AI Chatbots for Nutrition: A Handy Helper, Not a Dietitian

More people are typing food questions into AI chatbots: how many calories are in this meal, what should I eat this week, is this food healthy? The answers sound confident, which makes them easy to trust, so researchers have started testing how accurate they really are. Let’s see what they found, and how to get the best from your new kitchen helper. This is general education, not medical advice.

What they are good at

Chatbots handle general questions well: explaining what protein or fiber does, suggesting meal ideas, comparing two foods in broad terms or turning a recipe into a shopping list. For brainstorming and understanding, they are a wonderfully handy helper.

Where the numbers get shaky

The studies so far focus on the numbers, and the results are mixed.

• Nutrient estimates. A 2025 study in Nutrients compared five AI models, including a version of ChatGPT, against nutrition labels and dietitian calculations. Calorie, protein and fat estimates were roughly 70 to 90 percent accurate, which is a solid start, but sodium and saturated fat were badly underestimated.
• Single foods. Another study tested ChatGPT-3.5 and ChatGPT-4 on 222 foods against a Taiwanese food database. Energy estimates landed within 10 percent of the database value for only about 35 to 48 percent of the foods, and the chatbots tended to overestimate protein.
• Meal plans. Dietitians reviewed 1,800-calorie menus generated by ChatGPT for weight management. The menus did not fit dietary recommendations: calories were low, and fats, carbohydrate spread, potassium and calcium were off.
• Following guidelines. In another dietitian review, accuracy against guidelines varied by case, and in one case no chatbot reached 50 percent.

The pattern across these studies is consistent: useful for rough ideas, weaker on exact figures, micronutrients and complete plans.

Getting more out of a chatbot

• Ask for a range, not a single number, and treat it as an estimate.
• Check the numbers against a database or a label. Weight Wire’s Food Lookup searches USDA FoodData Central and Open Food Facts, and it is free to use.
• Take extra care with sodium and saturated fat, where the errors were largest.
• Treat meal plans as a first draft. A registered dietitian can tailor one to your needs.
• Leave medication and medical conditions to your clinician. They are the right people for those questions.
• Think about privacy. You do not need to share personal health details to ask a general question.

How it compares with other tools

Chatbots join a long list of tools that estimate calories with some error. See our guides to calorie-tracking apps and to photo-based calorie counting.

A helper, not a replacement

These tools are improving quickly, so results may look different next year, and that is exciting. For now, they are best used as a quick assistant whose answers you double-check, not as a substitute for a label, a database or a professional.

Sources

• Comparison of five AI models’ nutrient estimates with nutrition labels and dietitian calculations. Nutrients, 2025;17:3044.
• Cross-sectional comparison of ChatGPT-3.5 and ChatGPT-4 energy and macronutrient estimates for 222 foods against a Taiwanese food composition database.
• Can artificial intelligence (AI) chatbot tools be used effectively for nutritional management in obesity? 2025 (PubMed 40111941).