AI, ChatGPT and your rheumatology care

Tools like ChatGPT, Gemini, Claude and Copilot are always available, answer instantly and explain things in plain language, with no waiting room and no sense of asking a silly question. It is no surprise that many people now use them as a first step when something does not feel right, or when a letter or result leaves them none the wiser.

This is a recognised and reasonable thing to do. The World Health Organization, in its guidance on AI in health, lists "patient-guided use, such as for investigating symptoms and treatment" as one of the legitimate ways these tools are used. The question is not whether to use AI, but how to use it well, and where its answers need checking.

At its best, AI is an excellent explainer. It can take dense, jargon-heavy medical language and turn it into something you can actually understand. In one 2024 study in JAMA Network Open, AI chatbots rewrote pathology reports from a level pitched at university graduates down to around that of a 12- to 14-year-old, while keeping most of the meaning intact. A 2025 study found the same for rheumatology specifically: large language models condensed complex articles into readable summaries that expert rheumatologists judged accurate.

In practice, that makes AI useful for:

• Translating test results, clinic letters and unfamiliar terms into plain English
• Explaining a condition or medication in general terms, before or after an appointment
• Helping you organise your symptoms and draft the questions you want to ask
• Offering calm, patient, round-the-clock information when worry strikes at 2am

Used this way, to understand and to prepare, AI can make you a more informed patient and your appointments more productive.

The limits matter just as much as the strengths, and they are not always obvious.

• It cannot examine you. A chatbot cannot feel a swollen joint, watch how you move, or examine you in person, and so much of rheumatology rests on that physical examination.
• It does not know your full story. It has no access to your history, your previous results, your other conditions or your medication list, all of which change what any single symptom or number actually means.
• It is not a regulated medical device. In the UK, software intended for a specific medical purpose is regulated by the MHRA, but a general-purpose chatbot is not assessed or approved in that way. It is a consumer tool, not a clinical instrument.
• It cannot take responsibility. As the British Medical Association puts it, AI is "only a tool". The duty of care, and the accountability for advice, rests with a clinician, not a chatbot.
• Accuracy is uneven. A large 2025 review of language models answering clinical questions found their accuracy varied widely by topic and specialty, that newer versions were not reliably better than older ones, and that human experts still outperformed them at reaching the correct diagnosis.

There is also a simple privacy point: think twice before pasting your full reports or identifying personal details into a public AI tool.

The most important thing to understand about AI is that it can be confidently wrong. These tools generate plausible-sounding text; they do not actually know whether what they are saying is true. When they state something false as though it were fact, it is often called a "hallucination", and in medicine it is a real risk.

How real? In a 2025 study in Communications Medicine, researchers slipped a single invented detail, a fake test or a fictional disease, into clinical questions. The leading AI models went along with the fabrication and built on it in 50% to 82% of cases, depending on the model. A safety-focused prompt reduced the problem but did not remove it.

References are no safeguard either. When researchers asked ChatGPT to support medical content with citations, one 2023 analysis found that 47% of the references were completely made up, another 46% were real but contained errors, and only 7% were both real and accurate. An answer that "cites its sources" can be just as wrong as one that does not.

This is also why the way you ask changes the answer you get. Research has shown that the phrasing of a prompt, and how unfamiliar or specialised the topic is, both push these tools towards inventing more. A leading question ("this is just normal aches and pains, isn't it?") can nudge the AI into agreeing with you rather than challenging the assumption. The more unusual or specialised your problem, and rheumatology is full of uncommon, easily-missed conditions, the more caution an AI answer deserves.

None of this means AI has no place in good medicine. In the right hands it is genuinely promising, and rheumatology is one of the areas where it is advancing. The British Society for Rheumatology, the UK's professional body for the specialty, has set out seven ways AI is beginning to change rheumatology, from imaging to clinical decision support.

• Imaging and ultrasound. AI is being developed to help grade inflammation on scans and to make musculoskeletal ultrasound more consistent, helping to reduce the operator-dependence that has always been ultrasound's main limitation.
• Bone health. AI can opportunistically flag low bone density on scans done for other reasons, and assist with fracture-risk assessment, though performance still varies and results need careful interpretation.
• Personalised treatment. Machine-learning models can estimate an individual patient's risk of a flare when reducing biologic medication, or how likely a particular biologic is to work, which can help tailor treatment.

Some of it is striking. An autonomous robot, the ROPCA ARTHUR system, built on a KUKA robotic arm, can now scan the small joints of the hand for inflammation on its own, with a companion AI reading the images. It carries a European CE mark and has been put into routine use in Danish hospitals. Even so, the evidence stays grounded: peer-reviewed trials in 2025 judged it promising but not yet ready to work unsupervised, with assessment by an experienced rheumatologist still well ahead. It is a real glimpse of the future, and, for now, a reminder that the specialist's judgement remains the benchmark.

The common thread is that these are decision-support tools, used and interpreted by a specialist. They make an expert's judgement sharper; they do not replace it.

AI in medicine still attracts caution, and sometimes outright scepticism. Dr Borukhson takes a more open view. She welcomes the genuine advances AI is bringing, embraces its usefulness for patients, and would far rather you arrived curious and well-informed than not ask at all.

The boundary she holds is simple and important. AI can inform a conversation, but it cannot replace the examination, the clinical judgement and the personal context that turn information into a safe, individual plan. Bring what you have read; she will help you make sense of it.

Do:

• Use it to understand terms and to prepare questions for your appointment
• Bring what it told you to your consultation, screenshots and all
• Cross-check anything that matters against trusted sources; NHS guidance is clear that you should discuss any worries about your health with a clinician rather than rely on what you find online

Don't:

• Don't use it to diagnose yourself, or to rule a condition in or out
• Don't start, stop or change any medication on its advice
• Don't paste full reports or identifying personal details into a public chatbot
• Don't mistake confidence for correctness, or assume that cited "sources" are real

If AI has raised a possibility that worries you, a result looks abnormal, or your symptoms are persistent, changing or simply unexplained, that is exactly the moment to see a specialist.

A rheumatology assessment offers what no AI can: a hands-on examination, point-of-care ultrasound where it helps, the right tests interpreted against your own history, and a plan that is genuinely yours. Dr Borukhson is glad to review what AI told you, separate the useful from the misleading, and give you clear, reliable answers. Bring your questions, and your screenshots.

Can ChatGPT diagnose my arthritis?

No. It can explain what arthritis is, what the different types are and which tests are used, which is genuinely useful. But it cannot examine your joints, scan them with ultrasound, or interpret your results in the context of your history. A diagnosis needs a clinical assessment.

Is it safe to ask AI about my medications?

For general understanding, yes. But never start, stop or change a dose based on AI. Many rheumatology medications interact with others and need monitoring, so any change should be checked with your clinician or pharmacist first.

Can I trust AI to interpret my blood test results?

Treat it as a rough guide at best. AI does not have your full picture, and the same number can mean very different things depending on your symptoms and history. Our guide to rheumatology blood tests explains the common ones, and it is always worth bringing your results to your appointment.

Should I bring what AI told me to my appointment?

Yes, please do. It is genuinely helpful, and Dr Borukhson would far rather discuss what you have read than have you worry about it alone. It often makes the consultation more focused and productive.

What is Dr Borukhson's view of AI?

She welcomes the advances AI is bringing to medicine and embraces genuinely useful tools, while keeping the boundaries that protect you: AI can support a clinician's work, but the examination, judgement and plan remain hers.