People are using AI tools to self-diagnose, but research shows they are very likely to be getting bad advice

New polling from AXA Health, claims that artificial intelligence is directly shaping when and how people seek medical help in the UK despite a growing body of evidence showing that AI models are not very good at it. One new study, titled “Large Language Model Performance and Clinical Reasoning Tasks”, published in JAMA Network Open, found that AI chatbots misdiagnosed medical conditions in over 80 per cent of early clinical cases.

The findings from AXA health suggest that large language models (LLMs) are leading many to delay essential care (59 percent) while pushing others to seek unnecessary appointments (59 percent). According to AXA Health, the dual dynamic of reassurance and alarm shows that LLMs is not simply changing how people understand their symptoms – it is transforming routes to care.

The research, commissioned by AXA Health and conducted by Censuswide amongst an equal split of 2,000 AI users and non users, found a dual dynamic of reassurance and alarm. While AI is helping people understand their symptoms more clearly, it is also contributing to rising health anxiety and inaccurate diagnoses. 78 percent of AI users say it has helped them understand medical terminology, test results or treatment plans, yet 37 percent say the tech made them more anxious after checking symptoms.

Key findings include:

• 59 percent of people who check symptoms with LLMs have delayed speaking to a healthcare professional because the tool reassured them
• The same percentage (59 percent) say LLMs made them more worried, leading them to seek help they later discovered they didn’t need
• Over a third (36 percent) turn to LLMs first (e.g ChatGPT, AI symptom checker), nearly double those who visit the NHS website in the first instance (19 percent)
• 93 percent of AI users have used it to check symptoms late at night, fuelling health anxiety (37 percent feel more anxious)
• A quarter (25 percent) have received health information from LLMs that later turned out to be incorrect or misleading
• However, over two-thirds (68 percent) say AI has made them feel more confident discussing symptoms at appointments
• And over three-quarters (78 percent) of AI users say it has helped them understand medical terminology, test results or treatment plans

The data reveals concerning statistics. 93 percent of AI users say they have used it to check symptoms late at night, with many describing habits that spiral into more checking. 59 percent say one question ‘always’ or ‘often’ leads them to ask more, forming what AXA Health has termed the ‘AI Health Anxiety Loop’. In addition, a quarter (25 percent) of users say they have received health information that later turned out to be incorrect or misleading, highlighting risks around self-diagnosis.

AI also appears to be intensifying the urge to check symptoms against serious medical conditions. Over the past year, over a third (36 percent) checked symptoms related to mental health conditions, while 27 percent explored women’s health issues. More concerningly, 11 percent turned to LLMs to check symptoms linked to sepsis. This growing reliance may be fuelling anxiety, with over a third (35 percent) of users saying they felt more anxious as a result. Meanwhile, nearly two-thirds (63 percent) report that AI has made minor symptoms sound more serious at least some of the time.

Comparisons with non-AI users underline the scale of the shift. Users are more than twice as likely to delay seeking help after digital reassurance (59 percent vs 23 percent), and twice as likely to seek unnecessary care (59 percent vs 27 percent).

However, AI can be a positive source of support when used to empower patients in real-life health scenarios. Over two-thirds (68 percent) of users say AI has made them feel more confident discussing symptoms with a clinician, and three-quarters (78 percent) say AI has helped them understand medical terminology, test results or treatment plans.  The scale of AI adoption is striking, over a third (36 percent) turn to AI first, nearly double those who visit the NHS website (19 percent). And for many, the experience is a positive one: 78 percent say AI has helped them understand medical terminology, test results and treatment plans

AXA Health’s previous research showed that 48 percent of UK adults self-diagnosed online, with 30 percent turning to social media for health content and 78 percent calling for stronger regulation of digital health information. The rise of AI marks the next stage in this trend – shifting digital search from passive reading to conversational checking through live conversations with AI bots, and now directly influencing when people decide to seek care.