How AI Calorie Estimates Work
What happens when you tap AI Estimate — why the numbers lean conservative on purpose, what data is sent, and where estimates should not be trusted.
Updated August 2026
Quick answer
Type a description, tap AI Estimate, and a number comes back in seconds. Your body metrics personalise it, so the same meal returns different numbers for different people. The estimates lean conservative by design — slightly high on food, slightly low on exercise — so the running total errs against you rather than flattering you. Pro includes 20 estimates per day.
The underlying math
Loggnit tracks net calories: food and drink add, exercise subtracts, and the daily figure is the balance. You can always type a number yourself — the estimate is a shortcut for when you do not have one, not a required step.
Why the numbers lean conservative
| Side of the ledger | Bias | Effect on your total |
|---|---|---|
| Food and drink | Estimates slightly high | Assumes you ate a bit more |
| Exercise | Estimates slightly low | Assumes you burned a bit less |
Both biases push the same direction — a net figure that is pessimistic rather than encouraging.
This is deliberate, and it is a correction for how estimation error normally compounds. Calorie figures carry real uncertainty in both directions; when a tool resolves that uncertainty optimistically on both sides, the errors stack, and the daily total drifts steadily flattering. Erring the other way keeps the number closer to reality even when any single estimate is off.
The practical consequence: treat the total as a consistent yardstick for comparing your own days, not as a measurement. Its value is in the trend line, not the individual figure.
What personalises the estimate
Body metrics feed the math, which is why a sedentary office worker and a competitive cyclist get different numbers from an identical description. For exercise especially, body mass changes the answer substantially — a generic figure would be wrong for most people most of the time.
For lifts, the session's shape carries through too: total volume across your sets feeds the estimate, so a heavy five-set session and a light one do not resolve to the same number.
What gets sent, exactly
An estimate is generated remotely, so something has to leave the device. What that is:
- The activity description you typed
- Its category
- Any notes and metric values on the entry
- Your body metrics
Nothing is retained after the response, and nothing trains a model. Your entry history is not sent — each request carries only that one entry. Everything else in Loggnit stays on the device and syncs through your private iCloud.
Where estimates are weakest
- Restaurant meals — portion sizes and cooking fat vary enormously and are invisible in a description
- Homemade dishes — “stew” covers a very wide range
- Anything where the description omits quantity — “pasta” carries far less information than “two cups of pasta with pesto”
- Unusual activities the description does not pin down precisely
The pattern is consistent: the estimate is only as specific as the sentence. A few extra words about quantity or preparation improves it more than anything else you can do.
Common questions
How accurate are the estimates?
Accurate enough to compare your own days, not accurate enough to treat as measurement. Calorie estimation carries real uncertainty even from laboratory data, and a text description carries less information than a weighed portion. The deliberate conservative bias is there because of that uncertainty, not in spite of it.
Do I need Pro to track calories?
No. Typing calorie counts yourself, unlimited entries, all 16 categories, charts, Apple Health sync, and iCloud sync are free. Pro adds the AI estimates, body metrics, custom categories, and Copy Day for a one-time $7.99.
What happens after 20 estimates in a day?
You can keep logging normally and enter counts yourself — the cap applies to the AI estimates, not to logging. It resets the next day.
Is my food diary used to train AI?
No. Each request sends only the single entry being estimated plus your body metrics, nothing is retained afterwards, and nothing trains a model. Your history is never part of the request.
Related guides
Get Loggnit
AI Calorie & Life Log · Free on the App Store.
