Many clinicians are already using ChatGPT — and Claude, and Gemini — to answer clinical questions. They are fast, they take plain English, and they are always to hand. The honest short answer to "can I use them?" is: they are genuinely useful for learning, explaining concepts and drafting non-clinical text, but they are not built to be a clinical reference at the point of care, and they are not an appropriate place for patient-identifiable data. Here is why, and what is purpose-built for the job.
Why doctors reach for general-purpose AI
The appeal is obvious. A general-purpose chatbot answers in seconds, in natural language, without navigating a database or constructing a search. For a quick conceptual reminder or to rephrase something, that is a real convenience. The problem is not that the technology is weak — the underlying models are strong. The problem is what is, and is not, built around them.
Three reasons general-purpose AI is not a clinical reference
1. It is not grounded in current UK guidance. A general chatbot answers from its training, not from a live, curated set of UK sources. It will not reliably reflect the latest NICE, CKS or SIGN position, it can be confidently wrong (the familiar "hallucination" problem), and by default it does not show you a source you can verify. It may also default to US guidance or US-licensed medicines, which is a poor fit for UK practice. An answer you cannot trace is an answer you cannot stand behind.
2. It is not a medical device. General-purpose AI carries no clinical intended-purpose statement, no clinical risk-management process, and no regulatory accountability for clinical use. That is not a criticism of the tool — it was never designed or certified for that job. It simply means the governance a clinician should expect around a clinical tool is not there.
3. Information governance. Entering patient-identifiable information into a public, consumer AI tool raises real issues under UK GDPR and the common-law duty of confidentiality. As a rule of thumb, patient details do not belong in a consumer chatbot. (More on this in our piece on the information-governance question every clinician should ask.)
It is not the model — it is what is built around it
A useful way to think about this: the same underlying AI models can power a general consumer chatbot or a purpose-built clinical tool. Heidi Evidence, for instance, is built on Anthropic's Claude models — but as a clinical product with its own sourcing, citations and governance, not as the consumer Claude app. The difference between "AI you can use clinically" and "AI you should not" is rarely the model. It is the grounding in trusted current sources, the source-linked citations, the defined intended purpose, and the governance wrapped around it.
What "good" looks like for a clinical question
For point-of-care reference, the bar a clinician should hold is straightforward:
- Answers grounded in current, named UK sources — and the source shown, so you can verify it.
- A clear sense of what the tool is for: reference and retrieval that informs your decision, not a black box that makes it.
- UK relevance: NICE, CKS, SIGN and the SmPC, not generic or US-default content.
- Sensible handling of your patient's data.
Where iatroX fits
This is the gap AskIatroX is built for. It is a UK-registered, MHRA-listed Class I clinical reference tool, grounded in NICE, CKS, SIGN and the SmPC, that returns source-linked answers a clinician can verify — and its core clinical reference is free to use. Unlike a general chatbot, it is designed to inform the clinician's decision rather than make it, which is exactly why it is built and registered as a reference tool. For exam revision and reasoning practice, the same platform includes tutor and question-bank workflows.
None of this means general-purpose AI has no place in a doctor's life — for learning, brainstorming and non-clinical writing, it is excellent. It means that for a clinical question about a real patient, a purpose-built, source-linked, UK tool is the safer instrument.
Frequently asked questions
Can doctors use ChatGPT for clinical decisions? General-purpose AI should not be relied on for clinical decisions about real patients. It is not grounded in current UK guidance by default, does not reliably cite verifiable sources, and is not a regulated clinical tool. It can be useful for background learning, but the decision and its verification remain the clinician's, ideally using a purpose-built reference source.
Is it safe to put patient information into ChatGPT? As a general rule, no. Entering patient-identifiable data into a public consumer AI tool raises issues under UK GDPR and the duty of confidentiality. Clinicians should follow their organisation's information-governance policies and avoid entering identifiable patient data into consumer tools.
Is ChatGPT accurate for medical questions? It can be plausible and often broadly correct, but it can also be confidently wrong, may reflect outdated or non-UK guidance, and does not show a verifiable source by default. For clinical use, accuracy you cannot check is not enough.
What is a UK alternative to ChatGPT for clinical questions? Purpose-built clinical reference tools grounded in UK sources are the better fit. AskIatroX, for example, is a free, UK-registered clinical reference tool grounded in NICE, CKS, SIGN and the SmPC, returning source-linked answers for the clinician to verify.
Can I use ChatGPT, Claude or Gemini for exam revision? For general learning and explanation they can help, but they are not exam-specific and can be inaccurate on detail. Purpose-built question banks and tutors — such as iatroX's — are designed for exam preparation and clinical reasoning practice.