Logging a Meal With No Recipe

Name the parts, then widen the number#

When somebody else cooked and no recipe exists, stop trying to identify the dish and start identifying its components. Estimate the mass of what is on your plate, break it into the things you can actually see — a protein, a starch, a vegetable, a sauce — match each of those to several database entries rather than one, and log the result as a band wide enough to hold the cooking fat you cannot see. That last step is not a hedge. It is what professional dietary assessment does when it hits exactly this problem, and it has a written procedure.

The procedure is the FAO/INFOODS Guidelines for Food Matching, the standard national nutrition surveys follow when a person reports eating something the food composition tables do not contain. Its central instruction is the one home trackers almost never follow: "Do not just match to only one food item unless the food is infrequently consumed... Calculating a mean will at least reduce the bias" (FAO/INFOODS, 2012). Picking the single closest-sounding entry from a database and moving on is, in that framework, the worst available method — and it is what an app's search box invites you to do every time.

The protocol grades your guess before it uses it#

What makes the FAO document useful to someone standing over an unlabeled plate is that it does not pretend a good match is always available. It assumes it often is not, and tells you to sort matches into quality tiers and then say out loud which tier you used. The guidelines assign three grades:

Grade	What it means	A kitchen example
A — high	The food and all its descriptors match an entry exactly	A packaged item you scanned
B — medium	Averaged across several entries, or built from a recipe with yield and retention factors, or matched to a genuinely similar food	"Chicken curry" averaged over four database curries
C — low	"The food is very different but it is the closest match possible"; or raw values applied to a cooked dish with no adjustment	Camel meat logged as beef

Source: FAO/INFOODS Guidelines for Food Matching, Version 1.2 (2012), Table 1.

Two things in that table matter more than the rest. The first is that the document names the exact failure most trackers commit without noticing: applying raw-food values to a prepared dish "will lead to major errors in nutrient intake estimations," and it is a grade-C move rather than a small imprecision. The second is that grade B has a specific technique inside it — average across entries — that costs you thirty seconds and moves you a full grade up.

The guidelines are also candid that not every food deserves the same care. Quality "is mainly determined by the quality of the food matches of foods consumed in significant quantities," and "a lower quality food match is more acceptable for foods consumed infrequently." Translated to a plate: the 200 grams of rice and the 150 grams of braised pork carry the calories, and they are where the effort belongs. The coriander is a rounding error and can be logged carelessly on purpose.

Water first, then fat — the two checks that do most of the work#

Before averaging anything, the guidelines make you compare two properties between your dish and the entries you are considering, and the order is not arbitrary. "In general, water is the most important nutrient to check the food description and the concordance between two foods. Therefore, the water contents always have to be compared when matching foods."

Water is the right first check because it is the single largest thing separating two foods that share a name. A stewed version and a roasted version of the same ingredients differ mostly in how much water is still present, and water carries no energy while diluting everything that does. The rule that follows is concrete: when the water content of your food differs from the entry's by more than 10 percent, adjust all nutrients accordingly; when the fat content differs by more than 10 percent, adjust the fat-related values. You will not do this arithmetic at the table. But you can do the qualitative version in two seconds — is this wetter or drier than the picture in my head of the database entry, and is it glossier? — and let the answer push your estimate up or down before you commit it.

That is also the reason a photograph and a name together are still not enough. A food composition review of complex dishes puts the limit plainly: "many composite dishes with the same name show different characteristics between countries," and values arrived at by calculation "should almost always be observed as approximations"³. The entry you land on is a model of a category, not a description of the object in front of you — a point worth holding alongside why two apps hand you different numbers for the same food.

Someone else's kitchen is fattier, not bigger#

Here is the part that changes how you set the band, and it is the most useful single result for this problem. Researchers in Changsha, Hunan collected eight commonly consumed dishes from 20 small restaurants, 20 large restaurants, 20 urban households and 20 rural households, and chemically analysed them². Same dishes. Different cooks.

The restaurant versions were not simply richer versions of the home ones. They were recomposed. Across all dishes, fat ran 3.66 against 3.22 g per 100 g and saturated fat 1.10 against 0.84, with the share of energy coming from fat at 21.0 percent versus 16.0. Among vegetable- and legume-based dishes the gap was much larger: 3.03 against 1.89 g of fat per 100 g, and 30.8 percent of energy from fat against 18.2. Meanwhile protein in animal-based dishes was about 15 percent higher at home, and the energy density of those animal dishes was actually lower in restaurants — 211.5 against 231.3 kcal per 100 g.

The dish somebody else cooked is not reliably a bigger number. It is a different mixture wearing the same name — more oil, less meat — and a calorie total can absorb that swap without moving.

Read that carefully, because the naive correction is wrong. Adding a flat percentage to everything you did not cook would have overshot the animal dishes and undershot the vegetable ones. The transferable finding is directional and specific: the further a dish is from obviously fatty, the more its fat has probably moved. A stir-fried green vegetable made by someone else is the single least trustworthy thing on the plate, because it looks like the cheapest item and it is the one whose fat content nearly doubled. This is a study in one Chinese province with pooled composite samples rather than a global constant, so take the mechanism — someone cooking to taste, at volume, with a free hand on the oil bottle — rather than the decimal places. It is the same mechanism that makes homemade meals hard to count even when you were the cook.

Your confidence in the estimate carries no information#

There is one more reason to widen the band rather than sharpen the number, and it is about you rather than the food. When 197 adults estimated the energy content, sugar content and portion size of six real food items and then rated how accurate they thought each of their own estimates was, the relationship between how right they felt and how right they were was essentially nil — correlations of rs ≤ |0.20| between perceived accuracy and actual percentage deviation⁴. Participants underestimated the energy of a quiche in 84.8 percent of cases while rating their confidence no differently than on items they got closer.

That result is the whole argument against logging a single number for an unfamiliar dish. If confidence tracked accuracy, you could log tight when you felt sure and wide when you felt shaky, and the log would be self-calibrating. It does not, so the feeling of having recognised a dish is not evidence that you have. The band has to come from the structure of the problem — how much of the plate is sauce, how far the cook could plausibly have gone on the oil — rather than from how confident the guess felt.

Working a plate you did not make#

Five moves, in order, and the first two carry most of the accuracy:

Estimate total mass, not the dish. Grams on the plate is the number with the most leverage and the least ambiguity, and it is the half of the problem photographs actually help with when you shoot at an angle with something of known size in frame.
Split into visible components. Protein, starch, vegetable, sauce. Four buckets, roughly weighted by eye. You are not reconstructing a recipe; you are apportioning a mass.
Match each bucket to three entries and average them. For the big components only. This is the single move that takes you from the guidelines' grade C to grade B, and common food values are close enough to memorise for the repeat offenders.
Add the cook's fat as its own line. One to three teaspoons for a sautéed or dressed component is a defensible starting assumption, and keeping it separate means a later correction touches one number instead of five.
Log the band, and record which grade you were on. A dish where you saw every component and know the cuisine is a different claim from a covered casserole at a party. Both are legitimate entries; only one deserves a narrow range, and a tracker that lets you widen the estimate rather than forcing a digit is recording the difference instead of erasing it.

None of this makes an unfamiliar dish precisely countable, and it is not supposed to. It moves you from a guess with unknown quality to a guess with a stated quality — which, given that the whole stack carries roughly a fifth of error on a good day, is as much as the instrument can deliver. The dish you did not cook is not the weak link in your log. Pretending you know it to the calorie is.

FAQ#

How do I log a meal when I don't know what's in it?#

Log its components rather than its name. Estimate the total grams on the plate, split that mass into protein, starch, vegetable and sauce, match the large components to three or four database entries each and average them, then add a separate line for the cooking fat you cannot see. Averaging across entries rather than picking one is the explicit recommendation in the FAO/INFOODS food-matching guidelines, because a mean reduces the bias a single arbitrary choice introduces.

Is a friend's home cooking lower in calories than the restaurant version?#

Not dependably — the difference shows up in composition more than in the calorie total. Analysing the same eight dishes from restaurants and households, restaurant versions carried more fat (3.66 vs 3.22 g/100 g) and a higher share of energy from fat (21.0% vs 16.0%), while home animal-based dishes carried about 15% more protein and were actually higher in energy density². Assume the oil moved, not the calorie count.

Should I pick the closest database entry or average several?#

Average several, for anything you are eating a meaningful quantity of. National dietary surveys are instructed not to match an unspecific food to a single item, and to match at least three entries chosen to "reflect the range of different nutrient values available for the food," then take an arithmetic or weighted mean. One entry for a whole category imports whichever version of the dish happened to get sampled; three entries import the category's middle, which is what you actually know.

Sources#

This article was researched and drafted with AI assistance and reviewed for accuracy by the BurnWeek team. It is general information, not medical advice. How we research and correct our articles →