK businesses covers the wider data questions.
The data rule to agree before you buy a scribe
Write this before a demo, not after. It's short, and it saves arguments later:
- Name the classes: list which tools sit in each row of the table above, and who approved them.
- Match data to class: patient-identifying information only in tools covered by an enterprise health agreement your practice has signed.
- Ban personal accounts for patient work: no patient details in personal chat accounts, ever, however convenient it feels at the end of clinic.
- Keep the clinician in charge: every AI-drafted note, letter or appeal is reviewed by the clinician before it's filed or sent.
- Check the device question: anything with a medical purpose needs the maker's classification, not a home-built prompt.
- Run the risk check: use the ICO toolkit, or your own regulator's equivalent, before patient records go anywhere new.
Where I come in
I'm not a healthcare consultant and I don't pick clinical software. I help teams turn AI from scattered experiments into rules and workflows people follow. I've been in marketing since 2003, I built, scaled and sold a digital marketing agency, and I've spoken at over 100 events internationally. The teams I work with are mostly SaaS marketing teams, telecom companies and small businesses, and for a practice the useful services are:
- AI training and workshops for teams: a live session for clinicians and admin staff that ends with the data rule agreed.
- AI enablement: getting the approved tools set up so the rule is easy to follow.
- AI automation: taking non-clinical admin off the front desk, built in Make.com, Zapier or n8n.
- GEO and AI visibility: checking how ChatGPT and other assistants describe your practice when patients ask.
My advice for doctors and practice owners
Sort before you shop. Agree which class of tool each kind of data may go into, write it on one page, and make sure every clinician and receptionist knows it. Use consumer chat for writing that holds nothing about a patient. Look at clinician assistants only on enterprise terms your practice has signed, and keep a clinician's review on every output. In the UK, treat the MHRA's device rules and the ICO's toolkit as the two checks that come before any purchase.
If you're comparing products, my roundup of the best AI tools for healthcare covers the options. If you'd like help writing the rule and training your team on it, book a free 30-minute call through my work with me page. If it's not a fit, I'll tell you before we hang up.
Frequently asked questions
Can doctors use ChatGPT with patient information?
Not in a personal account. As of October 2026, OpenAI's ChatGPT Health is a consumer product for patients. Patient-identifying information belongs only in tools covered by an enterprise health agreement your practice has signed.
What is ChatGPT Health?
A health experience OpenAI introduced on 7 January 2026 that lets individuals connect medical records and wellness apps. OpenAI says it's designed to support, not replace, care from clinicians.
Is Claude for Healthcare for doctors or patients?
Both, in different products. Anthropic says its HIPAA-ready products are specific to healthcare providers and payers, while its personal health integrations are for individual users.
Is AI software a medical device in the UK?
It can be. The MHRA says many software and AI products are regulated as medical devices, and its change programme is reforming how they're qualified and classified.
Who can train a practice team on safe AI use?
Lilach Bullock runs AI training and workshops that end with an agreed data rule for each tool class. Book a free 30-minute call through her work with me page.