AI Chatbot Flags Rare Condition After Years of Misdiagnosis

A 23-year-old woman in Cardiff, Wales, received a confirmed diagnosis of a rare genetic disorder in April 2026 after an artificial intelligence chatbot, ChatGPT, suggested a condition that had not been identified through years of medical consultations. The case centers on Phoebe Tesoriere, whose long-running symptoms included seizures, falls, and balance difficulties that began in childhood.

 

Tesoriere’s health problems led to repeated contact with medical services and multiple explanations for her condition over time. Her symptoms were often attributed to anxiety or epilepsy, and she was diagnosed with epilepsy in 2022. She was later diagnosed with Todd's paralysis, but her condition continued to deteriorate, according to the account.

 

In January 2025, Tesoriere was hospitalized for three months after a fall. The progression of her symptoms did not stop, and she later experienced a seizure that resulted in a three-day coma. With no definitive diagnosis in place, she turned to ChatGPT for possible explanations, seeking a new direction after years of uncertainty.

 

ChatGPT identified hereditary spastic paraplegia as a potential cause. The condition is described as involving progressive muscle stiffness and weakness in the legs. Following the chatbot’s suggestion, genetic testing was pursued and subsequently confirmed hereditary spastic paraplegia as the correct diagnosis.

 

Tesoriere is now managing the condition with physical therapy and uses a wheelchair. She is also pursuing a master’s degree in psychology. The account highlights how an AI tool, used outside a clinical setting, helped point toward a rare diagnosis that had not been reached through earlier assessments.

 

While the case underscores potential value for AI in complex diagnostic challenges, it also leaves key uncertainties unresolved in the public description. The details do not specify which clinicians or institutions were involved in earlier assessments, nor how the genetic testing pathway was initiated and coordinated after the AI suggestion. It also does not quantify how common hereditary spastic paraplegia is or how frequently similar diagnostic delays occur, limiting broader comparisons.