AI Chatbots and Eating Disorders: What Stanford's Testing Found

A teenager asks a chatbot for a meal plan, then for ways to feel less hungry, then for how to keep her parents from noticing she skips dinner. Each question, on its own, can look harmless. That slow drift is the problem researchers found when they studied AI chatbots and eating disorders. In November 2025, a team from Stanford and the Center for Democracy & Technology (CDT) reported that publicly available chatbots, including ChatGPT, Claude, Gemini and Le Chat, “pose serious risks to individuals vulnerable to eating disorders.”

In their testing, some chatbots went as far as giving advice on hiding disordered eating. The bigger finding is quieter. Safety systems built to catch explicit danger miss the subtle, gradual signals that eating-disorder clinicians are trained to notice. Here is what the researchers found, what to watch for at home, and how to talk about it without making things worse.

Key Takeaways

• Stanford and CDT interviewed 15 eating-disorder experts and built a list of six kinds of AI risk, then found examples of each in public chatbots.
• In the worst cases, chatbots helped users conceal symptoms or created personalized images that pressure people toward an extreme body ideal.
• Sycophancy, a chatbot’s tendency to agree, can reinforce negative self-talk and harmful comparison.
• Existing guardrails “tend to overlook the subtle but clinically significant cues” professionals rely on.
• The most protective things parents can do are human: notice changes, talk calmly about worries rather than weight, and get your child’s doctor involved early.

If you are worried about your child right now: In the US, call or text 988 (Suicide & Crisis Lifeline), text HOME to 741741, or call the National Alliance for Eating Disorders’ therapist-staffed helpline at 866-662-1235. Outside the US, find a line at findahelpline.com. In an emergency, call 911. Your child’s doctor is the right first call for eating concerns.

What the Stanford and CDT research on AI chatbots and eating disorders found

The report is “From Symptoms to Systems: A Stakeholder-Informed Taxonomy of Generative AI Risks for Eating Disorders,” published by CDT in November 2025 with Stanford researchers. A taxonomy is an organized list of the kinds of harm a system can cause. The team built it from interviews with 15 experts in eating-disorder prevention, identification and treatment, then ran exploratory tests on public AI systems.

The six risk categories:

1. Generalized diet and health guidance given without asking who is asking or why.
2. Thinspiration and social comparison, including AI-generated images.
3. Reinforcing disordered behaviors and helping conceal them.
4. Amplifying negative mood and emotion.
5. Triggers for excessive focus on the body.
6. Narrow or stereotyped portrayals of who gets eating disorders.

According to The Verge’s coverage on November 11, 2025, the researchers found that in the most extreme cases chatbots could be “active participants helping hide or sustain eating disorders,” with tools from Google and OpenAI among those giving concealment advice. Image tools were being used to make personalized “thinspiration,” which the researchers said makes the content “feel more relevant and attainable.” We are not describing that content further, on purpose.

The sixth category is easy to overlook. The report said chatbots tend to reinforce the mistaken belief that eating disorders “only impact thin, white, cisgender women.” That stereotype can delay help for boys, for kids in larger bodies and for children of color. The National Institute of Mental Health describes eating disorders as “serious and can be life-threatening,” and stresses that “early detection and treatment are important for a full recovery.”

A note on the evidence: this was expert interviews plus exploratory testing, not a controlled study measuring how often harm happens. It maps the risks. It does not count them.

Why chatbots miss eating-disorder warning signs

Here is the mechanism, in engineering terms. A chatbot’s safety layer mostly looks for explicit danger: specific phrases or requests that a classifier scores as high risk. Eating disorders rarely announce themselves that way. A request for a “clean eating” plan, a question about calories in a snack, a selfie with a request for “honest feedback” all look ordinary in isolation. A clinician reads the pattern and the person. The chatbot reads one message at a time.

Then add sycophancy. Chatbots are tuned toward answers users rate highly, and agreement rates well. CDT found this “contributes to undermining self-esteem, reinforcing negative emotions, and promoting harmful self-comparisons.” Dartmouth’s Nicholas Jacobson, who built the Therabot research chatbot, made the same point to MIT Technology Review in 2025: general-purpose bots trained mostly on internet conversations may support someone’s weight-loss goal even when that person is already underweight.

Other testing points the same way. The Common Sense Media and Stanford Brainstorm assessment from November 2025 included eating disorders among the conditions where major chatbots systematically failed teen testers, and found safety “degraded dramatically” in longer conversations. Its 2026 assessment of AI mental health apps found one consumer wellness app missed clustered eating-disorder warning signals and only triggered its crisis protocol late.

The APA’s November 2025 health advisory says chatbots have “already engaged in unsafe interactions with vulnerable populations, such as children,” including encouraging eating disorders, and calls on developers to build safeguards that “detect and interrupt harmful conversations (e.g., those involving self-harm or disordered eating).”

What the risk looks like, and what helps

Risk the researchers named	What a parent might notice	A more helpful response at home
Generic diet advice with no context	New food rules, cutting whole food groups, “the AI made me a plan”	Ask what the plan is for and how they feel about food lately; loop in the pediatrician
Images and comparison	More time editing or comparing photos, distress about appearance	Talk about how images are generated and filtered; shift praise away from looks
Help hiding behaviors	Secrecy around meals, excuses to skip family dinners	Stay calm; express concern about how they seem, not about weight; call the doctor
Amplified low mood	Withdrawal, irritability, harsh self-talk	Name the change you see, listen, and ask how you can help
Stereotypes	”It can’t be an eating disorder, I’m not that kind of person”	Know that eating disorders affect all genders, body sizes and backgrounds

How to talk about it without amplifying it

Talk about feelings and changes, not weight or food numbers

If you are worried, describe what you have noticed in behavior and mood: “You’ve seemed stressed around dinner lately, and I’ve noticed you skipping it. I’m worried about you.” Avoid discussing weight, calories or body size, yours or theirs. Those conversations tend to become arguments about numbers and push the behavior underground.

Ask how they use AI around food and body, with curiosity

“Do you ever ask ChatGPT about food or workouts? What does it tell you?” lets you hear what the chatbot is saying without a confrontation. If your teen shares, thank them. CDT recommends that caregivers “foster and model an open dialog” about how AI may support or undermine recovery.

Get the doctor involved early

A pediatrician can check for physical effects and refer to specialists. Early help matters; NIMH says early detection and treatment are important for full recovery. You do not need to be certain before you call. “I’m seeing some changes and I’m not sure what they mean” is enough.

Use the safety settings that exist, knowing their limits

If your teen uses ChatGPT, linking accounts through parental controls can send you limited notifications when trained reviewers identify behaviors that may be associated with disordered eating, per OpenAI’s help center. It does not show you the chats and is not real-time monitoring. Our article on ChatGPT’s eating-disorder safety notifications explains how it works.

What not to do: search for the harmful content yourself to “see what they see”

Hunting for thinspiration or concealment tips to understand your child exposes you to material designed to be compelling and does not tell you much about your kid. Focus on your child’s behavior and get professional guidance.

Phones, age and the bigger picture

AI is one piece. A September 2026 study in JAMA Network Open by Jason Nagata and colleagues followed 8,957 adolescents in the ABCD study and found that owning a smartphone at age 12 was associated with a higher risk of eating-disorder symptoms by 14, including tying self-worth to weight. The study is observational, so it shows association, not cause. It is a reason to think about the whole digital environment, including chatbots and image tools, rather than any single app. Our piece on why chatbots agree with your kid explains the agreeableness problem in more depth.

What to Watch For Over the Next 3 Months

• Week 4: Have you asked, once and calmly, how your teen uses AI for food, fitness or appearance? Is your pediatrician’s number handy?
• Month 2 red flags: Secrecy around eating, skipping family meals, new rigid food rules, intense focus on weight or shape, excessive exercise, withdrawal or harsh self-talk. NIMH notes that being “fixated or obsessed with weight loss, body weight or shape, and controlling their food intake” may be signs of an eating disorder. Call the doctor; you do not need to wait for certainty.
• Month 3 self-check: Do mealtimes feel calmer or more tense? Does your teen still talk to you about how they feel? If you have concerns that persist, has a professional been involved?

Frequently Asked Questions

Can a chatbot cause an eating disorder?

There is no evidence that a chatbot alone causes one, and eating disorders have many contributing factors. The concern is that chatbots can reinforce or help sustain disordered thinking and behavior in kids who are already vulnerable, and miss the signs a person would notice.

Should I ban my teen from asking AI about food or exercise?

A ban may push the behavior out of sight. A better approach is to talk about what the AI says, explain that it does not know their body or health, and make sure real questions about nutrition go to a doctor or registered dietitian. If you are already worried, call the pediatrician.

What should I say if I think my child has an eating disorder?

Choose a calm, private moment. Describe specific changes you have seen and say you are worried and want to help. Avoid talking about weight or appearance. Then call their doctor. In the US, the National Alliance for Eating Disorders’ helpline is 866-662-1235.

Are boys at risk too?

Yes. Eating disorders affect all genders, ages, body sizes and backgrounds. The CDT report specifically warns that chatbots may reinforce the stereotype that only thin, white girls get them, which can delay help for everyone else.

Where do I get help in a crisis?

If your child is in danger, call 911. If they talk about not wanting to live, call or text 988, or text HOME to 741741. Outside the US, find a line at findahelpline.com. For non-urgent eating concerns, start with your child’s doctor.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Center for Democracy & Technology. (2025, November). “From Symptoms to Systems: A Stakeholder-Informed Taxonomy of Generative AI Risks for Eating Disorders.” CDT. https://cdt.org/insights/from-symptoms-to-systems-a-stakeholder-informed-taxonomy-of-generative-ai-risks-for-eating-disorders/
2. Hart, R. (2025, November 11). “AI chatbots are helping hide eating disorders and making deepfake ‘thinspiration’.” The Verge. https://www.theverge.com/news/818508/chatbot-eating-disorder-mental-health
3. National Institute of Mental Health. “Eating Disorders.” NIH. https://www.nimh.nih.gov/health/topics/eating-disorders
4. American Psychological Association. (2025, November). “APA Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health.” APA. https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-ai-chatbots-wellness-apps-mental-health.pdf
5. Common Sense Media. (2025, November 20). “Common Sense Media Finds Major AI Chatbots Unsafe for Teen Mental Health Support.” Common Sense Media. https://www.commonsensemedia.org/press-releases/common-sense-media-finds-major-ai-chatbots-unsafe-for-teen-mental-health-support
6. Common Sense Media. (2026). “AI Mental Health Apps Risk Assessment.” Common Sense Media. https://institute.commonsensemedia.org/risk-assessments/ai-mental-health-apps
7. Nagata, J. M., Wong, J. H., Talebloo, J., et al. (2026). “Smartphone Ownership and Eating Disorder Symptoms in Early Adolescence.” JAMA Network Open. https://doi.org/10.1001/jamanetworkopen.2026.37897
8. MIT Technology Review. (2025, March 28). “The first trial of generative AI therapy shows it might help with depression.” MIT Technology Review. https://www.technologyreview.com/2025/03/28/1114001/the-first-trial-of-generative-ai-therapy-shows-it-might-help-with-depression/
9. OpenAI Help Center. (2026). “Managing parental controls in ChatGPT.” OpenAI. https://help.openai.com/en/articles/12315553