When Julie Bernstein\u2019s doctor recently recommended that she start eating chicken and steak to get more protein, she felt a bit irritated. Bernstein, 76, had been vegan for decades, and she was not about to start eating meat again, she said. Her physician, lacking ideas on how to guide her toward vegan protein sources, suggested she turn to ChatGPT.<\/p>\n
A few weeks later, Bernstein typed her goals into ChatGPT on her iPad and marveled at how quickly it churned out protein-forward menu ideas, grocery lists and meal-prep instructions that incorporated lentils, quinoa, protein powder and chia seeds.<\/p>\n
\u201cIt was like a cookbook tailored for me,\u201d said Bernstein, who lives in Harbor Springs, Michigan. The responses seemed to \u201cget\u201d her, she said.<\/p>\n
Since ChatGPT was released in 2022, chatbots have become a popular source of health information. In a survey of more than 5,500 US adults published in April, 1 in 4 respondents said they had recently used chatbots for health guidance. Another survey, published in January, found that among 1,000 US adults, a third reported having used ChatGPT or another artificial intelligence-powered tool to create nutrition or weight loss plans.<\/p>\n
We recently asked New York Times readers to share their experiences with using chatbots for help with nutrition; more than 500 people responded. Their stories were overwhelmingly positive, with many delighting in the bots\u2019 abilities to provide quick and helpful advice.<\/p>\n
Nutrition experts say that while chatbots can be useful for straightforward tasks like brainstorming meal ideas, they can also sometimes lead people astray.<\/p>\n
Here are some of the stories readers shared, with experts\u2019 thoughts on the promises \u2013 and potential pitfalls \u2013 of relying on chatbots for nutrition advice.<\/p>\n
The pros: Meal planning, nutrient tracking and accountability <\/p>\n
After Vanessa Crain, 47, was diagnosed with heart disease in 2025, she vowed to get serious about her diet. On her cardiologist\u2019s recommendation, she started trying to follow the heart-healthy DASH diet, which emphasises fruits, vegetables and whole grains.<\/p>\n
For hours on weekends, she\u2019d pore over cookbooks and plan meals for the week \u2013 tasks that quickly became burdensome along with her work and family obligations. So Crain, a kindergarten teacher in Longwood, Florida, outsourced that planning to Claude.<\/p>\n
Within a week of following its suggestions, she \u201cfelt like a different person\u201d, she said, with more energy and fewer cravings for junk food. The bot even recommended healthier takeout options for when she didn\u2019t have time to cook. She lost 10 pounds (4.5kg) in two months.<\/p>\n
The mental load of figuring out what to eat is often the most difficult aspect of eating healthfully, said Dr Lisa Oldson, an obesity medicine physician at the Northwestern University Feinberg School of Medicine.<\/p>\n
In her own practice, Dr Oldson embraces AI. She counsels patients on how to use chatbots for healthy eating, and uses ChatGPT herself to estimate the nutrient content of meals \u2013 like protein and fiber levels \u2013 that her patients are logging.<\/p>\n
Andrew Weir, 50, a kindergarten teacher in Arvada, Colorado, uses ChatGPT to track how much sodium and potassium he\u2019s consuming for blood pressure control. The bot has helped him make changes \u2013 like eating fewer Doritos, deli sandwiches and chicken nuggets, and more broccoli, spinach and bananas \u2013 that have helped lower his blood pressure, he said.<\/p>\n
He also asks ChatGPT to remind him every Friday that there may be doughnuts in the break room and that he might not want to overindulge. It has helped him eat fewer sweets, he said.<\/p>\n
Using chatbots as accountability partners can be a \u201cbeautiful use\u201d of the technology, said Wesley McWhorter, a dietitian in Houston. But pay attention if the reminders start feeling too rigid or cause anxiety, he added. That may do more harm than good.<\/p>\n
The pitfalls: Inaccurate, sometimes harmful advice <\/p>\n
Chatbots are convenient and fast, but sometimes they give \u201cincredibly bad advice\u201d, said Nick Tiller, who researches health misinformation at the Lundquist Institute in Torrance, California.<\/p>\n
Alex Rawdin, 39, a financial adviser in Philadelphia, recently asked the chatbot Tomo to help him lose weight. After several weeks of following the high-fat, low-carbohydrate keto diet, as it recommended, he remembered that his doctor had told him to limit animal protein because of his kidney stones.<\/p>\n
When Rawdin relayed that information to the bot, it advised that he stop the diet. He wished the chatbot had asked about his medical history before recommending it, he said. \u201cI found myself kind of ridiculing\u201d the bot, he added. \u201cThen I realised I was talking to a machine.\u201d (Tomo did not respond to a request for comment.)<\/p>\n
Chatbots aren\u2019t trained in assessing your health history or using clinical judgment the way a professional is, said Dawn Clifford, a dietitian and professor of health sciences at Northern Arizona University.<\/p>\n
In a study published in March, researchers from Turkey asked five popular chatbots to create three-day meal plans to help a few fictional overweight or obese 15-year-olds lose weight. The bots produced menus that contained, on average, about 700 fewer daily calories than a dietitian on the team did. If this guidance were followed long-term, the authors wrote, it could put adolescents at risk for malnutrition and eating disorders.<\/p>\n
Shelley Wood, a gastroenterology dietitian in San Jose, California, often sees patients who are following unnecessarily restrictive diets based on chatbot suggestions. She described one patient who \u2013 after many conversations with ChatGPT \u2013 had ended up eating only papaya, spinach, chicken and eggs for months to help relieve issues like bloating and stomach pain. The patient lost a concerning amount of weight, Wood said, and had fatigue and hair loss.<\/p>\n
Following a chatbot\u2019s nutrition advice could be \u201cbenign\u201d \u2013 or it could be \u201ccatastrophic,\u201d Tiller said. In a study published in April, he and his colleagues asked five chatbots various health-related questions. Among the 50 nutrition-related responses they received, 72% were rated as being ineffective or harmful if followed, he said.<\/p>\n
Because AI models are evolving rapidly, chatbots are likely giving more appropriate responses today, Tiller said. (Their study was conducted in February 2025.) Still, he added, it\u2019s doubtful that all of their problems have been resolved.<\/p>\n
Chatbots are only as good as the data they\u2019re trained on, which include a range of sources like books, websites, scientific studies, social media posts and Reddit threads, some of which can be contradictory and inaccurate, Tiller said.<\/p>\n
\u201cIt only takes a tiny amount of misinformation\u201d to confuse a chatbot, he added. Yet they \u201calways respond confidently and authoritatively,\u201d even when those responses are dubious, Tiller said. They will rarely acknowledge when they don\u2019t have a good answer, he added, making it difficult to know when the advice is bad.<\/p>\n
Drew Pusateri, a spokesperson for OpenAI (which makes ChatGPT), said via email that the bots are not designed to be a substitute for medical care.<\/p>\n
Using chatbots with caution <\/p>\n
For all of their potential pitfalls, many readers felt appreciative of the guidance that chatbots provided. Some said they offered something their doctors or other health professionals couldn\u2019t: support in seconds, at any hour, no appointment or copay required.<\/p>\n
There is no doubt that chatbots can be helpful, McWhorter said, but if you find yourself turning to them for tasks like diagnosing or managing a medical condition, or for guidance on supplements or major dietary changes \u2013 it\u2019s best to run the responses by a doctor.<\/p>\n
If possible, he said, schedule a visit with a dietitian. They have far more nutrition training than most physicians do and offer longer appointment times.<\/p>\n
It\u2019s also worth asking chatbots to supply references and to check that they\u2019re from reliable places, like academic institutions, peer-reviewed journals or health organisations such as the American Heart Association or the Academy of Nutrition and Dietetics, Clifford said.<\/p>\n
And try the same prompt on multiple chatbots, she added. Different answers are a clue that at least one of them may be inaccurate.<\/p>\n
The bottom line is that you should use chatbots with caution, Tiller said. They\u2019re not \u201call-knowing oracles\u201d.\u00a0\u2013\u00a0\u00a92026 The New York Times Company<\/p>\n