TL;DR

The Eve Appeal, a UK charity for gynaecological cancers, found that when ChatGPT was asked about bloating that would not go away, it offered tips for living with irritable bowel syndrome and did not raise ovarian cancer. It also missed vulval cancer when asked about vulval itching. On 27 October the charity launched Ask Eve Assist, its own AI information tool, which answers only from certified sources and passes harder questions to specialist nurses.

What the charity found

In the vulval case, the charity says the chatbot missed the cancer link even though it had been told how the symptom can appear in cancer. Chief executive Athena Lamnisos said the charity also found AI tools claiming, wrongly, that a PAP test can pick up cancer of the vagina, an error that “could leave someone falsely reassured and delay them seeking help”.

“The mistakes made by online AI platforms could be fatal,” she said, pointing to the roughly 600,000 people she said are waiting for gynaecological care in England.

OpenAI told Digital Health that it works with clinicians to improve its models, that ChatGPT is built to ask follow-up questions and point people to professional help, and that it “should be used for information and education, not to replace professional medical advice”.

How Ask Eve Assist works

The tool answers from sources certified under the PIF TICK quality mark, on topics from symptoms and screening to HPV and prevention. Anyone who needs a medical assessment is sent to health services, and users can ask to speak to one of the charity’s nurses. Patients, nurses and NHS partners helped design it, working with its digital partner, Digital Wonderlab.

Demand explains why. Enquiries to the nurse-led Ask Eve service rose by 1,141% after Jo’s Cervical Cancer Trust closed in 2024, and the service has one full-time and one part-time nurse. In a pilot with 359 users, the tool handled 340 without a nurse and passed 19 to the specialists.

Looking Forward

In our view, this is a different answer to AI health misinformation from simply asking general-purpose chatbots to improve. Narrow tools that answer only from vetted sources and hand off at clear points may appeal to other UK health charities and NHS services facing similar demand. The test is whether people actually choose them over the general chatbot already on their phone. The pilot figures suggest the hand-off works; they do not yet show how accurate the answers were.