AI can't replace a registered dietitian (yet)

I'm a registered dietitian. I also use ChatGPT and Claude for parts of my own work and recommend our app to clients. I want to be precise about what AI can and can't do for nutrition.

What AI does well

1. Recipe generation from constraints. "I have these ingredients, this calorie budget, this macro target." Excellent.

2. Ingredient substitutions. "What can I swap for buttermilk?" Useful.

3. General nutrition education. "Why do I crave salty foods?" or "What's the difference between soluble and insoluble fiber?" Solid answers.

4. Calculation work. "If I eat X, how much fiber is in it?" Reliable when the ingredient is unambiguous.

5. Behavior tactics. "How can I reduce late-night snacking?" Reasonable, evidence-aligned suggestions.

What AI does poorly

1. Individualized medical nutrition therapy. Diabetes, kidney disease, eating disorder treatment, food allergies, GI conditions, oncology nutrition — these require an actual professional with access to your full clinical picture.

2. Eating disorder evaluation. AI assistants are not equipped to recognize the early signs of restrictive or compulsive patterns. They will give the same calorie target to a 22-year-old with no risk factors and a 22-year-old in early-stage anorexia. A dietitian or therapist will catch what the AI misses.

3. Pediatric nutrition. Don't.

4. Pregnancy and postpartum. AI can be a starting point but the stakes are too high for the specifics.

5. Real-time clinical integration. A dietitian working with you over months will notice things — the slight weight pattern, the off-handed mention of binge episodes, the change in sleep — that an AI conversation, even a long one, will miss.

6. Accountability that isn't transactional. A standing appointment with a human you trust is a different kind of behavior anchor than a chat window.

Where the line is moving

Generic AI advice is getting better. The Claude/ChatGPT of 2026 is meaningfully better at nutrition Q&A than the version of 2023, and will keep improving. The lines between "what AI can do" and "what humans can do" will keep shifting.

But the lines that won't move soon:

• Diagnosis
• Liability
• Medication-aware planning
• Long-term therapeutic relationship

For these, see a human.

How I personally use AI in my practice

I use Claude regularly to:

• Draft client handouts that I edit before sending
• Sanity-check my own meal-plan calculations
• Generate recipe variations for clients with restrictions
• Summarize new research papers when I don't have time to read fully

I don't use AI to:

• Make diagnoses
• Make medication decisions
• Replace the actual conversation with the client
• Write things I'd be embarrassed to put my name on

How a layperson should use AI

If you don't have a clinical condition and you want general nutrition guidance: AI is a fine starting point. Use it to answer questions, generate recipes, learn the basics.

If you have a clinical condition or a complex history: see a registered dietitian. Bring the AI conversations with you if you want — most of us are happy to look at them and say "this part is right, this part isn't quite right for your situation."

What our app's role is

CalorieScan AI is a tracking tool. It is not a clinician. We are explicit in our onboarding about this. If you have an eating disorder history or an active medical condition, we suggest involving a clinician who can supervise your tracking, not just opening the app on your own.

We've had several clinicians ask us about a "clinician mode" that lets them collaborate with patients. We're working on it.

AI is a great study partner. It is not your physician.