AI chatbots drive 7.6% of eating-disorder helpline calls

Eating-disorder therapists say patients increasingly use AI chatbots for diet advice, challenging professional guidance during treatment sessions.

Eating-disorder therapists report patients are turning to ChatGPT, Claude and Gemini for diet and exercise advice, with 7.6% of helpline calls now originating from AI platforms — up from less than 0.5% a year ago.

"I've had patients hold up their phone and use their mic to ask ChatGPT what was wrong with what I just said," Dr. Anne O'Melia, chief clinical and quality officer at the Eating Recovery Center, said. "Half my patients are fact-checking me daily."

Patients upload photos asking for weight-loss tips, share daily meal logs for healthier suggestions, and receive responses that include ultralow-calorie meal plans, carb restrictions and fasting recommendations — advice that can be dangerous for those with eating disorders, which carry the second-highest mortality rate of any psychiatric illness after opioid addiction.

OpenAI, Anthropic and Google are working to improve detection, with Anthropic's latest models scoring 99.7% and 99.8% on identifying disordered-eating prompts. But therapists warn that even accurate detection doesn't solve the core problem: chatbots remain unable to distinguish between a bodybuilder seeking protein advice and a malnourished person restricting intake, creating a product-safety challenge for AI companies embedding their tools in everyday health decisions.

Patients Fact-Check Therapists Mid-Session

Therapists say correcting chatbot misinformation consumes valuable treatment time. Hannah Lindsey, a therapist in Louisville, Ky., said she must pause trauma processing to address meal plans generated by chatbots. "When a patient says a chatbot told her to go to the gym five times a week and all she eats is oatmeal, I have to put the trauma processing to the side and address the meal plan that's not appropriate," Lindsey said.

The problem extends beyond the therapy room. Allie Weiser, director of helpline services for the National Alliance for Eating Disorders, said patients with eating disorders often believe extreme exercise or bingeing and purging help them cope with anxiety or depression. When chatbots unknowingly reinforce those behaviors with generic advice, recovery becomes more difficult.

AI Companies Race to Close the Detection Gap

OpenAI said it continues to use input from clinicians and real use cases to refine its eating-disorder policies and safeguards, pointing to independent research showing GPT-5.4 consistently recognized eating disorders and recommended evidence-based treatment. An Anthropic spokesman said its Claude models have classifiers designed to recognize signs of disordered eating, avoiding calorie counts and weight-loss guidance when risks are detected. Google said Gemini encourages users to consult medical professionals.

But training AI to know when and how to offer advice has proven difficult. In 2023, a chatbot on the website of the National Eating Disorders Association, Tessa, dispensed unapproved weight-loss tips after being programmed with generative AI without the nonprofit's knowledge. The organization removed the tool and has no plans to bring back a chatbot, Chief Executive Jessica Scheer said.

Johanna Kandel, CEO of the National Alliance for Eating Disorders, noted that upward of 70% of people with eating disorders never get access to care. Chatbots are pushing more people to call the helpline — 7.6% of calls last month originated from AI platforms — but the technology still cannot replace human clinical judgment. "AI platforms have an opportunity here to learn from what happened with social media and eating disorders, and to do better," Kandel said.

For AI companies, the stakes are both ethical and financial. OpenAI, valued at more than $80 billion in its latest funding round, and Anthropic, which has raised over $7 billion, are racing to embed their models in consumer health and wellness — a market where safety failures can trigger regulatory scrutiny and reputational damage. Google, facing similar risks with Gemini, is investing heavily in safety classifiers. The challenge: building models that can reliably distinguish between a healthy fitness query and a disordered-eating prompt, a distinction that even trained clinicians sometimes struggle to make without a full patient history.

This article is for informational purposes only and does not constitute investment advice.