AI Food Logging

AI food logging is the fast path from photo, text, or labels to structured Apple Health nutrition entries, with a verification step that keeps your record consistent. If you want the evidence view on how accurate these drafts are in practice, read How Accurate Is AI Food Logging?.

AI is for interpretation. If the food is already a known row, a repeated meal from yesterday, or a package with a clear label, Fuel may have a faster path that uses lookup or recall instead of asking AI to infer the meal from scratch.

01How AI logging works

AI logging is usually correct on the first pass when the input is unambiguous. Treat the draft as a proposal you verify before saving.

Fuel converts what you submit into an estimate of calories, macros, micronutrients, water, and caffeine, then you confirm the entry so the Health record stays stable across days.

The confirmation step exists for edge cases, not for busywork. Most edits happen when the input is low quality or ambiguous, such as a blurry photo, a partial label, or a description that leaves portion size unstated.

02When AI is not the fastest path

Use the path that has the strongest evidence for the meal.

Food Library lookup is not an AI estimate. It searches Fuel's offline corpus and logs the selected row. Smart Recall exact mode is also local, because Fuel rebuilds the previous meal from your saved history. Edited Smart Recall uses remote AI only for the requested change, after the app has already chosen yesterday's source meal.

03Photo based logging

Photo logging works best when the input contains portion cues.

1. Capture the full plate and any relevant packaging or labels.
2. Avoid extreme angles that hide volume and portion size.
3. Confirm the result and correct obvious misses before saving.

If you repeat the same meal often, correct it once and reuse it as a template so the next logs are consistent.

04Text based logging

Text logging is useful when you want speed with less camera work.

Fuel can handle vague prompts, but precision improves when quantity is explicit. Start with what you have and then refine the draft with feedback until it matches what you ate.

If the first pass is off, revise the text and run it again. A short feedback pass is faster than manual entry across a whole week.

05Feedback and corrections

Treat the first draft as a hypothesis. If something looks wrong, correct it before you save the entry.

Common corrections include portion size, missing ingredients, wrong preparation method, and label mismatches. The fastest feedback is specific and directional, since it tells the model what changed.

1. This was two servings, not one.
2. Add olive oil and the sauce on the side.
3. The protein was chicken thigh, not breast.
4. The rice portion was about half of what you estimated.
5. Use the nutrition label in the photo for macros and serving size.

Before and after examples

Photo log portion correction

Before

Chicken and rice bowl

After

The rice portion was half of what you estimated and add one tablespoon of olive oil used in cooking

Text log ingredient correction

Before

Greek yogurt with berries

After

It was 300g nonfat Greek yogurt, 100g berries, and 30g granola

Label anchored correction

Before

Protein bar

After

Use the nutrition label in the photo and set the entry to one full bar, not one serving

06Limits by plan

Fuel Free includes a limited number of AI logged meals per week. Fuel Pro removes the limit and unlocks the higher volume use case where AI logging becomes the default capture path.

See Free and Pro for the current plan limits.

07Privacy

AI features may process the content you submit to produce a structured entry. Use Privacy and Data to understand how Fuel is designed to handle health data and what choices you have.