People describe their symptoms of illness to artificial intelligence in much less detail than to a human doctor. This is the central finding of a new study by Julius-Maximilians-Universität Würzburg (JMU) and other partners. The limited quality of information could jeopardize the reliability of digital initial assessments and patient safety.

In the experiment, 500 participants described simulated symptoms of headaches and flu. They believed their reports were being read by either an AI chatbot or a doctor. The evaluation showed that the descriptions were significantly shorter and less meaningful compared to the AI. While the texts to human professionals comprised an average of 255.6 characters, the AI only had 228.7 characters.

This loss of quality also occurs in individuals who have actually suffered from the corresponding complaints. The researchers attribute the phenomenon to psychological barriers. Many people assume that AI cannot adequately capture their individual situation and only compares standardized patterns. In addition, there is skepticism about the diagnostic ability of algorithms and data protection concerns.

The study, which has been published in the journal Nature Health , thus shows a weakness in the increasing use of AI systems in initial medical assessment and patient management that has received little attention so far. Even high-performance models can only perform as well as the information provided allows.

The scientists therefore recommend that the user interfaces of AI chatbots be improved in a targeted manner. Concrete examples of good symptom reports and active enquiries about missing details could significantly increase the quality of the information. This is the only way to prevent digital systems from making incorrect or inadequate recommendations.

The results are particularly relevant for the debate on digital initial counselling and self-triage in healthcare. The authors emphasize that technical developments alone are not enough if there is no human willingness to communicate with the machine in detail.

Original Paper:

Reduced Symptom Reporting Quality During Human-Chatbot Versus Human-Physician Interactions, Moritz Reis, Florian Reis, Yeun Joon Kim, Aylin Demir, Jess Lim, Matthias I. Gröschel, Sebastian D. Boie, Wilfried Kunde. Nature Health, DOI: 10.1038/s44360-026-00116-y

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability ected.