Calorie app does not recognise food: why that happens

The three most common causes

1. A gap in the database. Regional brands, supermarket own-labels and bakery items are often missing. Apps with community entries have wider coverage but less reliable individual values.

2. The barcode is not on file. A scan only finds what someone entered before. New products or changed packaging come up empty.

3. The photo AI misjudges the amount. This is the biggest and least understood cause. An AI usually recognises what is on the plate quite reliably — but not how much. One extra spoon of oil is 90 calories, and no photo shows it.

What actually helps

• Weigh once, then estimate. Weigh your usual portions a handful of times and your estimates get surprisingly accurate. That beats any app.
• Create your own entries. For meals you eat often, one carefully entered recipe pays for itself.
• Correct rather than discard. With AI estimates the number is a starting point, not a measurement — adjust and move on.
• Consistency over precision. If you estimate the same way every day, the weight trend still tells you what you need to know.

How accurate does it need to be?

For a calorie deficit, being off by a few percent is fine. The body does not work in round numbers anyway, and water shifts in bodyweight are larger than most estimation errors. What matters more than the exact figure is using the same method for weeks.

If searching takes too long

Some apps therefore use description instead of database search: you write or say what you ate and correct the estimate if needed. GymLog AI works this way and captures training in the same chat.