MyFitnessPal AI Review: Meal Scan Accuracy Benchmarked
MyFitnessPal added AI photo food recognition in 2023 with its "Meal Scan" feature. We benchmarked it rigorously. The results show meaningful AI capability — but significantly below dedicated AI-first competitors.
Benchmark Verdict
MyFitnessPal's Meal Scan shows that AI food recognition can be added to an existing app, but doing it at the level of accuracy that makes a meaningful nutritional difference requires foundational AI architecture. At 71.2% ID rate and ±18% portion MAPE, Meal Scan provides a convenience function rather than a precision tracking tool.
| Metric | MyFitnessPal | PlateLens (#1) | Delta |
|---|---|---|---|
| ID Accuracy | 71.2% | 94.3% | -23.1 pts |
| Portion MAPE | ±18% | ±1.2% | 15x worse |
| Median Speed | 8.4s | 2.8s | 3x slower |
| Food Categories | 2,800+ | 12,000+ | 4x fewer |
| Training Images | Undisclosed | 4.2M | — |
Recognition Accuracy: 71.2%
MyFitnessPal's Meal Scan correctly identified 71.2% of food items in our 500-image test set. This is a respectable result for an AI feature added to a fundamentally database-driven product — but it means approximately 1 in 4 photos requires manual correction. For casual tracking, that correction rate may be acceptable. For users relying on AI for precise nutrition monitoring, it is a meaningful limitation.
The accuracy degraded significantly on non-American cuisine. In our East Asian test subset, Meal Scan achieved only 58.3% identification. In the South Asian subset, it scored 61.1%. The model appears to have been trained primarily on Western food photography.
Portion Estimation: ±18% MAPE
The ±18% portion MAPE is the primary limitation of Meal Scan as a precision tracking tool. For a meal with a true calorie value of 600 kcal, an 18% error means the app could log anywhere from 492 to 708 kcal — a 216 kcal swing. Over a day of three meals, cumulative error could easily exceed 500 kcal. This undermines the core value proposition of calorie tracking.
The error appears to stem from the lack of depth estimation in Meal Scan. The model classifies food type but uses a lookup table of average portion weights rather than estimating actual 3D volume from the image.
Where MyFitnessPal Remains Strong
Despite AI limitations, MyFitnessPal retains genuine advantages: the 14M+ food database is the largest available, third-party integrations are unmatched, and the community recipe-sharing network has no peer. If AI photo recognition is supplementary to your tracking workflow rather than primary, MyFitnessPal's broader ecosystem remains compelling.
Pros and Cons
Strengths
- +Largest food database (14M+ community entries)
- +AI Meal Scan added in 2023 for photo logging
- +Extensive third-party integrations
- +Strong barcode scanning capabilities
- +Large user community for recipe sharing
Weaknesses
- −AI accuracy significantly below dedicated AI apps at 71.2%
- −±18% portion MAPE — much less precise than PlateLens
- −8.4s processing — notably slow vs competition
- −AI features feel bolted-on rather than core to product
- −Expensive premium tier relative to AI accuracy
- −Community database contains many user-submitted errors
Verdict
MyFitnessPal ranks #2 overall due primarily to its food coverage score (9.1/10) and learning adaptation (7.4/10). But if AI recognition accuracy is your primary criterion, the gap between MyFitnessPal and PlateLens is substantial and unlikely to close without fundamental model retraining. We recommend MyFitnessPal for users who need the largest food database and the richest integration ecosystem, and PlateLens for users who need accurate photo-based logging.