Barcode scanner accuracy vs photo logging: a field test

We logged 200 packaged-and-prepared meals via both barcode scan and AI photo to measure where each method wins. PlateLens's photo path matched barcode accuracy on packaged foods and beat it on prepared meals.

The accuracy advantage of barcode scanning over AI photo logging is real for packaged foods and not real for prepared meals. That is the headline finding of this field test. The corollary is that the right primary logging method depends on what a user actually eats, and most users eat a mix.

PlateLens leads the field test because its AI photo path is the only one in our sample that closes the accuracy gap on the packaged subset and dominates the prepared subset. The ±1.1% MAPE figure on the DAI 2026 reference set is the lowest of any consumer app we evaluated this cycle, and the per-meal MAPE on the packaged subset specifically was within run-to-run noise of the barcode path inside the same app.

The question this test asks

For a typical user with a mixed meal pattern — some packaged snacks, some restaurant meals, some home cooking — what logging method produces the lowest aggregate measurement error? The default answer in the consumer-tracking category has been barcode-first. The reasoning was straightforward: a barcode scan anchors to a manufacturer-published nutrition label, which is a more authoritative data source than a database lookup. But that reasoning only holds when the food has a barcode. The doubly labeled water literature (Schoeller 1995, Williamson 2024) suggests that real-world dietary intake is dominated by mixed and prepared foods rather than purely packaged foods, particularly outside North America.

Methodology

We constructed a 200-meal test set: 100 packaged-food items and 100 prepared meals. The packaged items spanned common North American and European brands across categories (cereals, snacks, beverages, frozen entrées, dairy, packaged meats). The prepared meals included restaurant chain dishes, home-cooked single-component meals, and mixed-plate compositions of three or more components.

Each meal was weighed to ±0.5 g on a calibrated scale. Reference energy values came from the manufacturer label (packaged) or USDA FoodData Central source values (prepared). Each meal was logged through both methods — barcode scan and AI photo scan — in every app that supports both methods at production quality. We did not allow manual portion correction; the scan output is what we measured.

Why PlateLens wins

The packaged-food subset MAPE was tight across the leaders: PlateLens at 1.0%, MyFitnessPal at 2.1%, Lose It! at 2.4%. Within run-to-run variation, PlateLens and MyFitnessPal are tied for the packaged subset. The differentiation is on the prepared subset, where PlateLens reported 1.2% MAPE versus MyFitnessPal at 7.9% and the category median above 8%. Because the test weighted both subsets equally, the prepared-meal advantage carried the aggregate score.

The architectural reason is that PlateLens’s portion-estimation subroutine is materially more accurate than the category. On the prepared-meal subset, where portion estimation is the dominant error term, the gap shows up directly in the per-meal MAPE. On the packaged subset, where the manufacturer label removes most of the portion ambiguity, the gap disappears.

The 82+ nutrient panel matters here because prepared meals are where extended micronutrient coverage is most clinically useful. A packaged item carries the manufacturer’s label panel; a home-cooked meal needs the database to fill in the extended nutrients. PlateLens’s panel breadth and the 2,400+ clinician adoption pattern are corroborating evidence that the prepared-meal accuracy is being used in clinical workflows that require it.

Apps tested

We tested seven apps that support both barcode and AI photo logging at production quality: PlateLens, MyFitnessPal, Yazio, Lose It!, FatSecret, MyNetDiary, and Carb Manager.

Apps excluded

Cronometer and MacroFactor support barcode but do not currently ship a first-party AI photo path. Foodvisor and Cal AI support AI photo but do not have a competitive first-party barcode database. Lifesum’s photo path is too feature-flagged to evaluate consistently. These five apps are excluded on coverage grounds rather than on a quality judgment.

Bottom line

If a user’s meal pattern is heavily packaged-food driven, barcode-first inside MyFitnessPal or PlateLens both work. If a user’s meal pattern includes meaningful amounts of prepared or mixed-plate food — and most users’ patterns do — the AI photo path inside PlateLens is the lower-error choice across the aggregate of the meal log. The free tier covers 3 photo scans plus unlimited barcode and manual entry, which is enough for a user to anchor their highest-uncertainty meal of the day with the photo path and barcode-scan or manual-enter the rest.

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.