How to log restaurant and takeaway meals

Home cooking is trackable because you did the shopping. You know it was 300 g of chicken thigh, you know you used two tablespoons of oil, and if you are unsure you can read the packet. A restaurant removes every one of those facts at once.

The fat. This is the one that moves the number most, and it is invisible. Fat carries 9 kcal per gram — more than twice what protein or carbohydrate carry — so small differences in how generous the kitchen is with the oil bottle turn into large differences on the plate. The arithmetic is worth looking at directly.

A restaurant vegetable stir fry and a home vegetable stir fry can contain the same vegetables and differ by 300 kcal, entirely because of what happened in the wok. Nothing in the photograph shows you which one you are looking at.

The portion. Restaurant portions are not standard, and they are not the portion a food database means by “one serving”. This is the unknown you have the most control over, because you are sitting in front of it — the same hand-size and reference-object tricks that work at home work at a table, and estimating portions without a scale covers them properly.

The recipe. Was there cream in the sauce? Sugar in the marinade? Butter under the skin? You cannot know, and asking the waiter usually produces a confident answer from someone who did not cook it. Assume the more calorific version. Kitchens are not trying to trick you; they are trying to make the food taste good, and fat, sugar and salt are how that is done.

Before estimating anything, check whether the number already exists. A surprising share of eating out happens at businesses that are legally required to tell you.

• In England, out-of-home food businesses with 250 or more employees have had to display calorie information on menus since April 2022.
• In the United States, chains with 20 or more locations have to show calories on menus and menu boards, and must provide fuller nutrition information on request.
• Elsewhere it varies, and independents mostly publish nothing. But many chains publish a full nutrition table on their website even where no rule obliges them to — it is worth a search before you guess.

When the full table exists, take the macros and not just the calories. A chain burrito and a chain pasta dish can carry identical calories and 30 g of protein between them, and that difference is the whole point of what your macro split is actually for.

One practical caveat: check what the figure covers. “Chicken burrito, 720 kcal” may or may not include the guacamole, the sour cream and the tortilla chips that arrived with it. If the menu lists extras separately, log them separately.

The single most useful habit is to stop trying to price a meal as one object. “A curry” is unloggable. “A tub of rice, a piece of chicken, a creamy sauce and a naan” is four small estimates, each of which you can reason about, and errors in four small estimates partly cancel rather than compounding into one wild guess.

Work through a takeaway chicken tikka masala with pilau rice and a plain naan. The per-100 g values below are ordinary food-database figures; almost all of the width in the range comes from the portion and the fat, not from the lookup.

The honest answer is “somewhere between about 1,150 and 1,850”. The useful answer is 1,500. Both are true, and the second one is the one you can put in a diary and act on.

Notice where the uncertainty concentrates. The chicken is the part you are most confident about and the smallest contributor to the spread. The rice tub and the sauce carry most of it — one because takeaway containers are not a standard size, the other because you cannot see how much ghee went in. If you want to narrow the estimate, weigh nothing and simply ask yourself how much rice is left in the tub.

The same four questions, every time

1. What is the protein, and how big is it? A palm-sized piece, two palms, half a palm. This is usually the easiest part to judge.
2. What is the starch, and how much of it? Rice, chips, bread, pasta. Judge it as a number of fist-sized volumes.
3. What fat did the kitchen add? The one nobody asks. Was it grilled, fried, roasted in oil, finished with butter, dressed, sauced with cream? Use the table above and add a line for it.
4. What came alongside? Bread and oil before the meal, a shared side, the last three chips off someone else’s plate. These are the entries that quietly disappear.

Everything above is estimation. This part is not. Alcohol carries 7 kcal per gram, ethanol has a density of 0.789 g per millilitre, and the strength of the drink is printed on the bottle or the menu. So:

grams of alcohol = millilitres × ABV × 0.789, and calories = grams × 7.

Three pints of lager and a double gin and tonic is roughly 780 kcal, and not one calorie of it appears in a photograph of your food. For many people this is the largest single thing missing from their log, and it is also the easiest to fix, because it is the only part of the evening that comes with its own specification printed on the side.

Two additions the formula does not cover. Sugar syrup in a cocktail is carbohydrate at 4 kcal per gram, and a bar measure of it is 15 to 30 g — so add 60 to 120 kcal per cocktail on top of the spirit. A regular 330 ml cola is about 35 g of sugar, which is 140 kcal, and unlike almost everything else on the table the label tells you exactly.

A restaurant plate is the hardest input a photo-based tracker gets. It is worth being precise about which parts of the job a camera does well and which parts remain yours.

It is genuinely good at identification and decomposition: naming the dish, listing the ingredients it can see, and giving you a starting portion in grams. That is the tedious part of manual logging, and it is the part a camera removes. It cannot see dissolved fat, and it cannot see the bottom of a bowl. When we tested three vision models on 30 meal photographs, the widest disagreements were on exactly the dishes a restaurant serves most — ramen, soup, shakshuka — where the portion is hidden below a rim and the fat is dissolved into the liquid. The narrowest were on countable things in plain view, like a plate of pancakes.

So use it as a first draft, and correct it in this order.

1. Photograph the plate before you start eating, with the whole plate in frame and something of known size next to it — a fork, a standard wine glass, your own hand. Half an eaten plate is a much worse input than a full one.
2. Fix the portion first. It is the largest lever, and it is the one thing you can judge better than any model can, because you are looking at the real object rather than a photograph of it.
3. Add the cooking fat as its own line. If it arrived glossy, something put that there. One or two tablespoons of oil, per the table above, is a defensible addition to almost any restaurant main.
4. Correct the ingredient list. If the sauce is cream and the model has read it as tomato, that is a 200 kcal difference you can fix in one tap.
5. Log drinks separately, computed rather than photographed.

This is why every number in Biteline’s result sheet is editable rather than final — the portion slider sits at the top and rescales every macro as you drag it, the ingredient list can be corrected, and a low-confidence read says so on screen instead of quietly presenting itself as a fact. On a restaurant plate you should expect to use all of that.

Two rules do most of the work, and they matter more than the precision of any individual estimate.

Round up. The errors in restaurant logging are not symmetrical. Nobody forgets the salad and remembers the bread; the things that go missing — the oil, the dressing, the bread basket, the two chips off a friend’s plate, the second glass of wine — are almost all additions. If you take the midpoint of your honest range you will still, on average, under-log. Take the upper half.

Be consistent. If you decide the local curry house portion is 1,500 kcal, log 1,500 every time you order it. This is the part people get wrong by trying to be clever: re-estimating the same meal from scratch on each visit adds noise without adding accuracy. A log that reads 1,500 every Friday tells you something real about your week. A log that reads 1,100 one Friday and 1,900 the next, for the same order, tells you about your mood.

A steady bias is a solved problem. If you consistently log 15 per cent under and your weight is not moving as the target predicted, the target gets adjusted and the bias is absorbed — which is the whole logic of setting a target from your own numbers and then correcting it against the scale, as working out how many calories to eat to lose weight goes through. Random error cannot be absorbed that way, because there is nothing stable to correct.

Some meals defeat component estimation entirely. Each of these has a practical answer, and in every case the answer is better than logging nothing.

Shared plates and tapas

Count items, not dishes. You did not eat “the padrón peppers”; you ate four of them. Tally what actually reached your plate and log that as one combined entry at the end, rather than trying to log six dishes at a fraction each.

Buffets and all-you-can-eat

Do not attempt to itemise. Set a per-plate figure once — a reasonable starting point is to treat a full plate as equivalent to a substantial restaurant main — and then simply count plates. Counting plates is something you can actually do accurately. Estimating a buffet is not.

Tasting menus

Log the event, not the courses. Ten small courses with sauces you cannot name is beyond any estimation method, including a camera. Pick a flat figure for the whole evening, log it as one entry, and accept that this particular day is a wide error bar. One such meal a month does not decide anything.

Late-night food you did not photograph

Log it the next morning, generously, as a single line. The temptation is to skip it because the estimate would be poor. A poor estimate is worth far more than a blank — a missing 1,000 kcal entry does not make the calories missing, it makes your log wrong in a way you will later mistake for evidence.

Someone cooked for you

Ask. People are pleased to tell you what went into their own food, and the one question worth asking is about the fat: how much oil, whether there was butter or cream. That single answer usually narrows the estimate more than everything else combined.

A business dinner you did not choose

Log the shape rather than the specifics: a protein, a starch, a sauce, a side, three glasses of wine. You will not get the dish right. The shape is enough to keep the week honest.

Log it and carry on. The most common damage from a restaurant meal is not the meal — it is the reaction to seeing a large number in the diary, which tends to be either abandoning the log for three days or eating almost nothing the next day.

Run the arithmetic instead. On a 2,000 kcal daily target, a week is a 14,000 kcal budget. A 1,500 kcal dinner where you would normally have eaten 700 puts you 800 over — under 6 per cent of the week. Spread across the following six days that is around 130 kcal a day, which is an adjustment you would not notice making. Nothing about that requires a recovery plan.

Deliberately under-eating the next day is worse than doing nothing, because it makes eating out feel like a debt to be repaid, and that is the relationship with food that ends a tracking habit. The point of logging a restaurant meal badly is not self-discipline. It is that the log stays continuous, and a continuous log with known bias is a usable instrument.

That is also the honest limit of every method on this page. None of it makes a restaurant meal precise. It makes it consistent, which is the property that lets a food diary tell you something true over a month. What Biteline does, and how the estimate works is on the home page — including the parts a photograph genuinely cannot know.