AI, ChatGPT and your rheumatology care

ChatGPT, Gemini, Claude, Copilot and their like never close, reply in seconds and put things in everyday words; nobody keeps you waiting, and no question feels too small to ask. Many people now turn to one first when something feels wrong, or when a letter or a result leaves them no wiser.

That is a recognised use. The WHO's guidance on these tools in health names patients using them for themselves, to look into symptoms and treatment, among five broad applications. So the real question is how to get the good out of these tools, and which of their answers to check.

At its best, AI explains. A 2024 study in JAMA Network Open gave 1,134 pathology reports to two chatbots and asked for each in simple terms: the reading level fell from about that of a first-year university student to a level a 12- to 14-year-old could follow, and most of the simplified reports were medically correct, though some held errors or invented details, and the authors advise that a clinician review such a report before a patient reads it. A 2025 study in rheumatology asked ChatGPT to rewrite twelve research articles for a lay reader: their reading level fell from one suited to a university graduate to that of a 15- or 16-year-old, and two rheumatologists judged the summaries accurate, though the authors caution that such summaries should not be anyone's only source of health information.

Used in that spirit, AI can help with:

• Turning a result, a clinic letter or an unfamiliar term into everyday words
• Giving a general account of a condition or a medicine, ahead of an appointment or after it
• Putting your symptoms in order and drafting the questions you want to ask
• Answering calmly at any hour, when a worry arrives at 2am

In that role, AI can leave you better informed and your time in the clinic better spent.

Its weaknesses matter just as much, and they are harder to spot.

• No examination. No chatbot can feel a swollen joint or scan you, and so much of rheumatology rests on what hands and a probe find.
• No history. It knows only what you tell it, not the whole of your past illnesses, earlier results, other conditions and medicines, and each of those changes what a symptom or a number means.
• Probably not a medical device. In the UK, much software made for a medical purpose, AI among it, is regulated by the MHRA as a medical device. The MHRA's view is that a language model aimed at general purposes, whose maker claims no medical use for it, is unlikely to count as one, and so unlikely to have been held to those rules.
• Only a tool. The British Medical Association calls AI only a tool, one that can improve care when it is used appropriately and worsen it when it is not. The duty of care, and the accountability for advice, rest with a clinician, not a chatbot.
• Uneven accuracy. A 2025 review in the Journal of Medical Internet Research, pooling 168 studies, found accuracy varying from one model and one kind of question to another, newer versions not necessarily doing better than older ones, and human experts still more accurate than the models at naming the most likely diagnosis, even where the models did well on exam questions.

Privacy matters too: a public AI tool is no place for whole reports, or for details that identify you.

Above all, AI can be wrong and sound entirely sure of itself. These tools build an answer by predicting likely words, not by checking facts, so text that reads as right may not be. Researchers call a fact a model has made up a "hallucination", and in a medical question it can do real harm.

How real? A 2025 study in Communications Medicine slipped a single invented detail, a laboratory test, a sign or a condition that does not exist, into each of 300 clinical cases. The six models tested repeated or built on the invention in 50% to 82% of their answers, depending on the model. An instruction to use only clinically validated information and to say when they were unsure brought the average down from about two thirds to under half, without removing the problem.

Nor do references make an answer safe. When a 2023 study asked ChatGPT, then on version 3.5, to write short medical papers with references, 47% of the 115 references were made up, 46% were real but held errors, and just 7% both existed and were cited correctly. A reply that lists its sources is no more trustworthy for doing so.

It follows that how you put a question shapes the reply. The Communications Medicine authors point to many of these models' tendency to be overly confirmatory, and in their study carefully worded instructions cut the errors without ending them. A leading question ("it's only wear and tear, isn't it?") invites agreement rather than challenge. The more complex your problem, and rheumatic diseases often are, varying widely in how they show themselves, the more caution an AI answer deserves.

AI still has a real place in good medicine. In rheumatology it is already in clinic, says the British Society for Rheumatology, which has set out seven ways it is changing the specialty, from imaging and remote assessment to the language models themselves.

• Imaging and ultrasound. The BSR reports AI already matching experienced clinicians at spotting changes on X-rays, and a 2023 review in Frontiers in Medicine describes work to make joint ultrasound, which depends on the skill of whoever holds the probe, more consistent and faster to score.
• Remote assessment. Systems that recognise hand movement are being tested to judge how active rheumatoid arthritis is without an in-person examination, which could in time make routine monitoring easier between clinic visits.
• Personalised treatment. Applied to large patient datasets, AI is uncovering groups of patients that were not previously recognised, which is expected to help refine how conditions are classified and how treatment is chosen for each person.

One example is striking. ARTHUR, a robot developed in Odense, Denmark, scans the small joints of the hand by itself, while a companion AI, DIANA, scores the images; both carry the CE mark. Svendborg Hospital became the first to use it in daily practice with patients, in February 2024. The evidence keeps it in proportion: in a 2025 trial, its agreement with a rheumatologist's own scan was moderate to good across the hand as a whole but low joint by joint, and the authors judged that version not yet ready for daily clinical practice; a second 2025 study, of a newer version, found its repeatability on a par with experts' and its agreement with a rheumatologist encouraging, with refinement still needed. A glimpse of what may come, then, and for now a reminder that it is measured against a specialist's own scan.

Caution about AI in medicine is common, and sometimes it hardens into outright scepticism. Dr Borukhson's view is more open: she welcomes what AI is bringing, sees its value for patients, and would far rather you came curious and well read than held back a question.

The boundary she keeps is a simple one. AI can feed a conversation, but it cannot stand in for the examination, the clinical judgement and the knowledge of you that turn information into a plan that is safe and your own. Whatever you have read is welcome in the consultation.

Do:

• Do ask it what a word in your letter means, or to help you list your questions
• Do show Dr Borukhson what it said; screenshots are welcome
• Do weigh anything important against a reliable source: NHS Knowledge and Library Services rates .nhs.uk sites the most reliable, advises talking over any worry about your health with your GP, and warns that not everything online is trustworthy, accurate or up to date

Don't:

• Don't treat its answer as a diagnosis, whether it reassures or alarms
• Don't alter a medicine, or its dose, because a chatbot suggested it
• Don't paste in your name, date of birth, NHS number or anything else that identifies you
• Don't take confidence for correctness, or assume that a cited "source" exists

If a chatbot has raised something that frightens you, a result seems out of range, or your symptoms carry on, change or remain unexplained, what you need next is a person: your GP, the clinician who ordered a test, or a specialist. A joint that has stayed swollen beyond a couple of weeks, or morning stiffness lasting over half an hour, is a reason to see a GP within days, or to book a specialist assessment directly, as our early inflammatory arthritis guide sets out. For a symptom that might not keep, the guide to your condition says under When to seek help which signs need help sooner and where each should go, and no chatbot's answer replaces it; where nothing has been named yet, our guides to joint swelling and joint pain without swelling hold those lists for the joints, and our conditions page lists the rest.

What an assessment adds is everything a chatbot lacks: hands that examine the joints, a scan in the consultation where it helps, tests read against your own history, and a plan made with you. If you cannot tell how urgent something is, asking is never a mistake.

Can ChatGPT or another chatbot tell me whether I have arthritis?

No. It can explain what arthritis is, the forms it takes and the tests used to tell them apart, which is useful background. It cannot examine your joints or scan them, and a diagnosis rests on the history, the examination and the results read together.

Can I ask AI about my medicines?

To learn what a medicine is and how it works, yes; for deciding anything, no. Whether to begin a medicine, stop it or alter the dose is for your doctor, your specialist team or a pharmacist, since many medicines used in rheumatology need regular blood tests.

Can AI read my blood test results for me?

Only roughly. It does not have the rest of the picture, and the same number can mean quite different things beside different symptoms and histories. Our guide to understanding blood tests in rheumatology explains the common tests, and the results themselves are worth bringing to your appointment.

Should I tell Dr Borukhson what AI told me?

Yes, please. Bring the answers, screenshots and all: talking through what you have read is far better than worrying over it alone, and it often helps the consultation get to what matters most to you.

How does Dr Borukhson see AI?

She is glad of the progress AI is making in medicine and of the tools that help. The line she holds is the one that protects you: a tool can support her work, but examining you, weighing what is found and agreeing the plan with you remain hers to do.