Nearly one in two patients now seeks medical advice from an AI. But while professionals get spot-on medical results with precise queries, for many laypeople the journey leads into the unknown. In the end, it’s often not the computational power of the code that matters, but the quality of our questions.
Although generative AI models like ChatGPT possess an enormous amount of medical knowledge, laypeople hardly benefit from it in their daily lives. According to a study published in the journal *Nature Medicine*, ordinary users receive the correct preliminary diagnosis in only 35% of cases. By comparison, experts achieve a 95% accuracy rate when using precise input.
According to experts, the problem lies less in the AI’s computing power and more in the interaction: Laypeople often omit important information in their prompts (search queries) or unconsciously steer the model in a direction that leads to an incorrect diagnosis. In addition, users tend to anthropomorphize the AI, which reduces their critical distance from the results.
Despite this error rate, demand remains extremely high, as chatbots are available around the clock and allow for anonymous counseling on sensitive topics. Experts are therefore urgently calling for greater AI literacy and the use of specialized bots to prevent dangerous misdiagnoses.
The Trend Toward Digital Second Opinions
The reluctance to have medical symptoms assessed by AI is rapidly declining. An estimated half of those with public health insurance in Germany have already used ChatGPT or similar models to ask health-related questions (“Digital Health 2025” by Bitkom Research). The main reasons for this are the constant availability (24/7) and the difficulty in securing appointments with primary care physicians, especially in rural areas. AI models such as ChatGPT, Claude, or Gemini also offer anonymity, which is seen as an advantage, particularly when it comes to sensitive topics such as mental health, addiction, or sexual health.
Data Set Factor
A key risk associated with general-purpose chatbots is their data source. Because they draw on the entire Internet, reliable medical knowledge is mixed with inaccurate information. This can lead to dangerous self-diagnoses that delay necessary visits to the doctor.
Experts see the solution in specialized systems based on verified expert data. Examples such as “Uro-Bert” (urology) or “Lupus-GPT” demonstrate how AI can serve as a precise interface for expert knowledge. However, data protection remains an unresolved issue: To receive an accurate, personalized response, users must disclose highly sensitive data. This presents a dilemma between medical accuracy and the protection of privacy.
The Role of the Pharmaceutical Industry
For the pharmaceutical industry, integrating AI into patient counseling can open up new possibilities along the patient journey. Specialized bots could be used to explain complex package inserts in an understandable way or to support the management of side effects in chronic conditions. AI thus serves as a bridge between scientific evidence—which is often difficult to understand—and patients’ need for information. Nevertheless, the issue of liability remains central: As long as AI models are not permitted to make binding diagnoses, their use serves primarily to provide information and promote adherence, but not to replace medical expertise.
