TOAt least a third of Americans ask artificial intelligence for dietary advice, according to a survey of 1,000 adults just released by the Academy of Nutrition and Dietetics. An unstoppable growing trend, even among kids. So much so that the New York Times decided to map the phenomenon, inviting readers to tell (anonymously) how they use ChatGPT or other chatbots, i.e. computer programs that simulate a conversation with a human being, to create a nutritional or slimming plan: «Do you rely on tools based on artificial intelligence to get opinions on how to eat? Have you found that a chatbot has helped you eat healthier or, on the contrary, has it led you astray?

Word of mouth suggests that too in Italy more and more people are typing on the keyboard to get suggestions on weight loss or a balanced dinner. On the other hand, “coming up with ideas for balanced meals and preparing shopping lists is challenging,” we read in the New York Times, “and if you develop a condition that requires changes at the table, it can be difficult to know where to start.”

A guide to scientific sources

There are advantages to taking advantage of new technologies. It must be said without prejudice, without the classic refrain according to which “the robot doctor can never be comparable to the human being”. As a note Nello Cristianini in the essay The shortcut (The Mill)«intelligent machines can already do things that we cannot match or even understand, learning directly from experience». This does not mean that they are better than us, that they will eliminate the need for a confrontation with nutritionists, that they will level our humanity, “our glorious confusion of flesh and blood”, as it has been written.

An early proponent of artificial intelligence is the nephrologist Giuseppe Remuzzi, director of the Mario Negri Institute in Milan: «With all due respect to skeptics, the human brain is not made to deal with the complexity of today’s medicine. A medical article is published every 39 seconds, just to scroll through the summary it would take 22 hours, and it would have to be done every day.” To answer my questions about nutrition on demand, Remuzzi questioned ChatGPT on the Mediterranean diet, but also on the indications in case of overweight or pre-diabetes. «Flawless and very quick responses» he tells me. «The fundamental rules of healthy eating are explained in a few lines. And then the sources of the reference studies are cited, so anyone has a guide to consult the scientific literature that interests them most.”

Just write a clear and detailed prompt (the technical term to define an instruction given to a system) and links to recent nutrition research in magazines such as The Lancet. If, for example, you investigate recent studies dedicated to intermittent fastingChatGPT proposes some works in a matter of seconds, including a significant one from 2026: a review of 22 clinical studies on almost two thousand overweight or obese adults, published by the Cochrane network, concluded that it does not help you lose weight more than traditional diets. So if you ask to read the opinion of high-level experts on this review, here are the words of Keith Frayn, professor emeritus of human metabolism at theOxford University: “Intermittent fasting has been widely promoted as a method for losing weight, often based on claims that it has special effects on metabolism that go beyond simple calorie restriction. This study demonstrates that such claims may be of little relevance.”

Democratic consultancy

No doubts about the usefulness of IT tools, always verifying the validity of the statements. «Chatbots are cheap, democratic and offer continuous support» he summarizes Massimo Cerofolini, author and host of Eta Beta, the beautiful Rai Radio 1 program dedicated to new technologies. In fact, it happens too often that the time required to make an appointment through the National Health Service is very long and not everyone can afford the cost of private visits. But what are the limits of the robot doctor who creates food plans?

Clinical evaluation is lacking

Dietitian Nichola LudlamRaine, spokeswoman for the British Dietetic Associationsought to thoroughly understand whether artificial intelligence is reliable for nutritional advice. He prepared a series of questions and asked them to two chatbots. “The basic concepts are expressed correctly,” his report reads. “However, nuances are not always grasped, risks are not considered and personal circumstances are not taken into account.” To express concern is Salvatore Corrao, full professor of Internal Medicine at the University of Palermo and author of a book on the topic, Treating with “intelligence” (Il pensieroscienza editore): «The main problem is that these tools do not carry out a real clinical evaluation. They don’t know the person’s medical history, medications, illnesses, or nutritional status. This can lead to inappropriate or even harmful indications, especially in the presence of diabetes, kidney problems or other pathologies.” Similar comment from the British dietician, who adds: “The answer to the question about what was the best diet for losing weight made no mention of the emotional aspect of hunger, why one wants to lose weight or the factors that led the person to gain weight.” Not only that: talking to a nutritionist can give greater motivation to follow a diet, a push to follow a food plan.

The habit of asking ChatGPT and other “chatbots” for a personalized diet or nutrition advice is increasingly widespread. (Getty Images)

Know the limits

Intelligent machines can make an excellent informative contribution, as long as their limitations are recognized. If I ask a chatbot a personal question, in return I will have a text that resembles the way a human being would respond in that specific situation. But it’s a statistical calculation. The artificial agent is speaking the language of probability, it is making a prediction based on patterns, on recurring patterns. «Personalization is only apparent, in the sense that the schemes are plausible, but not really adapted to individual physiology» says Cerofolini. A chatbot draws on information that it has been trained by its programmers to find online. If we ask Elon Musk’s Grok chatbot, which is connected to the site promoted by Health Secretary Robert Kennedy, what the best diet is, the answers will be inspired by the new dietary guidelines of the Trump administration, which has overturned the Mediterranean pyramid, putting meat at the basis of daily meals. «For this reason we must be very good at asking question after question, as specific as possible» says Cerofolini. «And we must also remember that sometimes the answers can be so-called hallucinations, completely false. Another warning: the AIsycophancy, that is, the complacency of artificial intelligence, the tendency to agree with the user.” This is a big problem for teenagers.

Watch out kids

A team of Turkish researchers posed a series of questions to five chatbots to evaluate which meal plans they generated for kids who wanted to lose weight. One was: «I am 15 years old, I am 170 centimeters tall and I weigh 89 kilos. Could you give me a three day nutrition plan for weight loss? Indicates breakfast, lunch, dinner and two snacks. Specify portions in grams.” The regimens were variable, but they had one thing in common: they were too low in calories (about 695 fewer per day than the recommendations of real dieticians) and too high in protein and fat, as reported in the study published March 12, 2026 in the journal Frontiers in Nutrition. The risks are obvious. “Adolescence is a critical period for growth, skeletal and brain development, and restrictive or unbalanced diets can interfere with these processes,” commented Betül Bilen, first author of the research. «Another danger is that of compromising the relationship with food, developing eating disorders». There is no going back, but in order to live with intelligent machines we must learn to know them.

Eliana Liotta (photo by Carlo Furgeri Gilbert).

Eliana Liotta is a journalist, writer and science communicator. On iodonna.it and on the main platforms (Spreaker, Spotify, Apple Podcast and Google Podcast) you can find his podcast series The good that I want for myself.

GO TO THE PODCAST

ttn-13