A chatbot cannot examine you
Millions of people are now typing their symptoms into a chatbot before calling a GP, and AI self-diagnosis has become a major health issue.
EY’s UK research found that 52% of respondents had requested, or would consider requesting, a test, treatment or prescription based on AI-generated information.
That figure shows how quickly trust is shifting. EY’s wider global survey found 56% of respondents said the same about AI, and 31% of those who delay care turn to AI tools instead.
So is AI helping people understand their bodies, or feeding anxiety and risky self-diagnosis?
The evidence points both ways, and the stakes are higher for communities that already face barriers in healthcare.
EY’s data also shows where self-directed care goes wrong. GLP-1 weight-loss drug users were nearly twice as likely to report negative outcomes after acting on AI health information (43% versus 23%), and only 37% of UK respondents obtained their prescription through a doctor.
Why People turn to AI First
The appeal is obvious: instant, free, private and available even at 2 am. A chatbot never rushes you and never asks you to wait, unlike any clinic in the UK.
AXA Health surveyed 2,000 UK adults and found that 78% of AI users said it helped them understand medical terminology, test results or treatment plans.
Some 68% said it made them more confident discussing symptoms with a clinician.
Medical letters are full of jargon, and a plain-language explanation can make an appointment far more productive.
Used this way, AI works as a translator of medical language rather than a doctor. The trouble is when people ask it to make the medical decision completely uninformed.
The Anxiety Loop
The same AXA research found that 37% of AI users felt more anxious after checking symptoms. Some 93% had used AI to check symptoms late at night, with many describing habits that spiral into more checking.
Anyone who has searched ‘headache’ and landed on a brain tumour will recognise the pattern. Chatbots can make it worse because they reply conversationally, which makes every worst case feel personal.
AXA’s headline finding was that AI is driving people both to delay care and to seek it unnecessarily.
Reassurance can keep one person away from a GP they need, while alarm can send another to book appointments they do not.
Accuracy is the bigger worry. A quarter of AXA’s AI users said they had received health information that later proved incorrect or misleading.
A University of Oxford study, published in Nature Medicine, tested nearly 1,300 participants on common health scenarios. Some used chatbots, while others took traditional routes such as consulting a GP.
The chatbots delivered a mix of helpful and harmful information, and users struggled to separate the two. Even correct details often arrived alongside misleading suggestions, creating confusion rather than clarity.
Dr Rebecca Payne, a GP and author, said AI is not ready to replace trained professionals in clinical decisions. Chatbots perform well in standardised knowledge tests, but real people describe symptoms messily.
A chatbot cannot examine you, hear your breathing or notice that you look pale. It also rarely asks the follow-up questions a clinician would, such as how long the pain has lasted or whether it spreads.
Confidence is part of the danger. A fluent answer sounds authoritative even when it is wrong, and non-experts cannot always tell the difference.
Regulation has not kept pace with use. AXA’s earlier research found that 78% of UK adults want stronger regulation of digital health information.
Worse for Brit-Asians
These risks land differently in British Asian communities. Reports note that access barriers, language differences and cultural stigma already shape South Asian experiences of healthcare, which makes a confident but flawed answer more consequential.
Stigma is the clearest example. Reproductive health, mental health and family medical history are often avoided at home, and many British Asians feel ashamed discussing reproductive health.
A private chatbot feels safer than a conversation with a GP who may know your family.
That privacy has value, but it can also let people avoid the appointment they need.
Patients who think in Punjabi, Urdu, Gujarati or Bengali may describe symptoms in English words that do not quite fit, or translate idioms literally.
The Oxford researchers found that users often leave out relevant details, and translation gaps can widen that problem. A GP can ask again, rephrase or book an interpreter, while a chatbot answers only what it was given.
There is a fair access argument for AI. An older relative with limited English might use a chatbot to understand a hospital letter in their first language, provided someone fluent checks the result.
In many families, younger relatives already translate for parents at appointments. If they also screen symptoms through a chatbot first, an unreliable answer can quietly delay care.
South Asian adults face a higher risk of type 2 diabetes and heart disease, and early signs of both can be subtle enough that an examination and blood test matter more than a chatbot’s best guess.
Where AI Fits
AI is not the enemy of good care. It can help people prepare questions, decode test results and describe symptoms clearly before they walk in.
Trust in doctors remains higher regardless.
EY’s global survey found 89% of respondents call their doctor reliable, compared with 68% for AI tools and 49% for social media.
People also want a human involved. In AXA’s study, 74% of AI users said they wanted access to a health professional who could validate or explain AI results.
That points to the right model: AI as preparation before an appointment, and the clinician as the decision-maker after it.
Community organisations can help close the gap. Health charities, pharmacists and local groups could run sessions on using AI safely, in the languages people actually speak at home.
Technology companies share responsibility. Health tools should be tested with varied accents, languages and ways of describing symptoms, rather than assuming every user writes like a medical textbook.
How to Use It Safely
Use AI to understand, not to decide. Ask what a term means, what questions to raise with a doctor, or how to describe symptoms in a way a clinician will follow.
Never use it to rule out an emergency. Chest pain, difficulty breathing, sudden weakness, a severe headache or heavy bleeding need urgent help through 999 or NHS 111, not a chat window.
Check anything important with NHS.uk, a pharmacist or your GP. If a chatbot suggests a medication, test or dose, treat it as a question to ask, not an instruction to follow.
Ask for support in your first language. Practices can arrange interpreters, and pharmacists can give quick advice without an appointment.
Trust ChatGPT to explain, organise and prepare. Do not trust it to diagnose, reassure or prescribe.
The 52% willing to act on AI-generated information are not being reckless; many are filling gaps in access and understanding. Closing those gaps, with shorter waits, better interpretation, and less stigma, is the real fix.
Until then, AI can be the notebook you bring to the consultation. It should never replace the person sitting across the desk.
