The latest
Chatbots are moving into one of the most delicate areas of mental-health care: eating-disorder treatment. Therapists say patients are increasingly uploading photos, asking for weight-loss plans, muscle-building routines and meal advice, then arriving in therapy with a competing voice in hand. The issue is not simply that the advice is wrong. It is that it can sound persuasive enough to disrupt treatment.
What may be harmless guidance for a general fitness user can be dangerous for someone with an eating disorder. Advice on cutting calories, reducing fat or increasing exercise may reinforce the very patterns therapy is trying to undo. Clinicians say they are increasingly forced to pause deeper treatment work in order to correct chatbot guidance that is detached from diagnosis, trauma history and nutritional risk.
AI companies say they are refining their models to detect signs of disordered eating and to steer users toward professional help. Some referrals appear to be improving. But therapists say the core problem remains: a model still struggles to distinguish between an ordinary nutrition question and a clinically dangerous request from a vulnerable patient.
Details
- Therapy room: Some patients now use ChatGPT during sessions to question or fact-check what a therapist has just said.
- Clinical risk: Eating disorders carry one of the highest mortality rates among psychiatric illnesses, making generalized health advice especially sensitive.
- Type of advice: Chatbots may suggest severe calorie restriction, cutting fats and carbohydrates, or excessive workout schedules.
- Company response: OpenAI, Anthropic and Google say they have strengthened safeguards, detection systems and referrals to professionals.
- Mixed effect: AI may harm some users, but it is also sending more people toward human-staffed helplines and professional support.
Between the lines
This is not only a story about bad nutrition advice. It is about a new authority entering the therapy room. The more confident, personalized and always available the chatbot seems, the easier it becomes for it to weaken the therapeutic relationship, even when its advice is not entirely false.
What’s next
The next test is whether AI companies can build more precise safeguards around eating disorders, ones that do more than offer generic referrals and instead block guidance that could reinforce harmful behavior before it causes real damage.
