How the numbers work.
Where calorie values come from, how reliable they are and what to do when something is wrong. You should be able to see the evidence behind nutrition numbers.
Updated 15 October 2026
Where do calorie values come from?
From USDA FoodData Central using Foundation Foods (April 2026), SR Legacy (April 2018) and FNDDS 2021–2023. Our local import contains 13,588 foods and preserves their USDA identifiers and dataset provenance. Everyday database lookups do not require a live USDA API request.
For source energy, we prefer specific Atwater energy (2048) , then energy (1008), then general Atwater energy (2047). Missing energy remains unknown; we do not silently derive it from macros. Database values can differ from a particular product's label. Salt calculated from sodium is marked as calculated.
A database record is not your recipe. USDA includes both ingredients and prepared dishes. A record may not describe your homemade version. Ingredient-based examples identify the weights and source records used. Results depend on the recipe, cooking fat and finished weight, and raw, dry and cooked foods must be distinguished.
For a specific packaged product, check its nutrition label. Barcode search uses Open Food Facts internationally. Generic database records do not establish the values of every brand.
How uncertain are the values?
We have not measured a general error percentage for the app's analyses. Uncertainty depends on recognition, food selection and portion size. These factors can make a photo estimate easier or harder; they do not guarantee accuracy:
| Food type | Factor | Why |
|---|---|---|
| Separate pieces | The count may be visible | Pieces can be counted, but their size and recipe can still vary |
| Served portions | Quantity must be estimated | Both identification and gram weights can be wrong; weighing provides a firmer quantity |
| Mixed dishes | Ingredients may be hidden | Ingredients overlap in stews, soups and sauces, and fat content is not visible |
Weigh portions and check ingredients when possible. A weekly average can summarise your records, but cannot correct a wrong estimate or determine your calorie needs on its own.
How does AI estimation work?
The model proposes ingredients and quantities. The server looks for reliable USDA matches and scales their nutrients by the gram weight. Without a reliable match, the model's estimate is used and labelled. Results distinguish database matches, mixed sources and estimates.
Quantity matters. Database values are scaled to the proposed gram weight. If the quantity is wrong, the total is wrong even when the food match is correct. The model can also misidentify or miss foods.
Hidden cooking fat and drinks outside the frame may be missed. You can edit each value before saving. Confirmed preview values are saved without another AI analysis; the initial estimate is a starting point to review.
OpenAI performs food recognition. Food photos submitted for analysis are sent to that service. Private journal photos and weight records are excluded from AI analysis. Read the details in our privacy policy.
What does a meal's source mean?
Each meal identifies its nutrition source: USDA, historical FRIDA, a nutrition label, Open Food Facts, a community contribution, your own values or an AI estimate. Mixed sources are identified when only some ingredients match. A database value and an estimate have different evidential support.
The source stays with the value. Copied meals, saved dishes and imported records retain their original source. No new AI analysis runs, and copying an estimate does not turn it into a database match. Historical FRIDA records are never relabelled USDA. Imported values without a source are labelled as file imports.
Unknown is not zero. Fibre, sugars, salt and saturated fat are complete meal totals only when every ingredient has the value. Missing data stay unknown, and daily totals identify excluded meals. Vitamins and minerals retain their units and missing values. A recorded zero remains zero; an absent nutrient is never silently filled with zero.
When do we update an article?
An article's update date changes when its substantive content changes, such as a value, recommendation or section. Deployments and spelling corrections alone do not make the information new.
An update date should help you judge the age of the information. Replacing it without a meaningful content change would make it less useful.
What if you find an error?
Report an error by emailing [email protected] or using the contact page . Include the page or food record and what appears incorrect.
When a substantive value is corrected, the article's update date should reflect that change. Material corrections should be explained where readers can find them.
Calorie.one is an independently maintained project. Reports help identify mistakes, but neither the app nor its articles are guaranteed to be error-free.
Is this medical advice?
No. Calorie.one helps you record food. It is not a medical product or a substitute for professional advice.
Calorie needs and weight changes depend on more than sex, weight and activity. General examples do not suit everyone. If you have a medical condition, are pregnant or breastfeeding, take medication affecting weight or appetite, or have a current or past eating disorder, speak with your doctor or a registered dietitian before changing your intake.
If calorie tracking makes you uncomfortable, you can stop. It is useful for some people and unsuitable for others.