AI Can Give You Nutrition Advice, But Can It Understand You?
Updated: Sep 17
What Recent Research Tells Us About AI, Nutrition, and Why Clinical Oversight Still Matters
There's something fascinating happening in nutrition right now. You can open an AI chat box, describe what you eat, list your symptoms, and within seconds receive a meal plan, supplement suggestions, and your macros calculated to 0.1 of a gram. It all appears genuinely personalised. So, where does that leave us nutritional therapists? More importantly, what happens when nutrition moves beyond general healthy eating into the complex territory of symptoms, medical conditions, medications, and biomarkers?
I've been reading the research on this, and the answer isn't simply good or bad. AI is capable of providing nutrition advice, but the real question is whether it can understand individuals in a uniquely human way.
Where AI Genuinely Helps
AI processes enormous amounts of information instantly. It explains concepts clearly, generates meal ideas, and turns complicated science into something readable. Ask it, "What are some good sources of fibre?" or "How do I get more protein into breakfast?" and you get a great answer while you're staring into the fridge. This doesn't require a practitioner, and studies back this up—generative AI in nutritional sciences shows real potential.
So, the question isn't whether AI is useful. It's where its usefulness ends, and this is where the can of worms opens. In 2024, Ponzo and colleagues published a study in Nutrients asking whether ChatGPT could provide dietary advice that held up against established guidelines. The questions addressed several non-communicable diseases, had nutrition experts assess the responses, and then repeated the exercise with a more complex scenario involving multiple conditions.
For single conditions, appropriateness ranged from 55.5% for sarcopenia (age-related muscle loss) to 73.3% for non-alcoholic fatty liver disease. Only two recommendations directly contradicted guidelines, which is a reasonable starting point for general advice. However, when researchers asked ChatGPT to consider multiple conditions simultaneously, recommendations became contradictory and sometimes inappropriate. The authors concluded that it could not replace consultation with a qualified professional.
This is key because people rarely present with a single diagnosis and a single symptom. They have histories, medications, and supplements they might forget to mention. Stress varies hour to hour, and digestive symptoms often go unmentioned until the third consultation. Family history and genetics also play a role. These factors, which clients may not think are relevant, often matter most. AI can only be as good as what you tell it, and most people struggle to articulate their experiences, let alone identify clinically significant details.
Even the Numbers Don't Always Hold Up
A 2026 study compared AI-generated diet plans with dietitian-designed plans for adolescents. Five different AI models were tested. All of them underestimated energy needs by an average of 695 kcal a day. They recommended slightly more protein but around 115g less carbohydrate. This is a finding I want any parent to focus on, it isn't that the advice sounded wrong; it's that a plan can look complete and still fall hundreds of calories short of what a growing body needs.
Good Advice Isn't the Same as the Right Advice
A recommendation can be perfectly sensible in isolation and still be wrong for the person in front of you. If you tell AI, "I'm exhausted. What should I eat?" it suggests iron-rich foods, adequate protein, complex carbohydrates, hydration, B12, and vitamin D. None of those suggestions are unreasonable. But do they actually need more iron? Are they eating enough overall? Has their training load changed? Is their sleep adequate? Have they been restricting? Has their cycle changed? Are they on medication? Has this gone on long enough to warrant investigation?
The chatbot answered the question it was asked, but it can't ask follow-up questions that put the client's concerns in context. This is where a clinician's skills come into play—knowing which questions should come first and what follow-up questions need to be asked. Ponzo's team found something similar: many responses weren't contradictory to guidelines but weren't fully aligned either. That's not the same as being wrong; it's missing nuance, priority, and many caveats. To the client, a confident, well-written answer makes that gap very hard to spot.
"Personalised" — Based on What?
When an AI offers you a personalised plan, it's worth asking what the personalisation is actually built on. Is it based on your food preferences, goals, lifestyle, symptoms, blood tests, medical history, genetics, or responses to things you've already tried? The list goes on, and these factors are not interchangeable. Most consumer AI personalises based on preferences and lifestyle, but that is a long way from biological personalisation.
So Should You Stop Using AI for Nutrition?
Not at all! I encourage clients to learn how to use it effectively. This means asking more questions, understanding terminology, exploring ideas, generating recipes, and making sense of research papers. It can make nutrition feel less intimidating. It's another tool in the toolbox. However, it's crucial to recognise when it crosses from general information into individual health advice, particularly concerning persistent or unexplained symptoms, diagnosed conditions, medications, significant restrictions, complex digestive issues, or multiple concurrent issues.
That's where professional input earns its place. Not because I have secret nutrition facts that a chatbot doesn't, but because my job isn't merely to provide more information. It's to work out what's actually happening, what's missing, what matters most, what to change first, what to leave alone, how we'll know if it's working, and how we can pivot if it isn't. Then, we reassess and adapt because nutrition isn't static.
The Future Isn't AI Versus Practitioners
In my opinion, we must embrace the partnership; it should be AI AND practitioners. I genuinely believe this could be beneficial. AI can help analyse dietary records, spot patterns, summarise research, and reduce administrative work. The practitioner retains the elements that need clinical judgement: interpretation, prioritisation, context, monitoring, and knowing when to refer.
Perhaps the better question isn't, "Can AI give you a nutrition plan?" It's, "Does it understand why that plan is right for you?"
Good nutrition isn't just about knowing what foods contain; it's about being curious and asking better questions. It's about making sense of the answers, choosing what to act on, observing outcomes, and knowing when something needs a proper investigation. The ultimate goal isn't to make you dependent on AI or on me; it's to help you understand your own body well enough to make better decisions for yourself.
As I've said before, every person is beautifully unique, and that's why there will never be a one-size-fits-all answer. If you'd like support understanding what your own body needs, I'd love to help you navigate the evidence and find a personalised approach that's right for you.
References:
Hieronimus, B., Lopez-Aguirre, M.-L., Birringer, M. & Podszun, M. (2025). GenAI in Nutritional Sciences (GAINS): A Systematic Review and Reporting Framework for Future Research. Nutrition Research.
Ponzo, V., Goitre, I., Favaro, E., Merlo, F.D., Mancino, M.V., Riso, S. & Bo, S. (2024). Is ChatGPT an Effective Tool for Providing Dietary Advice? Nutrients.
Bilen, A.B., Kalkan, G.E. & Önal, H.Y. (2026). Artificial Intelligence Diet Plans Underestimate Nutrient Intake Compared to Dietitians in Adolescents. Frontiers in Nutrition
Comments