When a nutrition label, published restaurant data, and an AI estimate disagree, start with the source that most directly matches the exact food and portion in front of you. For an unopened packaged food, the package usually comes first. If a chain restaurant provides current data for the same item and size, use that. For home cooking, cafeteria meals, and takeout without clear nutrition data, AI can provide a practical estimate.
nosh is useful for filling in the parts of real life that do not come with labels. When you have more direct, verifiable information, you can edit the food name, calories, and portion instead of letting the AI’s first draft override the package or an actual recipe.
With a nutrition label, check the unit before the number
A yogurt label might list energy per 100 grams while the cup contains 200 grams. If you finish the cup, you need to calculate for 200 grams. Reading only the number in the nutrition table can easily turn a full cup into a 100-gram serving.
Some labels list values per serving; others use 100 grams or 100 milliliters. Check the unit first, then compare it with the net weight and the proportion you actually ate. If two bags of cookies sit inside one larger package, find out whether “one serving” means one bag, a few cookies, or the whole package.
Package data is closer to the specific product, but it still requires context. A label from an old flavor, size, or formula should not be applied to a different version automatically.
Restaurant data helps when the item, size, and order match
If a chain restaurant publishes nutrition information for a standardized item, it usually knows the recipe better than a photo does. But a large and a medium drink, regular and zero sugar, extra sauce and no sauce, or an item and a full combo are not the same data point.
Published calories for a burger do not automatically include the fries and drink beside it. Nutrition data for a bowl of noodles may assume all the broth and a fixed set of toppings, while you left half the broth and added something else.
Before using restaurant data, check the product name, size, options, and publication date. If any of them do not match, the number becomes a nearby reference rather than direct data.
AI estimates matter most when there is no standard answer
A home-cooked stir-fry, a cafeteria plate with several dishes, or takeout without nutrition information rarely has one published entry that matches your exact portion. AI can identify foods from a photo, infer portions, and match them with nutrition data to create a quick everyday reference.
Its advantage is broad coverage with little effort. Its limitation is that it cannot see exact weights, the amount of oil or sugar, fillings, or the actual recipe. AI is not a final authority. It is the most convenient first draft when direct information is missing.
After nosh produces a result, still check the main foods, the portion you actually ate, and anything clearly missing.
When three numbers disagree, ask four questions
First, are they describing the same food? Plain and fruit yogurt, or poached and fried chicken, cannot share one nutrition entry just because their names are similar.
Second, are they describing the same amount? Values per 100 grams, per serving, and per package need to be converted to the same basis before they can be compared.
Third, are they describing the same way of eating it? Finishing the sauce, choosing a sweetened drink, or leaving some rice all change what you actually consumed.
Finally, which source is most direct? When a package label and measured weight match the food in front of you, use them first. Fully matching restaurant data comes next. When only a photo is available, keep the AI estimate.
Many apparent conflicts are really mismatches in units, sizes, or what was eaten.
Do not average the three numbers
If the package says 420 calories, the restaurant page says 480, and AI estimates 560, taking the average will not make the answer more accurate. Find out what each number describes. The package may refer to one item, the restaurant may include a full combo, and AI may have interpreted the portion or sauce differently.
nosh currently lets you edit calories, food names, portions, and log times. If package or restaurant data clearly matches what you ate, correct the entry directly. If you cannot confirm the match, keep the estimate—and its uncertainty.
Reliability does not mean always trusting one type of source. It means matching the data to the food in front of you. The closer the source is to that food, the more consistent its units are, and the easier it is to verify, the more weight it should carry.