Why portions matter (more than the prompt)
If you tell ChatGPT “chicken and rice,” it can only guess. If you tell it “6 oz chicken, 1 cup rice, 1 tbsp olive oil,” it has something to work with. Most “accuracy” problems come from missing amounts, not the AI being bad at nutrition tables.
The fastest portion formats (ranked)
- Best: measured units (cups, tbsp, oz, grams)
- Great: countable units (slices, eggs, tortillas, scoops)
- Good enough: hand-based units (palm, fist, thumb, handful)
- Last resort: “small/medium/large” plus context
Copy-paste prompt for “range” estimates
Estimate calories and macros (protein, carbs, fat). Meal: [what I ate] Portions: [rough amounts] Details: [brand/recipe/cooking method, sauces, oils, drinks] Output: 1) Low / typical / high estimate 2) Biggest uncertainty 3) The one detail I should specify next time
For the full workflow, start here: ChatGPT calorie tracker.
Portion phrase cheat sheet
Carbs
- Rice/pasta/oats: “1 cup cooked”, “2 cups cooked”
- Bread: “2 slices”, “1 bagel”, “1 tortilla”
- Potatoes: “1 medium”, “a fist-sized”
Protein
- Meat/fish: “palm-sized”, “about 6 oz”
- Eggs: “2 eggs”, “3 eggs”
- Beans: “1/2 cup”, “1 cup”
Fats (easy to undercount)
- Oil/butter: “1 tbsp”, “2 tbsp”, “buttered”
- Dressing: “1 tbsp”, “2 tbsp”, “on the side”
- Nuts/nut butter: “a handful”, “1 tbsp”
Restaurant portions (the honest approach)
Restaurants vary. The best you can do is describe the item, the size, and the add-ons. If you split it, say “ate half.” For restaurant-specific prompts and examples: ChatGPT calorie tracker for restaurants.
Two small habits that improve accuracy a lot
- Learn 5 “anchor” portions you eat often (rice, pasta, chicken, oil, dressing) once, then reuse that mental model.
- Always call out the multipliers (oil, sauce, cheese, drinks) even when you skip the exact measurement.