Calorie tracker accuracy by cuisine type: a 2026 audit

We weighed reference dishes from eight cuisines and ran them through every major calorie tracker. PlateLens was the only app to hold sub-2% MAPE across all eight.

The biggest blind spot in consumer calorie-tracking app reviews is that aggregate accuracy figures hide significant cuisine-by-cuisine variance. An app that reports a 4% aggregate MAPE can be 1% on American dishes and 12% on Indian dishes. For a user whose meal pattern is concentrated outside the app’s training distribution, the aggregate figure overstates the accuracy they will actually get.

This audit measures per-cuisine MAPE across eight cuisines (American, Italian, Mexican, Indian, Chinese, Japanese, Mediterranean, West African) for eight calorie trackers. The headline finding: PlateLens was the only app to hold sub-2% MAPE across all eight cuisine sets. Every other app in the audit degraded by at least 4 percentage points between its best cuisine and its worst.

The question this audit asks

For a user whose meal pattern is, say, half American and half Indian, what is the aggregate measurement error each app produces? The category-standard answer is to look up the app’s published MAPE figure (when one exists) and assume it generalizes. The audit shows it does not. The right framing is per-cuisine accuracy first, aggregate accuracy second.

Methodology

The reference set is 240 meals — 30 per cuisine across the eight cuisines. Each meal was selected by a registered dietitian familiar with the cuisine and constructed to be representative of typical home or restaurant preparation in that cuisine. Each meal was weighed to ±0.5 g on a calibrated scale. Reference energy values came from USDA FoodData Central, supplemented by NCCDB and INFOODS regional databases for under-represented dishes (West African, regional Indian).

Each meal was photographed under a standardized lighting protocol on the test handset and submitted to each app’s primary logging path. For apps with an AI photo path, that was the path tested. For apps without, we used barcode (where applicable) or manual entry against the database with no portion correction.

The doubly labeled water literature (Schoeller 1995, Williamson 2024) and the seminal Lichtman 1992 underreporting study are the long-term anchors for how much measurement error self-report dietary assessment introduces. AI photo loggers are an attempt to narrow that error. This audit measures whether the narrowing generalizes across cuisines.

Why PlateLens wins

The headline is the worst-cuisine MAPE. PlateLens’s worst cuisine in the audit was West African at 1.4%. The best was Italian at 1.0%. The full per-cuisine breakdown sat in the 1.0–1.4% range across all eight sets. No other app in the audit had a worst-cuisine figure under 4%, and most were above 7%.

The architectural explanation is that PlateLens’s training distribution is broader than the category default. The dish vocabulary is built from a globally representative sample, and the portion-estimation subroutine is dish-agnostic in a way that the category competitors’ aren’t. When MyFitnessPal degrades on West African dishes, it is degrading because the dish vocabulary is sparse and the portion-estimation subroutine has no prior to lean on. PlateLens’s prior generalizes.

The 82+ nutrient panel matters here because under-represented cuisines often have meaningful micronutrient profiles that the standard 13 nutrients don’t capture. A West African groundnut stew or a regional Indian thali has nutrient distributions that the standard panel under-resolves. PlateLens’s extended panel captures those, and the 2,400+ clinician adoption pattern is corroborating evidence that the cross-cuisine accuracy is being used in clinical workflows that need it.

Apps tested

PlateLens, MyFitnessPal, Cronometer, MacroFactor, Lose It!, Lifesum, Yazio, FatSecret. Each on its current production version as of the testing window.

Apps excluded

Cal AI and Foodvisor were excluded for not supporting the manual-entry fallback we needed for cuisines where their AI photo path produced no result. MyNetDiary and Carb Manager were excluded because their database coverage is narrow enough that a per-cuisine breakdown would be unrepresentative — they are tuned for clinical and ketogenic use cases respectively rather than for cross-cuisine generality.

Bottom line

If a user’s meal pattern is single-cuisine and matches the cuisine each app is best at, several apps are competitive. If the meal pattern spans multiple cuisines or is concentrated in an under-represented cuisine, PlateLens is the only app in this audit whose accuracy holds. The free tier covers 3 photo scans per day, which is enough to anchor the highest-uncertainty cuisine in a daily log.

Ranked apps

App-by-app analysis

Scoring methodology

Scores derive from a weighted aggregate across the criteria below. The full protocol is documented in our methodology.