Crowdsourced Database
Crowdsourced Database, This type of database relies on contributions from users for entries related to food, rather than being compiled by paid editorial teams or official organizations like the USDA. Although crowdsourced databases boast extensive coverage with millions of entries, they often suffer from inconsistent quality and higher error rates in user-reported nutritional information.
What is a crowdsourced database?
A crowdsourced food database is one that primarily gathers its entries from users of the app. A notable example is MyFitnessPal, which created a database encompassing over fifteen million entries by enabling users to add new foods and replicate entries made by others. Other major consumer tracking services such as Cronometer and Lose It employ a hybrid approach, combining a carefully curated selection of verified foods with a much larger array of user-submitted entries.
Crowdsourcing addresses a coverage gap that curated databases cannot fill. While the USDA’s FoodData Central contains high-quality data for many whole foods and branded items, it lacks specific entries for unique items like “Joe’s Coffee Shop blueberry muffin, downtown Boston.” However, a crowdsourced database, with a sufficient user base, can fill that gap. The downside is that quality can suffer; when users submit calorie data, no authoritative body verifies it against the manufacturer's information. The individual may have interpreted the label correctly, or they might have guessed the portion size, or even mistakenly input values from a different brand.
How is it measured in our testing?
The database quality evaluation criteria at Independent Reviews (20% of the total 100-point scale) specifically assess how well an app differentiates between crowdsourced and verified data. This includes four sub-criteria:
- Coverage. A search panel consisting of 50 items, including grocery SKUs, items from restaurant chains, and regional specialties. Coverage assesses whether the app contains the food item at all.
- Verification. We sample 20 entries from each app to verify if the displayed values align with manufacturer labels or the USDA FoodData Central entries. Apps that indicate verification status (allowing users to know if the entry is from a verified source or another user) receive higher scores.
- Freshness. Restaurant menus change regularly. We examine whether the entries for chain restaurants in the app reflect the current menu offerings.
- Noise resilience. We submit ambiguous search terms (like “pizza”) and evaluate how effectively the app presents reliable entries compared to low-quality user submissions.
Applications that blend user-submitted and verified entries without making a clear distinction in the user interface are subject to penalties. Those that default to displaying user-submitted entries without verification flags face more significant repercussions.
Why it matters in calorie tracking apps
For users, the trade-offs related to crowdsourced databases can pose a daily risk: the application you rely on to log your lunch might provide a calorie count that was entered by an unknown individual on the internet three years ago, without any verification. In MyFitnessPal’s database, for instance, a single food item may have ten or twenty different user-generated entries with calorie amounts differing by 30% or more. Users who select the first result without verifying can introduce systematic inaccuracies into their tracking.
To mitigate this, current apps suggest the following: (1) always prioritize entries marked as verified or sourced from the USDA, if the app provides this option; (2) for branded products, use the barcode scanner instead of searching by name (the barcode search directs to the manufacturer’s verified label); (3) confirm the calorie content against the product packaging or USDA FoodData Central if uncertain. In the context of a verified food database, the trust assessment changes, so refer to that entry. Also, check our free-tier entry for insights on whether the verified database requires a subscription.