The Conversation
We cannot deny it: it is a revolution. The arrival of artificial intelligence (AI) is bringing about such a profound change that it forces us to even rethink how we take care of our diet. But can we trust it as a reliable ally in nutrition?
Queries to AI are much easier than those we can do through search engines on the internet. He understands what we ask him, handles large volumes of information and presents his conclusions convincingly.
However, we give it a credibility it has not yet earned. Its probabilistic nature means that its responses cannot be considered verified information, which in health matters entails considerable risk.
App Binge
A revealing fact: according to OpenAI data, one in four weekly ChatGPT queries is about health. Although there are no specific figures for nutrition, technology has been present in this area for years.
(They investigate the use of artificial intelligence to generate child pornography in Texas)
In 2022, the apps of nutrition had 1.4 billion users worldwide and no less than 30% of mobile phone owners already have this type of application installed. The integration of AI in them is advancing rapidly: data from 2025 is expected to confirm that half of those apps They already incorporated it.
Reliability, the great challenge
The ChatGPT, Gemini, Claude and Copilot generative AI models are the most used in the world, and all of them are mainly powered by the Internet. The problem is that the reliability of the Internet in certain matters is very low.
As we discussed in a previous article, much of the information on nutrition on the internet is unreliable. For example, a 2022 study analyzed the information on-line about health claims made about yogurt, kefir, kombucha, fiber and prebiotics. The result of the evaluation was a score of just three out of ten: a resounding failure.
As the advantaged heir of the internet, AI has our trust implicitly. In a survey conducted in the United States, only one in four people believed that health information generated by AI is not very reliable or not at all reliable.
But if we analyze it, the AI does not approve either. A study that compared the responses of dietitians and ChatGPT to 928 questions about diet resulted in an agreement coefficient of 0.42, below the 0.5 considered the minimum acceptable.
Mistakes to personalize the diet
If general AI already shows these reliability limitations, the challenge becomes more complicated when we try to apply it to personalized nutrition. In order for you to act as a go-to nutritionist, your recommendations must have been validated in rigorous studies that confirm that they are accurate, safe and appropriate for the profile of each user. And today, that validation is still non-existent.
The nutritionist bases his work on a comprehensive evaluation: what and how much the patient eats, his biochemical and physical parameters, his physical activity and whether he suffers from any pathology.
One of the main obstacles that AI must overcome is accurately recording intake. This includes: identifying the food that appears in the images, separating the food into its different components or parts, calculating portion sizes and leftovers to estimate how much has actually been eaten, and estimating nutrient intake.
At the moment, general AI models do not perform this task accurately. ChatGPT makes a good assessment of weight in small portions, while in medium or large portions it underestimates it by up to 50%. A study in adolescents also showed that commonly used AI models underestimate nutrient intake compared to dieticians’ estimates.
The apps Nutrition policies present similar limitations, especially in dishes made with several ingredients, such as stews. On the other hand, the use of devices available on the market to record physical activity and calculate energy expenditure accumulate errors that cannot be ignored.
How to use AI responsibly
Given this panorama, it is worth asking: can we use AI in nutrition responsibly? The answer is yes, but with two essential conditions.
The first condition is to develop what is known as digital health literacy. The WHO defines it as:
“The ability to search, find, understand and evaluate health information from electronic resources, as well as to apply the knowledge acquired to solve health-related problems.”
Without this competence, the user is exposed to following incorrect, incomplete or downright dangerous recommendations. The digital divide is not just a question of access to technology, but also the ability to use it critically.
The second condition is not to lose sight of the role of the professional. AI has enormous potential to support the practice of nutrition: it can help process large volumes of data, identify patterns in diet or facilitate access to general information. However, It cannot replace the individualized clinical evaluation carried out by a professional. Nutrition is a discipline that requires context, judgment and empathy, qualities that AI does not yet possess.
The most sensible path is that of complementarity. A well-validated AI, used by a person with sufficient digital literacy, at the service of a professional who interprets its results. Only in this way can the bridge between artificial intelligence and nutrition be built on solid foundations.
