Can you trust the nutrition in an AI-generated recipe?
In short: Don't assume the nutrition values supplied with an AI-generated recipe are accurate. A general-purpose language model may produce plausible calorie and macronutrient estimates without calculating them from verified food composition data. To check the recipe, identify and quantify each ingredient, match it to an appropriate food composition database entry and calculate the nutrition using consistent quantities, preparation states and serving sizes. Review the result before using it in a client's meal plan.
1 min 16 s videoCaptions onUpdated 12 October 2026
Why the printed numbers need checking
A general-purpose chatbot asked for 'a 550 kcal chicken curry for two' may produce a recipe labelled as containing 550 kcal without independently verifying that figure. Unless the system uses a reliable food composition database and calculation process, the stated nutrition values may be generated estimates rather than the result of adding up the ingredients.
The stated energy may therefore match the requested target even when the ingredient quantities suggest otherwise. For example, 30 mL of cooking oil contributes roughly 240 to 250 kcal before accounting for the other ingredients. A recipe might still report an unrealistically low total if the oil or other ingredients haven't been included correctly.
The solution is to calculate the nutrition from the ingredients rather than relying on the model's stated totals. Each ingredient needs an appropriate food composition match and a defined quantity, with adjustments where relevant for edible portions, preparation state and cooking changes. How to calculate calories and macros in a recipe explains the calculation process.
Where generated recipes can go wrong
- Vague ingredients. Terms such as 'yoghurt', 'cheese' or 'milk' may refer to products with very different nutrient compositions. Check the type, fat content and any relevant product details before selecting a database entry.
- Quantities that are not weights. Descriptions such as 'a handful', 'one can' or 'to taste' may require clarification or conversion into measurable quantities. Household measures can be used when reliable conversion information is available, but assumptions should be documented.
- Raw or cooked. Food composition values differ according to preparation state. For example, 200 g of dry rice contains substantially more energy than 200 g of cooked rice because cooking changes its water content. Match the recorded quantity to the correct food state.
- Servings. Check whether the nutrition values refer to the entire recipe or an individual portion. Confirm the total recipe yield and how many servings it provides.
- Missing ingredients. Cooking oil, butter, sauces, dressings and garnishes can make meaningful contributions to the final nutrient totals. Check that ingredients mentioned in the cooking instructions are also included in the calculation.
One bad match can move the whole dish
Take coconut milk in a curry. Canned coconut milk and the carton drink sold for cereal share a name and are nothing alike.
The figures below are illustrative estimates rather than universal values. Nutrient composition varies between products, so the actual food or brand should be checked where possible.
| Matched to | kcal per 100 mL | 200 mL in the recipe |
|---|---|---|
| Canned coconut milk | about 200 | about 400 kcal |
| Coconut milk drink (carton) | about 20 | about 40 kcal |
Using these illustrative values, the difference is approximately 360 kcal across the recipe, or 180 kcal per serving if the recipe makes two equal portions. This demonstrates why selecting the correct ingredient entry matters. However, an accurate match must also reflect the quantity used and any relevant preparation details.
How to check a generated recipe
- Check the recipe itself. Review the ingredients and method for missing quantities, inconsistencies or ingredients mentioned in the instructions but omitted from the list.
- Verify ingredient matches. Match each ingredient to an appropriate food composition entry, checking the food type, preparation state and relevant product information.
- Confirm quantities. Use grams, millilitres or reliable household-measure conversions. Account for edible portions and relevant cooking changes where necessary.
- Check servings and yield. Confirm the total recipe quantity and how it is divided into portions before interpreting per-serving nutrition values.
- Review the nutrition calculation. Check the calculated energy and macronutrients for inconsistencies. A simple energy calculation can help identify potential errors, but it does not independently establish that the ingredient data or recipe totals are accurate.
Approximate energy (kcal) = (protein g × 4) + (available carbohydrate g × 4) + (fat g × 9)
Example: (38 × 4) + (45 × 4) + (20 × 9) = 152 + 180 + 180 = 512 kcal
This calculation uses general energy conversion factors and is intended as a sense-check. It does not account separately for energy from dietary fibre, alcohol or other components, and reported energy values may differ depending on the food composition database and calculation method used. In UK and EU nutrition data, carbohydrate generally refers to available carbohydrate, with fibre reported separately. A substantial discrepancy should prompt a review of the data and calculations, but there is no universal 10% threshold that confirms whether a recipe is accurate. To convert kilocalories to kilojoules, multiply by 4.184; 512 kcal is approximately 2,142 kJ.
Where AI recipes help, and where to be careful
AI-generated recipes can be useful starting points for meal ideas, dietary variety and adapting recipes to a client's preferences. For example, a practitioner might request a vegetarian dinner with a particular protein target or preparation time. However, the resulting recipe still needs to be checked for practicality, ingredient suitability and nutritional accuracy before it is used in a meal plan. The time needed for verification will depend on the recipe's complexity.
Extra care is needed for food allergies and therapeutic diets. An AI-generated recipe may include ingredients or products that conflict with a client's restrictions, even when those restrictions were included in the original request. Check ingredient lists, relevant product labels and potential allergen cross-contact rather than relying solely on the recipe description. Planning around allergies and foods to avoid explains these checks. For medical diets, such as those used in kidney disease or coeliac disease, verify the relevant nutritional and ingredient requirements against the client's individual clinical needs.
How to do it in Foodzilla
Foodzilla writes the recipe from your description, then matches every ingredient to its food database before you can save it, so the saved nutrition comes from those matches rather than the AI's estimate. Generating recipes needs the Professional plan or above.
- 1Go to Recipes and press Auto Generate (Generate on a small screen).
- 2In Describe your recipe, give the dish, how many it serves, any targets and anything to avoid, for example "A high-protein Thai chicken curry for 2, around 550 calories, no coconut milk". You can also tap one of the examples under Need a starting point?. Press Generate and wait about 40 seconds.
- 3In the Recipe preview, the Ingredients tab shows each line matched to a Foodzilla food, with its database, amount and energy. Where it says "matched to", check it picked the food you meant. Press Swap for an alternative, or type a new amount.
- 4A line marked No match found needs Find match, to search the database yourself, or Ignore to leave it out. The recipe can't be added until every line is resolved, unless you choose Use original as custom.
- 5Check the servings and energy at the top, then the Nutrition and Cooking Steps tabs. Press Add to My Recipes, and edit it later from My Recipes like any other recipe.
Good to know
- Specify ingredient quantitiesAsk the AI to provide measurable amounts for each ingredient, ideally in grams or millilitres, and to identify ingredients precisely, such as '15 mL olive oil' rather than 'a little oil'. This makes the recipe easier to check and calculate, although the quantities still need reviewing.
- Check nutrition before using custom recipesFoodzilla's 'Use original as custom' option retains the AI-generated ingredients and nutrition figures without matching them to food composition database entries. This can be useful for saving a recipe draft, but the nutrition values should be independently verified before the recipe is included in a client's meal plan.
- Prioritise accurate ingredient matchesCheck that ingredients are matched to the correct foods, quantities and preparation states before focusing on small differences in reported nutrient values. Ingredients contributing substantial amounts of energy or key nutrients, such as oils, dairy products and protein sources, deserve particular attention.