How accurate is this?

Short answer: the calories per serving are typically within about 10–15% for a normal home recipe, and the ranking of two recipes against each other is far more reliable than either absolute number.

That is genuinely useful for “is this a 400 or an 800 calorie dinner” and for “does swapping to whole wheat actually add much fibre”. It is not precise enough to count grams against a medical target without checking the specifics.

Here is what moves the numbers, roughly in order of impact.

1. How you divide it — up to 100% error

This is by far the biggest one, and it is the only one entirely in your hands. A lasagne written for six that two people finish over a weekend is a three-serving lasagne. Every number on the page doubles.

Recipe authors are also optimistic. “Serves 4–6” usually means four adults or six people who are also eating something else. When the recipe states a range, this calculator takes the lower number, which never understates a portion.

Fix: weigh the finished dish once, then divide by what you actually plate up.

2. Cooking losses and gains — 0 to 30%

Nutrient data is for the ingredient as listed, usually raw. Cooking changes the weight, and sometimes the nutrition:

Water loss. A 150 g raw chicken breast is about 110 g cooked. The calories are the same; the calories per gram rise by a third.
Fat loss. Grilling or draining mince removes a meaningful part of its fat. Browning 500 g of 15%-fat mince and draining it can remove 20–30 g of fat — about 200 kcal off the recipe.
Fat gain. Deep frying adds oil that no ingredient list mentions. Batter can absorb 10% of its weight.
Discarded liquid. Pasta water, blanching water and brine take some minerals and water-soluble vitamins with them.
Marinades. Most of a marinade goes in the bin, but the whole thing is in the ingredient list.

Fix: delete or reduce ingredients that do not end up on the plate. If you drain the fat off mince, switch to a leaner mince entry to approximate it.

3. Volume-to-weight conversion — 5 to 25% on that ingredient

Cups measure space, not mass, and how a food packs varies a lot. A cup of flour is 120 g scooped lightly and 150 g if you pack the cup — a 25% swing on the single biggest ingredient in most baking.

Volume conversions here use per-ingredient weights (flour 125 g/cup, sugar 200 g/cup, honey 339 g/cup) rather than a single density, which removes most of the error. What is left is how you fill a cup.

Fix: if the recipe gives grams, use grams. Rows converted from a volume or a count are tagged est. weight.

4. Ingredient variation — 5 to 20%

“Chicken breast” is a range, not a value. So is an onion, an apple, and especially anything branded. Bread, stock cubes, sauces and plant milks vary enormously between products — sodium in particular can differ threefold between two brands of the same thing.

USDA figures are averages across samples, which is the right choice for a generic recipe but will not match a specific packet.

Fix: for anything where the brand really matters — protein bars, sauces, plant milk, stock — read the label and enter the numbers you actually have.

What this gets right

Comparisons. Both versions of a recipe carry the same assumptions, so the difference between them is much more accurate than either total.
Ratios. The macro split, and the share of calories from saturated fat, barely move with weight errors.
Ranking contributors. “Most of the salt is the soy sauce” is reliable even when the absolute figure is not.

What it does not do

Model cooking losses automatically.
Know brand-specific values without you entering them.
Account for glycaemic response, bioavailability, or how a food actually affects you.
Constitute medical or dietary advice. If you are managing a condition, use this to explore and take the specifics to a professional.