Clinicians are using AI for medical questions. The GMC's own research confirms it — over a quarter of UK doctors have used some form of AI in practice. The question is no longer whether AI is being used but which tool is safest and most useful.
This article compares the four AI tools clinicians most commonly reach for — ChatGPT, Claude, Perplexity, and iatroX — across the dimensions that matter for clinical use.
The Comparison Framework
Accuracy. Does the tool give the correct clinical answer?
Citations and provenance. Does it show where the answer came from? Can you verify?
UK guideline grounding. Does it reference NICE, CKS, SIGN, and BNF — the sources that govern UK practice?
Hallucination risk. How often does it generate plausible but incorrect information?
Cost and access. Is it free? Does it require verification or institutional access?
Regulatory status. Is it classified as a medical device or clinical tool?
Additional clinical features. Does it offer learning, reasoning support, or CPD?
ChatGPT (OpenAI)
What it is: The world's most popular general-purpose LLM. Over 200 million weekly users. Free tier available; premium plans offer GPT-4o access.
Accuracy: Variable. Performs well on standard medical knowledge questions but can generate confidently wrong answers, particularly for dosing, interactions, and jurisdiction-specific guidelines. Published research shows hallucination rates of up to 83% when clinical vignettes contain planted errors.
Citations: Poor. ChatGPT generates plausible-looking references that frequently do not exist. Published studies show over 45% of AI-generated references had fabricated DOIs, authors, or publication dates.
UK grounding: Weak. ChatGPT does not reliably distinguish between UK, US, Australian, and European guidelines. A UK GP asking about hypertension management may receive a US-centric answer without any indication that the wrong guideline system is being referenced.
Hallucination risk: High. This is an architectural limitation, not a fixable bug. ChatGPT generates text from statistical patterns, not from verified sources.
Cost: Free tier available. GPT-4o requires Plus subscription ($20/month).
Regulatory status: Not a medical device. Not UKCA-marked, MHRA-registered, or FDA-cleared.
Additional features: None clinical. General-purpose tool.
Verdict: Useful for non-clinical tasks (writing, brainstorming, admin). Not safe as a primary clinical reference. Every clinical output must be independently verified.
Claude (Anthropic)
What it is: Anthropic's AI assistant. Known for nuanced reasoning and longer context handling. Free tier available; Pro plan at $20/month.
Accuracy: Generally strong on medical knowledge, with a tendency to be more cautious and qualified than ChatGPT. Less prone to confident wrongness — Claude more often acknowledges uncertainty.
Citations: Moderate. Claude is more transparent about its limitations and less likely to fabricate references, but it still does not retrieve from verified medical databases. Its citations, when provided, should still be verified.
UK grounding: Weak. Same limitation as ChatGPT — not grounded in UK-specific guideline databases.
Hallucination risk: Moderate. Lower than ChatGPT on many benchmarks due to more cautious calibration, but still present.
Cost: Free tier. Pro at $20/month.
Regulatory status: Not a medical device.
Additional features: None clinical.
Verdict: Better reasoning quality than ChatGPT for complex questions. Still not a clinical reference tool. Same verification requirements apply.
Perplexity
What it is: An AI-powered search engine that combines LLM generation with real-time web search and citation.
Accuracy: Variable. Perplexity searches the web and synthesises answers with inline citations. The accuracy depends on the quality of the sources it finds — which may include authoritative guidelines or patient-facing health content of variable quality.
Citations: Strong relative to ChatGPT and Claude. Perplexity provides inline citations with clickable links to actual web sources. This is a genuine advantage — you can verify.
UK grounding: Moderate. Perplexity can find UK sources if they rank well in web search. But it does not preferentially retrieve from NICE, CKS, or BNF — it retrieves from whatever ranks highest, which may be US-centric or non-authoritative.
Hallucination risk: Lower for factual claims (because it retrieves from web sources) but the sources themselves may be unreliable.
Cost: Free tier. Pro at $20/month.
Regulatory status: Not a medical device.
Additional features: None clinical.
Verdict: The best general-purpose option for clinicians who want cited answers. But citation to a web source is not the same as citation to an authoritative guideline. Source quality varies.
iatroX
What it is: A UK-focused AI clinical reference platform using RAG over a curated corpus of NICE, CKS, SIGN, BNF guidelines, and peer-reviewed research.
Accuracy: High for UK clinical questions. Every answer is retrieved from verified guideline content, not generated from training data. The RAG architecture means the AI synthesises from known sources rather than predicting plausible text.
Citations: Excellent. Every answer includes inline citations linking directly to the primary NICE, CKS, SIGN, or BNF source. One click to verify.
UK grounding: This is the core design principle. iatroX is built around UK guidelines. Every answer reflects UK practice, UK drug names, UK referral pathways, and UK prescribing conventions.
Hallucination risk: Low. RAG-grounded synthesis from curated sources produces fundamentally fewer hallucinations than open-web generation. The risk is not zero — edge cases and synthesis errors can occur — but the profile is qualitatively different from general-purpose LLMs.
Cost: Completely free. No subscription, no trial period, no institutional login, no professional verification required.
Regulatory status: UKCA-marked and MHRA-registered for its UK guideline features. No other tool in this comparison has this status.
Additional features: Knowledge Centre for structured guideline browsing, Brainstorm for clinical reasoning, Q-Bank with spaced repetition for learning, CPD module for professional development.
Verdict: The strongest option for UK clinicians who need guideline-grounded, citation-first clinical answers. Free, UK-specific, and designed for clinical use.
The Recommendation
For UK clinical questions: iatroX. It is the only tool in this comparison that is architecturally grounded in UK guidelines, UKCA-marked, and designed for clinical use. Free.
For international clinical context: Perplexity offers the best citation quality among general-purpose tools. Verify every source.
For non-clinical tasks: ChatGPT or Claude for writing, brainstorming, and administrative work.
For any clinical decision: Verify against the primary source — NICE, CKS, BNF — regardless of which AI tool you used. The habit of checking is more important than the tool you choose.
Conclusion
The best AI for medical questions in 2026 depends on what you are asking, why, and where you practise. For UK clinicians, the answer is clear: a tool grounded in UK guidelines, with visible citations, regulatory status, and additional learning features. iatroX provides all of these, for free. Start there.