Benchmark · Benchmark protocol
Auditing food database completeness and barcode coverage
Our fixed-list method for measuring how completely each app's database covers branded, restaurant, fresh and international foods — and how often it falls back to a vague entry.
Key takeaway
Database size and database quality are different things: the largest databases had the most coverage but also the most inconsistent entries, while AI-native apps paired smaller, cleaner data with better international coverage.
Why database coverage decides everything else
Even a perfect AI estimate has to resolve to a food record. If the database can’t represent what you ate — or only offers a vague generic — your log is wrong before you start. So we audit coverage directly rather than trusting headline “millions of foods” claims.
The fixed audit list
We run every app against the same list of several hundred items, grouped so we can see where each app is strong or weak:
- National grocery and packaged brands (via barcode)
- Fast-food and sit-down restaurant menu items
- Fresh produce and whole foods
- International and region-specific dishes (including Asian, Latin and Middle-Eastern foods)
- Common unlabeled and homemade items
What we record
For each item we note whether the app returns a correct, complete entry, a partial or generic fallback, or nothing usable. For barcodes, we scan a standard set of packaged products and record hit rate. We also flag when an app silently drops nutrients like fiber or sodium from an otherwise-correct entry.
What we found
The biggest databases had the broadest raw coverage, especially for barcodes — but also the most inconsistent, user-generated entries, so users must vet matches. AI-native apps tended to carry smaller but cleaner datasets and handled international and unlabeled foods noticeably better, falling back to vague entries less often. Coverage of non-Western foods remains the clearest dividing line in the category.
Limitations
A fixed list is a sample, not the universe of foods. We refresh the list periodically and weight it toward the foods our readers actually eat, but local availability and regional menus mean your mileage will vary.