The AI Nutrition Reality Check

Good morning!

Ever stared into your fridge at a random mix of ingredients and asked ChatGPT to turn it into a meal plan? Or used an app that tells you exactly how “healthy” your food is after a quick photo? 

It feels seamless. A few taps and suddenly you’ve got diet advice, calorie counts, even a whole day of eating mapped out for you.

But as more people start relying on AI for ongoing nutrition decisions, not just one-off meals, a growing body of research is asking a simple question: how accurate is it, really?

Key Takeaways

• Consumer AI nutrition tools often show large errors (around ~36% for calories and weight estimates), making them unreliable for precise diet tracking.

• Meal plans generated by chatbots frequently miss the mark on calorie needs and macronutrient balance, sometimes by hundreds of calories per day.

• Accuracy improves dramatically only when AI is fed detailed personal health data (e.g., glucose, biomarkers), rather than just prompts or photos.

CORE

The App That Reads Your Plate

Snap a photo, get your macros back in seconds. It's one of the most appealing promises in nutrition tech right now, and apps built around it have grown rapidly. But the real question is whether they're accurate enough to actually guide your day-to-day eating.

A 2025 study published by researchers at the University of Gothenburg tested exactly this, putting three leading AI vision models through a rigorous evaluation using 52 standardized food photographs across small, medium, and large portion sizes.

Accuracy declined consistently as portion size increased, with large portions producing the greatest underestimation across every model tested. The app is most reliable when your meal is small and simple, which, let's be honest, is not when most of us need the help.

Food misidentification compounded the problem. In one instance, a model misidentified falafel as meatballs, producing a dramatic nutritional overestimate. In another, scrambled eggs were misidentified as pasta. Mixed dishes, unfamiliar cuisines, and anything with a sauce hiding what's underneath are exactly the meals that trip these systems up, and exactly the meals worth tracking.

A 2024 scoping review in the Journal of Medical Internet Research confirmed that lighting conditions, food presentation, and portion size are consistent challenges for AI image-based dietary assessment across the research literature.

The honest answer to whether these apps are worth using daily depends on what you're asking of them. Noticing general patterns in what you eat, probably fine. Treating the numbers as precise measurements, the research says not yet.

The Chatbot That Writes Your Meal Plan

The second, more consequential use of AI in nutrition is asking a chatbot to plan what you should eat. A study published in March 2026 in Frontiers in Nutrition put this to the test. They created four standardized profiles of overweight and obese 15-year-olds and gave five widely used AI chatbots, ChatGPT-4o, Gemini 2.5 Pro, Claude 4.1, Bing Chat, and Perplexity, simple, realistic prompts: age, height, weight, and a request for a three-day weight-loss meal plan.

Across all five models, AI plans systematically underestimated total daily energy by an average of 695 calories. Carbohydrate ratios fell to 32-36% of total energy, well below the 45-65% range recommended by the Institute of Medicine for adolescents, while protein and fat ratios exceeded guidelines across every model tested. No single model showed consistent accuracy across all 22 nutrients analyzed, with the largest gaps in vitamin D, folate, calcium, and iron.

The researchers point to a likely explanation: these models train on publicly available text, and online weight-loss content skews heavily toward low-carbohydrate approaches regardless of who they're appropriate for. The models absorbed that pattern and applied it uniformly, even when the prompts contained no low-carb framing whatsoever.

A Small Experiment

Here's a quick way to see these nutrient discrepancies firsthand. Take a simple food, like nonfat plain yogurt. Simple enough, right? Now ask your favorite chatbot for the macronutrients. Then go one step further and compare it to USDA’s Food Search for yogurt. This is the database nutrition professionals actually reference. You’ll find something as simple as yogurt is not reported correctly.

Try it with a few other foods and you'll notice the pattern, some will be closer, some won't, but the AI will deliver every answer with the same unshakeable confidence regardless. That's the part worth paying attention to.

A Possible Explanation

What photo-tracking apps and chatbot meal planners share is what they don't know about you. A photo app sees your plate. A chatbot sees a few typed sentences. Neither knows your health history, your medications, how you're sleeping, what your relationship with food looks like, or whether any kind of dietary change is even appropriate for you right now. And it’s probably better that way. A registered dietitian on the other hand works with you and gathers all of that before making a single recommendation.

That said, when AI is given real individual health data (like glucose levels, microbiome data, and blood biomarkers), results look very different from consumer apps.

A 2025 review in Healthcare (Basel) of 11 clinical studies found that these AI systems, built for continuous, personalized feedback, often outperformed standard dietary approaches. In some cases, outcomes included reduced IBS symptoms, improved metabolic markers, and higher diabetes remission rates compared to conventional care. 

Key point: these weren’t chatbots, they were data-driven medical systems adjusting in real time to each person’s biology.

But the evidence is still early, small studies, short timelines, and limited populations.

ENDNOTE

Final Thoughts

AI nutrition tools can be useful for building general awareness of what you’re eating day to day. Even with a ~36% margin of error in photo-based tracking, they can still help you spot patterns you’d otherwise miss. But that’s about where it ends. They’re not yet precise enough for strict calorie tracking or personalized nutrition planning, even if they sound extremely confident while telling you your yogurt is “basically pasta.”

So for now, you can use them as a rough guide… just maybe don’t let them plan your entire diet unless you’re also okay with your scrambled eggs occasionally becoming pasta.

For the finer details, you’ll still want a human dietitian, preferably one who knows what yogurt actually is.

Until next week!

Adrian Macdonald | Team Dietitian | The 8% Newsletter Author