Why "accurate" is the most overused word in calorie tracking

Open the App Store page for almost any AI calorie tracker and you will find a headline accuracy claim: 90 percent, 95 percent, "the most accurate AI food scanner." Almost none of them say what that percentage is actually measuring, and the ones that have been independently checked tell a less flattering story than the marketing copy.

What the actual documented numbers say

Cal AI, one of the category's largest apps by downloads, has been repeatedly reported by users and independent reviewers to be off by 20 to 40 percent on portion size, worse on calorie-dense foods where a small weight error becomes a large calorie error. Lose It's "Snap It" photo-logging feature, tested independently, correctly identified 64 out of 100 dishes. That is a real, checkable number, and it still means more than a third of dishes were missed outright, before portion size even enters the picture.

Neither of those apps is unusually bad. They are simply two of the only ones with a number attached to the claim at all.

The trick in the headline percentage

Most "95% accurate" marketing claims conflate two different things: did the model correctly say what the food is, and did it correctly estimate how much of it there is. Identification accuracy in this category is genuinely high and has been for years. Portion accuracy, the part that actually determines whether the logged calorie count is right, is a much harder problem, and it is the one almost never reported separately. A claim that says "accurate" without specifying which of those two it means is not describing a measurement. It is describing a marketing decision.

The other pattern worth knowing: several apps in this category publish head-to-head portion-error comparisons against named competitors with no independent source, no published methodology, and no benchmark dataset behind the number. Treat any specific accuracy percentage in this category, including a competitor's own claim about another competitor, as unverified until there is a citation attached.

Why Arcal does not publish a headline accuracy number

This is also why you will not find a big accuracy percentage on Arcal's own marketing. Portion estimation from a single photo is a hard, unsolved problem for every app in this category, our own included, and putting an uncited number on a page does not change that. What we do instead: publish pricing before you download, keep the plan the same at every tier, and build the camera system around giving the model better spatial context rather than around a number to put on a landing page. See how Arcal approaches AI food scanning or join the waitlist.

Sources: Cal AI vs MyFitnessPal 2026 — Welling, Best Calorie Tracking Apps 2026 — Every Calorie Tracking App Compared