They answer with confidence and authority, rarely acknowledging uncertainty.
In the age of algorithmic convenience, millions of Americans have begun turning to artificial intelligence not just for recipes, but for the kind of nutritional guidance once reserved for clinicians. The appeal is understandable — a patient listener, available at any hour, capable of tailoring a meal plan to a lifetime of preferences. Yet beneath this promise lies a quieter danger: a technology trained on the full spectrum of human knowledge, including its errors, answering with the confidence of expertise it does not possess.
- One in four American adults now consult chatbots for health advice, drawn by the speed and apparent personalization of tools that never tire or judge.
- Studies reveal that 72% of chatbot nutrition responses are ineffective or harmful — bots have generated meal plans dangerously low in calories and recommended diets incompatible with patients' existing medical conditions.
- Real people have paid real prices: restricted diets leading to malnutrition, hair loss, fatigue, worsened kidney stones, and eating disorders — all following advice delivered with algorithmic certainty.
- The fundamental flaw is structural: chatbots are trained on contradictory data including misinformation, and they lack the clinical judgment to recognize when a recommendation has crossed into harm.
- Experts are now drawing a careful line — chatbots may serve well as brainstorming partners and accountability tools, but any guidance touching medical history must be verified by a doctor or registered dietitian.
When Julie Bernstein's doctor ran out of ideas for helping his 76-year-old vegan patient get enough protein, he pointed her toward ChatGPT. Within weeks, she was marveling at meal plans built around lentils, quinoa, and chia seeds — complete with grocery lists. "It was like a cookbook tailored for me," she said.
Since ChatGPT's release in 2022, millions have followed a similar path. Surveys suggest one in four American adults have used chatbots for health advice, and when the New York Times asked readers about their experiences, more than 500 responded — most of them positive. A kindergarten teacher in Florida overhauled her diet after a heart disease diagnosis and lost 10 pounds in two months using Claude for meal planning. A Colorado teacher used ChatGPT to track sodium and potassium for blood pressure control. Some physicians now counsel patients on how to use these tools effectively.
But the convenience conceals a serious problem. A study published in April found that five popular chatbots, asked to create meal plans for overweight teenagers, generated menus averaging 700 fewer daily calories than a professional dietitian would recommend — a shortfall capable of triggering malnutrition. A Philadelphia financial adviser followed a chatbot-recommended keto diet for weeks before remembering his doctor had warned him against animal protein due to kidney stones. The bot had never asked.
Researcher Nick Tiller found that 72% of nutrition-related chatbot responses were ineffective or harmful if followed. A gastroenterology dietitian in San Jose regularly treats patients harmed by AI advice — including one who had restricted herself to four foods for months, developing fatigue and hair loss. The chatbot had no knowledge of her medical history and no capacity to recognize the danger.
The root issue is structural: chatbots are trained on vast, contradictory datasets that include misinformation, and they respond with uniform confidence regardless of accuracy. OpenAI has acknowledged its tools are not designed to replace medical care.
Experts now recommend a measured approach: use chatbots for brainstorming and accountability, but verify any meaningful guidance with a doctor or dietitian. Ask bots to cite sources, cross-check answers across multiple tools, and treat diverging responses as a warning sign. The technology is improving — but it remains a tool that handles easy questions well and hard ones dangerously.
Julie Bernstein's doctor had a problem. The 76-year-old from Harbor Springs, Michigan, had been vegan for decades, and when her physician suggested she eat chicken and steak for protein, she politely refused. Out of ideas, the doctor pointed her toward ChatGPT. Within weeks, Bernstein was marveling at how the chatbot seemed to understand her completely—generating protein-rich meal plans built around lentils, quinoa, and chia seeds, complete with grocery lists and cooking instructions. "It was like a cookbook tailored for me," she said.
Since ChatGPT's release in 2022, millions of people have turned to AI for nutrition guidance. A survey of more than 5,500 American adults found that one in four had recently used chatbots for health advice. Another study of 1,000 adults showed a third had used ChatGPT or similar tools to create nutrition or weight loss plans. When the New York Times asked readers about their experiences, more than 500 responded—and the overwhelming majority reported positive results. People loved the speed, the personalization, the sense of being understood by a machine that never got tired or impatient.
For many, the benefits were real. Vanessa Crain, a kindergarten teacher in Longwood, Florida, was diagnosed with heart disease in 2025 and needed to overhaul her diet. Following the DASH diet felt overwhelming alongside work and family, so she turned to Claude for meal planning. Within a week, she felt transformed—more energy, fewer cravings. She lost 10 pounds in two months. Andrew Weir, a teacher in Colorado, used ChatGPT to track his sodium and potassium intake for blood pressure control, and the bot even reminded him on Fridays about the doughnuts in the break room. His blood pressure improved. Dr. Lisa Oldson, an obesity medicine physician at Northwestern University, now counsels her own patients on how to use chatbots for meal planning and uses ChatGPT herself to estimate nutrient content. The technology, when used for straightforward tasks, can genuinely lighten the mental load of figuring out what to eat.
But the convenience masks a serious problem. In a study published in April, researchers asked five popular chatbots to create three-day meal plans for fictional overweight teenagers. The bots generated menus averaging 700 fewer daily calories than a professional dietitian would recommend—a shortfall that, if sustained, could trigger malnutrition and eating disorders. Alex Rawdin, a financial adviser in Philadelphia, followed a keto diet recommended by the chatbot Tomo for several weeks before remembering his doctor had warned him to limit animal protein because of kidney stones. When he told the bot about his medical history, it simply advised him to stop. He had to realize on his own that he'd been given dangerous advice.
Nick Tiller, who researches health misinformation at the Lundquist Institute in California, conducted a study in February 2025 asking five chatbots various health questions. Of 50 nutrition-related responses, 72 percent were rated as ineffective or harmful if followed. Shelley Wood, a gastroenterology dietitian in San Jose, regularly sees patients harmed by chatbot advice. One patient, after extended conversations with ChatGPT, had restricted herself to eating only papaya, spinach, chicken, and eggs for months. She lost a concerning amount of weight, developed fatigue and hair loss. The chatbot had no way to know about her medical history, no clinical judgment to apply, no ability to recognize when a recommendation had crossed into danger.
The core problem is that chatbots are trained on vast, contradictory datasets—books, websites, scientific studies, social media posts, Reddit threads. Some of that information is wrong. And when a chatbot encounters misinformation, it doesn't hesitate or qualify its response. It answers with confidence and authority, rarely acknowledging uncertainty. "It only takes a tiny amount of misinformation to confuse a chatbot," Tiller said, "yet they always respond confidently and authoritatively." OpenAI, ChatGPT's maker, acknowledged via email that the bots are not designed to replace medical care.
Experts now offer a careful framework for using these tools. Wesley McWhorter, a dietitian in Houston, calls using chatbots as accountability partners a "beautiful use" of the technology—but warns that rigid reminders can cause anxiety and do more harm than good. For anything beyond brainstorming meal ideas, the guidance is clear: run the chatbot's responses by a doctor. Better yet, see a dietitian, who has far more nutrition training than most physicians and can spend real time understanding your situation. Ask chatbots to cite their sources and verify them against reliable institutions like the American Heart Association or peer-reviewed journals. Try the same question on multiple chatbots; different answers suggest at least one is wrong. The technology is evolving rapidly, and responses may be more accurate today than they were months ago. But Tiller's warning remains: chatbots are not oracles. They are tools that can help with the easy questions and fail catastrophically on the hard ones.
Notable Quotes
It was like a cookbook tailored for me— Julie Bernstein, 76, on ChatGPT's meal planning responses
It only takes a tiny amount of misinformation to confuse a chatbot, yet they always respond confidently and authoritatively— Nick Tiller, health misinformation researcher at the Lundquist Institute
The Hearth Conversation Another angle on the story
Why do you think people are so willing to trust these chatbots with something as personal as their health?
Speed and availability matter more than we realize. A chatbot answers at midnight on a Sunday. Your doctor's office is closed. There's no copay, no waiting room, no judgment. For someone like Julie Bernstein, it felt like finally being heard.
But the studies show real harm. How do we square that with the positive experiences people are reporting?
The harm is often invisible at first. A restrictive diet feels like progress for weeks. Weight loss feels like success. By the time someone like that patient eating only papaya and spinach realizes something is wrong, they've already lost hair and developed fatigue. The chatbot never knew to ask the right questions.
What's the difference between a chatbot and a dietitian, really? Both are giving advice based on information.
A dietitian has clinical judgment. They know what questions to ask before answering. They notice when something doesn't add up. They can say "I don't know" and mean it. A chatbot will confidently recommend something harmful and never know the difference.
So should people just avoid chatbots for nutrition entirely?
No. They're genuinely useful for the grunt work—brainstorming meals, tracking nutrients you already know matter to you, staying accountable. The danger is mistaking convenience for expertise. A chatbot can help you plan; it shouldn't diagnose or manage a condition.
What would make you trust a chatbot's nutrition advice?
If it asked about my medical history first. If it admitted uncertainty. If it cited sources I could verify. If it said "this is a starting point, not a prescription." Right now, they do none of those things. They answer like they know, even when they're guessing.