Can AI Count Carbs for You? Why the Basics Still Matter
Despite concerns about AI replacing jobs in health and wellness, I’ve fully embraced tools that make my work more efficient. From AI scribe that document patient visits to generating meal ideas for different dietary needs, I’m all in. (If AI could do my laundry, I’d be even happier!)
But as a registered dietitian and diabetes care and education specialist, one question keeps coming up: Can AI accurately count carbohydrates?
“Can I just take a picture of my food?”
“ChatGPT counts my carbs. Isn’t that good enough?”
“Do I still need to learn carb counting?”
I get the appeal. Carb counting can be tedious, and tools like ChatGPT, Claude, and Gemini make estimates as easy as snapping a photo. But just because AI gives you a number doesn’t mean it’s accurate.
In the fifth and final installment of Five Nutrition Questions Your Patients Are Asking, I explore the potential and limitations of AI-powered carb counting and why the basics still matter.
Five Nutrition Questions, One Practical Series
I recently had the honor of presenting at Colorado’s Association of Diabetes Care and Education Specialists annual conference. As diabetes care expands to include cardiovascular, kidney, and metabolic health, I was asked to discuss what’s new and what’s next in nutrition therapy.
That conversation inspired this five-part series, “Five Nutrition Questions Your Patients Are Asking.”Each article explores a common question, what the evidence says, and how to translate it into practical, person-centered guidance.
Catch up here:
What Can AI Do for Carbohydrate Counting?
AI tools can analyze food photos, identify ingredients, and estimate carbohydrate content. They can also help interpret nutrition labels, estimate restaurant meals, and suggest serving sizes.
These tools have the potential to make carbohydrate counting faster, more convenient, and more accessible. However, carbohydrate counting requires more than identifying what’s on your plate.
What a Food Photo Can't Tell You
Consider a photo of a bowl of pasta. AI might correctly identify the pasta and tomato sauce, but can it determine:
- Whether you’re eating one cup or two?
- Whether the pasta was measured dry or cooked?
- If the sauce contains added sugar?
- What ingredients are hidden beneath the surface?
Portion sizes, ingredients, preparation methods, and mixed meals can significantly influence carbohydrate estimates. AI may generate a precise-looking number without having all the information necessary to calculate it accurately.
What Does the Science Say About AI Carb Counting Accuracy?
Recent studies suggest that AI can be a useful carbohydrate-counting assistant, but accuracy varies considerably.
A 2026 study by Goncalves and colleagues compared ChatGPT, Gemini, and Claude with expert dietitians evaluating 30 meals. ChatGPT’s accuracy was not statistically different from the dietitians’, although its average absolute error was higher (20 versus 13 grams of carbohydrate).
Another 2026 study by Soyer and colleagues found that providing detailed meal descriptions alongside food photos improved estimates from certain AI models, particularly Gemini.
Similarly, research by Zagaroli and colleagues found that ChatGPT performed particularly well when given detailed descriptions of foods and quantities. However, this study evaluated written meal descriptions, not photographs alone.
The takeaway? AI performs better under some conditions than others, and the information you provide matters.
Why Carb Counting Skills Still Matter
For someone using mealtime insulin, an inaccurate carbohydrate estimate can have real consequences.
Underestimating carbohydrates may contribute to hyperglycemia, while overestimating may increase the risk of hypoglycemia. The American Diabetes Association’s 2026 Standards of Care emphasize ongoing education about carbohydrate intake and its relationship to insulin requirements.
Understanding carbohydrate sources, portion sizes, nutrition labels, and meal composition helps you recognize when an AI estimate makes sense and when it deserves a second look.
How to Use AI as a Carb-Counting Assistant
Rather than relying exclusively on AI, consider using it to support your existing skills:
- Provide context. Include a photo along with portion sizes, ingredients, and preparation details.
- Question the estimate. Ask AI to explain its assumptions and identify uncertain ingredients.
- Double-check. Compare estimates with nutrition labels, reliable food databases, or your own carbohydrate-counting knowledge.
- Use clinical judgment. Don’t rely on an unverified AI estimate alone to make insulin dosing decisions.
Here's How I Respond to My Patients
“Absolutely use the technology. I do! AI can make carbohydrate counting easier, especially for restaurant meals, mixed dishes, and unfamiliar foods.
However, it’s important to understand the basics so you can recognize when an estimate makes sense and when it needs a second look. The goal isn’t perfection. It’s using technology to make diabetes management easier while still making informed decisions.”
The bottom line: AI can be a powerful carbohydrate-counting assistant, but it shouldn’t replace the skills and knowledge needed to use those estimates safely.
Ready to Strengthen Diabetes Care and Education in Your Practice?
Are you looking to build, refine, or expand diabetes care and education services within your practice? I provide flexible, consultative support designed to meet the needs of busy clinics and healthcare organizations.
Diabetes care and education is central to my work. My approach focuses on translating current evidence and standards of care into clear, practical strategies that support sustainable behavior change and informed decision-making.
Let me help you develop patient-centered diabetes care and education that integrates seamlessly into clinical care. Book a discovery call to explore how we might work together.
- Goncalves S, Coelho C, Pretre L, et al. Chat, Gemini and Claude at the dinner table: assessing general-purpose AI tools for carbohydrate counting in the context of type 1 diabetes. Diabetes Res Clin Pract. 2026;231:113031. doi:10.1016/j.diabres.2025.113031
- Soyer AK, Akarsu GÇ, Yelli BD, et al. Evaluation of large language models for carbohydrate counting in type 1 diabetes mellitus: a comparative study with patient estimations. Diabetes Res Clin Pract. 2026;236:113279. doi:10.1016/j.diabres.2026.113279
- Zagaroli L, Caione N, Zara S, et al. Accuracy of ChatGPT, Gemini, Claude and DeepSeek in carbohydrate counting. Diabetes Obes Metab. 2026;28(7):5627-5636. doi:10.1111/dom.70747
- American Diabetes Association Professional Practice Committee for Diabetes. 5. Facilitating positive health behaviors and well-being to improve health outcomes: Standards of Care in Diabetes—2026. Diabetes Care.2026;49(suppl 1). doi:10.2337/dc26-S005
Discover more from Sarah Hormachea
Subscribe to get the latest posts sent to your email.
