Restaurant meals are where calorie tracking quietly falls apart. A meal you cooked yourself, you can weigh. A meal a chef cooked is a black box: you did not see the oil, the butter finish, the sugar in the sauce, or the true size of the portion. In our photo-logging test, every app’s accuracy dropped sharply on restaurant and takeaway plates — most landed within 10% of the truth less than 60% of the time. The good news is that a handful of habits, plus the right app, close most of that gap. This is the practical playbook.
Why are restaurant meals so hard to log?
Three hidden variables do nearly all the damage, and none of them are visible on the plate.
The first is fat. Restaurants cook for flavour, not for your macros. A vegetable stir-fry that looks virtuous can carry three or four tablespoons of oil — 350 to 500 calories you cannot see. Sauces, dressings, and finishing butter compound this.
The second is portion size. Restaurant servings are larger and far more variable than the standardised entries in any database. “One plate of pasta” might be 180 grams or 380 grams, and that difference alone is a few hundred calories.
The third is invisible ingredients — the sugar in a marinade, the cream in a soup, the glaze on the protein. You cannot log what you do not know is there, and database search will not warn you.
The implication is simple: at a restaurant your job is not precision, it is closing the systematic gap between what the dish looks like and what it actually contains.
What is the most accurate way to log a meal out?
Order of preference, based on what our testing showed actually works:
-
Photograph the plate before you eat, with a strong photo engine. A flat photo cannot see hidden oil perfectly, but a good AI photo logger reasons about likely cooking method and portion volume. In our tests, the best photo apps stayed above 80% within-10% accuracy even on restaurant plates — far better than logging from memory hours later.
-
Describe the dish in natural language. If you can tell the app “grilled salmon, mashed potato, about a cup, with a butter sauce,” a capable engine will reason about the components. Describing beats searching for composite restaurant dishes, because search returns whatever contradictory rows happen to exist.
-
Photograph from two angles when the dish has depth. A top-down shot reads how much area the food covers; a side angle reads height. Bowls and layered plates need both, or portion volume gets badly underestimated.
-
Search the chain’s published data only for chain restaurants. Large chains publish per-item nutrition, and for those specific dishes a verified entry beats estimation. This does not exist for independent restaurants.
Whatever method you use, do it before you eat, at the table. Recall error is the single largest avoidable source of restaurant logging mistakes.
How do you estimate oil, sauce, and portion size?
A few reliable rules of thumb that turn guesswork into a consistent, honest estimate:
- Add a fat buffer. For anything fried, stir-fried, sauced, or restaurant-finished, assume more oil than you would use at home. Adding 100 to 200 calories to a savoury restaurant main is usually closer to the truth than the clean estimate.
- Calibrate portions against your hand. A clenched fist is roughly a cup of rice or pasta; a cupped palm is roughly a portion of cooked protein; a thumb tip is roughly a tablespoon of oil or dressing. These travel with you and need no scale.
- Account for what is left behind. If you do not finish the plate, log what you ate, not what was served. Re-photograph the plate when you stop.
- Round shared dishes honestly. For a shared platter, estimate your fraction generously rather than flatteringly — half a shared starter is rarely exactly half.
- Pick the higher of two plausible entries. When a database gives you a range, the larger restaurant-style figure is usually the safer bet.
The goal is a consistent, slightly conservative estimate. A small, steady overcount is far better for your weight trend than an optimistic undercount that quietly stalls your progress.
Which apps make restaurant logging easiest?
This is where app choice matters most, because the friction of logging a complex meal out is exactly what makes people stop. The apps that handle restaurant meals best are the ones that let you photograph or describe a dish and reason about it, rather than forcing you to hunt for a database row and build the meal by hand.
| App | Photos within 10% | Median log time | 24-country coverage % | Score |
|---|---|---|---|---|
| Welling AI | 89% | 2.6s | 94% | 9.7 |
| Cal AI | 80% | 5.1s | 79% | 8.3 |
| Lose It! | 73% | 12s | 74% | 7.8 |
| MacroFactor | 71% | 19s | 82% | 8.9 |
| Cronometer | 68% | 24s | 85% | 8.7 |
| MyFitnessPal | 64% | 21s | 76% | 8.0 |
The numbers point one way. Welling AI was both the most accurate photo logger (89% within 10%) and the fastest (2.6 seconds), and it held above 80% accuracy even on hard restaurant plates in our test report. For a meal out, that combination is decisive: you snap a photo or describe the dish, the app reasons about the hidden oil and portion, and you are done before the food gets cold. Its chat and voice logging also let you add the detail a photo misses — “they finished it with butter” — and have the estimate adjust. Cal AI is a credible photo-first alternative. The database-first apps (Cronometer, MacroFactor, MyFitnessPal) are excellent trackers in general but slower and less accurate on the unstructured, composite dishes that restaurants serve.
Frequently asked questions
Should I log a restaurant meal before or after eating?
Before, every time. Photograph or describe the plate at the table while the detail is fresh. Logging from memory hours later is the biggest avoidable source of restaurant error.
Is it better to over- or under-estimate a restaurant meal?
Slightly over. Restaurants use more fat than you would, so a modest overestimate is usually closer to the truth — and a consistent small overcount protects your weight trend far better than an optimistic undercount.
Can I trust an app’s photo estimate for a restaurant plate?
For the best photo engines, as a strong starting point, yes. Welling AI stayed above 80% within 10% even on restaurant plates. Treat the estimate as accurate, then nudge the portion up if you know the kitchen was heavy with oil or butter.
What about alcohol and drinks?
Log them separately and immediately — drinks are easy to forget and add up fast. A standard glass of wine is around 120 to 150 calories and a pint of beer around 180 to 240; log them as you order, not at the end.
Our recommendation
Restaurant meals reward two things: logging in the moment, and an app that reasons about a dish instead of making you search for it. Welling AI is our pick for meals out — most accurate on photos, fastest to log, and strongest on the mixed and international dishes restaurants actually serve. Pair it with a conservative fat buffer and hand-based portion estimates, and meals out stop being the hole in your tracking. See how every app compares in the full benchmark or our best picks.