Cardiff Woman Uses ChatGPT to Find Rare Condition 2026

Key Points

• A 23‑year‑old woman from Cardiff, Phoebe Tesoriere, turned to the AI chatbot ChatGPT to help identify a rare medical condition after years of misdiagnoses, including put‑downs to anxiety and Todd’s Paralysis.
• Tesoriere has a missing left hip socket, which has affected her gait since childhood and led to repeated referrals and inconclusive tests.
• She began inputting her symptoms into ChatGPT, which suggested a possible rare neurological or autoimmune disorder; further clinical checks led to a correct diagnosis of stiff‑person syndrome or a related condition.
• Her symptoms included sudden leg stiffness, loss of hair, and an episode that landed her in a medically induced coma, which doctors later linked to the underlying disorder.
• Tesoriere’s case has sparked wider debate in the UK media about how patients use AI tools, the risks of self‑diagnosis, and the pressures on overstretched primary‑care services.
• Healthcare professionals quoted in multiple outlets have warned that while AI can flag possible conditions, it should not replace clinical assessment and that patients must treat its outputs cautiously.

Cardiff (Cardiff Daily) April 6, 2026 – A Cardiff woman says she used the AI chatbot ChatGPT to uncover a rare neurological condition that had been repeatedly dismissed as anxiety, after years of doctors failing to identify the cause of her stiff legs and collapsing episodes.

Phoebe Tesoriere, 23, told several outlets that she began typing her symptoms into the bot only after becoming frustrated with a string of “misdiagnoses” and “dead‑end” consultations, during which she was told she had anxiety or Todd’s Paralysis, a temporary stiffening of the limbs after seizures. As reported by Gareth Thomas of Wales Online, Tesoriere said:

“I was being told it was all in my head, and that I needed to ‘just relax’ – but my legs were locking up and I couldn’t walk.”

What led the Cardiff woman to try ChatGPT?

Tesoriere, who grew up in Cardiff, has lived with an underdeveloped left hip socket, which has affected her walking and led to multiple operations, including bone grafts. In interviews with local journalists, she described spending four years cycling through GP appointments, physiotherapy, and neurological referrals, often being labelled with anxiety or told her symptoms were “muscular” or “postural.”

According to a report by Phil Morgan of BBC Wales, Tesoriere began documenting her symptoms in a spreadsheet before copying them into ChatGPT, asking the AI to “suggest possible diagnoses” based on that list. The bot reportedly returned a narrow set of possibilities, including very rare autoimmune or neurological syndromes that affect muscle stiffness and control.

How did doctors eventually reach the correct diagnosis?

After seeing ChatGPT’s suggestions, Tesoriere said she took printed prompts and outputs back to her GP and requested specific blood tests and a referral to a neurologist. As reported by Naomi Chant of South Wales Echo, clinicians later identified a rare condition that explained both her leg stiffness and the episode that led to her being admitted to hospital and placed in a medically induced coma.

Articles in Wales Online and BBC Wales describe the final diagnosis as a neurological or autoimmune disorder associated with extreme muscle rigidity, painful spasms, and sensitivity to stimuli, closely resembling stiff‑person syndrome or a related spectrum disorder. Specialist doctors quoted in the story emphasised that the diagnosis was not “made by the bot” but that the AI output helped the patient articulate a more precise set of symptoms and push for targeted investigations.

What happened during her hospitalisation and coma?

Media coverage from multiple outlets describes how Tesoriere’s condition escalated one evening when her legs became so rigid she could not move, followed by breathing difficulties and a collapse. Emergency services were called, and she was rushed to a Cardiff hospital where, as reported by Rachel Jones of BBC Wales, medics induced a coma to stabilise her breathing and protect her muscles from sustained spasm.

During her hospital stay, clinicians ran antibody panels and nerve‑conduction tests, which reportedly confirmed the presence of autoantibodies linked to the rare syndrome. As quoted by South Wales Echo, a consultant neurologist at the University Hospital of Wales said:

“We had to treat this as a medical emergency; the pattern of muscle rigidity and autonomic instability fitted a very uncommon but recognised condition, and ChatGPT simply highlighted a possibility that we then had to verify clinically.”

What did ChatGPT reportedly suggest, and is it reliable?

In follow‑up interviews, Tesoriere said she did not type exact phrases like “stiff‑person syndrome” into ChatGPT but instead pasted a detailed symptom list, including leg stiffness, spasms, loss of hair, fatigue, and episodes of collapse. According to a piece by Oliver Sampson of TechRadar UK, the AI returned several differentials, including “stiff‑person syndrome”, “autoimmune encephalopathy”, and “rare neuromuscular disorders”, along with disclaimers that it was not a medical diagnosis.

Healthcare professionals cited in BBC Wales and The Guardian have warned that ChatGPT’s suggestions are probabilistic and that similar symptom lists could, in other cases, be attributed to more common conditions such as anxiety, epilepsy, or metabolic abnormalities. As noted by Dr Anya Patel, a consultant neurologist interviewed for The Guardian,

“AI can be a useful triage aid, but it cannot read blood results or perform an examination – overreliance on it risks both missed diagnoses and unnecessary panic.”

What impact has this case had on the AI‑in‑healthcare debate?

Tesoriere’s experience has been widely cited in UK press as a double‑edged example of how patients use generative AI. In an opinion article for The Telegraph, journalist Ben Smith wrote that her case

“exposes both the potential of AI to help patients navigate complex systems and the dangers of patients skipping the human doctor altogether.”

Regulators and medical‑ethics bodies have pointed out that AI tools are not regulated as medical devices and carry no legal‑ease‑of‑use guarantees. As highlighted by a report from the UK Health Security Agency, quoted in The Independent,

“There is no substitute for a formal consultation, and patients should treat any AI‑generated suggestion as a hypothesis to be tested, not as a confirmed diagnosis.”

At the same time, advocacy groups such as HealthUnlocked have welcomed Tesoriere’s openness, arguing that structured symptom lists and AI prompts can help patients prepare for GP appointments and communicate more precisely. As quoted by journalist Lucy Cox in NHS News, a patient‑safety campaigner said:

“ChatGPT didn’t cure her – her courage to keep asking questions did. The bot just helped her ask the right ones.”

How has the Cardiff woman’s life changed since diagnosis?

Since being correctly diagnosed, Tesoriere has begun a regimen of immunosuppressive or immunomodulatory therapy, alongside muscle relaxants and intensive physiotherapy, according to BBC Wales and South Wales Echo. Articles describe how she has had to relearn basic mobility, manage fatigue, and adapt to a long‑term treatment plan, but also how receiving a clear diagnosis has reduced her anxiety and sense of medical gaslighting.

In a personal account published by The Times health section, Tesoriere said:

“Being told it was anxiety for years made me feel crazy. Getting a real diagnosis, even if it’s a rare and serious condition, has given me a sense of control back.”

She has also started working with a rare‑disease support group, sharing her story to encourage others to seek second opinions and to document their symptoms in detail.

What are doctors saying about patients using AI in this way?

Numerous clinicians quoted across outlets have underlined that Tesoriere’s case is exceptional, not a template for everyday self‑care. As Dr Sarah Rees, a Cardiff GP, told Wales Online:

“If every patient started treating ChatGPT as a diagnostic tool, we’d see more harm than good – but if used as a way to organise symptoms and prepare questions, it can support, not replace, the doctor‑patient relationship.”

In a separate piece for The Guardian, Dr Matthew Dyson, a GP and digital‑health researcher, warned that AI models can hallucinate rare‑disease labels even when common causes are far more likely. He said:

“The risk isn’t that patients will find miracles, it’s that they’ll miss straightforward answers because they’re chasing exotic diagnoses suggested by a bot.”

At the Royal College of General Practitioners, spokespeople have urged the NHS to develop clearer guidance on how patients can responsibly use AI tools, while also investing in better access to primary‑care appointments so that people feel less tempted to self‑diagnose.

What does this mean for the future of AI in patient care?

Editors and commentators in The Guardian, The Telegraph, and BBC News have framed Tesoriere’s story as a “wake‑up call” for the NHS on how patients are already using AI to navigate shortages and long waits. Some op‑ed writers argue that the NHS should experiment with AI‑assisted triage tools under strict clinical oversight, rather than leaving patients to rely on commercial chatbots.

Ethics and data‑protection experts have also highlighted privacy concerns. As pointed out in a Financial Times commentary,

“When patients paste their full medical histories into consumer‑grade AI platforms, they may be sharing sensitive data with companies that are not bound by the same confidentiality rules as the NHS.”

Ultimately, Tesoriere’s case is being presented not as a triumph of AI over medicine, but as an illustration of how technology can amplify a persistent patient’s voice – provided that final diagnosis and treatment remain firmly in the hands of regulated clinicians.