The growing use of artificial intelligence for personal health questions is raising a challenge that extends beyond the accuracy of its answers: whether those answers can be examined when something goes wrong.

A medical case involving a man who required hospital treatment after making a dietary change has brought this issue into focus. Seeking to reduce chloride in his diet, he replaced ordinary table salt with sodium bromide. He subsequently developed hallucinations and paranoia, leading to hospitalization.

Yet doctors faced a significant obstacle when trying to understand the circumstances behind the incident. The conversation in which the man had sought advice from a chatbot was no longer available, leaving medical professionals unable to establish precisely what guidance he had received.

The case was discussed in a research paper written by scholars from Binghamton University, Stanford University, Texas A&M University and Indiana University. The researchers argue that limited transparency around AI-generated health information can make it difficult to investigate potentially harmful advice after an incident has occurred.

The problem goes beyond the accuracy of answers

Kaicheng Yang, one of the paper's authors, told Newsweek that artificial intelligence can offer useful support when people seek health information, but its mistakes can be difficult to recognize because incorrect responses may be delivered with the same confidence as accurate ones.

He warned that relying on such information could have consequences, including postponing professional medical care or taking inappropriate medication.

The missing chatbot conversation in this case illustrates another dimension of the problem. Even when a medical complication is identified, the absence of a record of the original exchange can prevent doctors and researchers from determining whether the AI system contributed to the decision that preceded it.

This creates an accountability and research challenge: without access to the advice that was provided, it becomes harder to assess the system's role, understand how the misunderstanding developed or identify changes that might help prevent similar incidents.

Millions are turning to AI for health information

The case comes amid increasing reliance on artificial intelligence as a source of health guidance. A Gallup survey of more than 5,500 U.S. adults, conducted between October 27 and December 22, 2025, found that approximately 14 million adults reported skipping a healthcare visit after using AI.

The figure points to a wider issue in the relationship between digital tools and medical services. AI systems are becoming part of how some people seek answers about their health, while questions remain about the reliability of their responses and the extent to which those interactions can be reviewed.

The researchers argue that the limited visibility into AI-generated recommendations makes it harder for clinicians, researchers and regulators to identify potentially harmful guidance retrospectively.

OpenAI stresses that ChatGPT is not a substitute for medical care

Responding to Newsweek, an OpenAI spokesperson emphasized that ChatGPT is not a doctor and should not replace medical care, diagnosis or treatment.

The spokesperson added that the service's terms make clear it is not intended for medical diagnosis or treatment. However, because people continue to use it to obtain health information, the company said it takes the reliability and safety of those responses seriously.

The case therefore highlights two connected responsibilities. Users need to understand the limits of automated health information, while developers face the challenge of making AI responses safer and more transparent.

As AI becomes a more common source of health-related information, the ability to evaluate what systems recommend—and to examine those recommendations when problems arise—may become as important as improving the answers themselves.