explainer
Why two recipe sites give the same dish completely different calorie counts
Quick answer
<p><strong>Four undisclosed decisions account for almost all of it: serving size, ingredient specificity, whether cooking losses are modelled, and which composition database was used.</strong></p> <p>Serving size is the largest term by a distance — a lasagne divided into six versus eight servings changes the per-serving figure by a third with no change to the food. Ingredient specificity is next: "500 g minced beef" spans roughly 5% to 20% fat, which is a couple of hundred calories across the dish. Neither decision is usually shown.</p>
The same lasagne can be 420 or 780 calories a serving depending on four decisions the recipe never shows you. Here is how to read a recipe's nutrition panel and know which of those decisions it made.
Search a lasagne recipe and the nutrition panels will disagree by several hundred calories a serving. Neither site is being careless. They made four decisions you cannot see, and those decisions account for almost all of the spread.
1. Serving count — the largest term
A recipe’s per-serving figure is the total divided by a number somebody chose.
Divide the same lasagne by six instead of eight and every figure rises by a third. Nothing about the food changed.
Recipe sites pick a serving count that presents well. Readers divide the dish according to how many people are eating and how hungry they are. Those two numbers are almost never the same, and this single mismatch is the biggest reason a panel does not describe your plate.
The fix: check what the recipe assumed and compare it to how you actually cut it. It is one multiplication and it removes the largest error in the whole exercise.
2. Ingredient specificity
“500 g minced beef” is not a specification. Minced beef in a supermarket spans roughly 5% to 20% fat, and at 9 calories per gram of fat that is a couple of hundred calories across the dish.
The same applies to “a splash of olive oil” — which weighed between 11 g and 24 g in our own kitchen depending on the day and the pan — and to cheese, cream, and anything described by volume rather than mass.
A panel computed from 5% mince and one computed from 20% are both honest and describe different dinners.
3. Whether cooking losses were modelled
Meat loses water and renders fat. Some of that fat leaves the pan; some stays in the sauce. Vegetables lose water and concentrate.
Composition databases hold separate entries for raw and cooked states, and which one a recipe used is essentially never disclosed. The difference is commonly in the 5–10% range and it does not run in a consistent direction, which makes it hard to correct for even if you suspect it.
4. Which composition database
Different reference databases give slightly different values for the same food, because they were built from different samples at different times. Reputable ones agree closely; the residual spread is a few percent.
This is the smallest of the four terms and the one most people would guess is the largest.
Reading a panel properly
| Decision | Effect on the figure | Disclosed? |
|---|---|---|
| Serving count | ±33% | Number given, basis never explained |
| Ingredient specificity | ±15–20% | No |
| Cooking losses | ±5–10% | Almost never |
| Database choice | ±3–8% | No |
So a recipe panel is an estimate with a band of tens of percent, assembled from four undisclosed choices.
What we do
For dishes we cook regularly: log the ingredients. We weighed them, so we know them, and it removes the two largest error terms at once.
For dishes somebody else cooked: use the panel and accept the band. You are estimating anyway, and an estimate with a stated uncertainty is more useful than false precision.
Always check the serving count. That one correction does more than the other three combined.
Side-by-side comparison
| Decision | Typical effect on the per-serving figure | Usually disclosed? |
|---|---|---|
| Serving count (6 vs 8) | ±33% | Number given, basis never explained |
| Ingredient specificity (5% vs 20% fat mince) | ±15–20% | No |
| Cooking losses modelled or not | ±5–10% | Almost never |
| Composition database used | ±3–8% | No |
FAQ
Which recipe calorie count should I trust?
The one whose assumptions you can see, which is usually none of them. In practice, treat a recipe's panel as an estimate with a wide band and prefer logging the actual ingredients you used if the dish matters to your week. If you do use the panel, check the serving count against how you actually divided the dish — that single correction removes the largest error.
Why does serving size matter so much?
Because it is a division, so it scales the whole figure. A lasagne cut into six portions instead of eight has per-serving figures a third higher, with no change whatsoever to the food. Recipe sites pick a serving count for presentation and readers divide the dish however it suits them, and those two numbers are rarely the same.
Do cooking losses really change the numbers?
Yes, and in both directions depending on the dish. Meat loses water and renders fat; some of that fat leaves the pan and some stays in the sauce. Vegetables lose water and concentrate. A composition database entry for raw mince and one for cooked mince differ substantially, and which one a recipe used is almost never stated.
Is it better to log ingredients or the finished dish?
Ingredients, for anything you cook regularly. You weighed them, so you know them — and it removes both the serving-size guess and the ingredient-specificity guess in one step. Log the finished dish when somebody else cooked it and you are estimating anyway.