How to Optimize AI Tracking

AI Tracking makes logging food fast. Like any AI tool, the quality of what you get out depends partly on what you put in. A few small habits make a real difference in how accurately meals are recognized and how closely the numbers match what you actually ate.

This guide covers how AI Tracking works under the hood, what makes results more or less accurate, and what to do when a number looks off.

Why AI estimates can vary

AI Tracking works by identifying food from a photo, a voice note, or a typed description, then estimating nutrition based on what it recognizes. Estimation is the key word. The AI is reading visual or written information and inferring the most likely food, portion, and preparation method from it, the same way a person glancing at a plate would estimate calories rather than weigh each ingredient.

A few things affect how close that estimate lands:

• How much information the AI has to work with. A blurry photo, a cropped plate, or a vague description ("a salad") gives the AI less to work with, so it fills in gaps with its best guess.
• How the food is prepared. The same ingredient can carry very different nutrition values depending on brand, recipe, and cooking method. Grilled and fried chicken breast are not nutritionally the same, even though they look similar in a photo.
• Portion size. Without a clear scale reference, judging exact portion size from a photo is inherently harder than from a description with a stated weight.

None of this means a number is wrong. It means the AI made its best estimate with the information available, and that estimate can usually be tightened with a few adjustments, or corrected on the spot.

Best practices by input method

Voice input

• Speak clearly and at a normal pace, in a quiet environment when possible. Background noise is one of the most common causes of misheard entries.
• State quantity and preparation method, not just the food name. "150 grams of grilled salmon with steamed broccoli" gives the AI far more to work with than "salmon and broccoli."

Text input

• Log one food or meal per message rather than combining several unrelated items into one long description.
• Include quantities and units when known. "2 slices of whole wheat toast" outperforms "some toast."

Image input

• Use natural daylight or a well-lit room. Phone flashlight and dim restaurant lighting distort color and make food harder to classify correctly.
• Keep all edges of the plate in frame. A cropped plate removes the visual reference the AI uses to judge portion size.
• For layered or stacked foods like sandwiches, burgers, and bowls, shoot from a 45-degree angle showing both the top and side. A straight top-down shot hides what's inside.
• Add a fork, hand, or other familiar object for scale when a portion is unusually shaped or sized.
• Photograph mixed dishes, like salads, before stirring, so each component stays identifiable.

Barcode scanning

• Center the barcode, keep it in focus, and minimize glare.

Quick Track

• Use Quick Track for fast manual entries when the exact values are already known. It skips estimation entirely.

Seeing a number that looks off? Here's what to check first:

A value that feels wrong is usually explained by one of a few common causes, and most take seconds to fix.

1. Check what the AI actually logged. Open the entry and confirm the food, portion, and preparation method it identified match what was eaten. A misread ingredient or portion is the most common cause of an unexpected number, and it's a one-tap fix.
2. Compare against the specific product or recipe. Nutrition values vary by brand, recipe, and cooking method. A generic match ("pasta with tomato sauce") will differ from the exact product or homemade recipe used, even when the dish looks the same.
3. Review the photo or description that was submitted. A cropped plate, low light, or a vague description gives the AI less to work with. Retaking a photo with better framing and lighting, or adding quantity and preparation detail, usually resolves it.
4. Adjust the entry manually. Every logged item can be corrected after the fact. This takes a few seconds and makes that entry accurate immediately.
5. Correcting entries helps over time. Manual corrections refine future recognition for recurring meals, so the small effort compounds.

If a number still looks significantly wrong after checking the above, and the same issue keeps recurring across different meals or days, reaching out to support with a couple of example entries makes it easy to look into further.

The short version

AI Tracking is built to make logging effortless, not to replace judgment entirely. Clear photos, specific descriptions, and a quick check after logging go a long way toward accurate results, and every entry stays fully editable if something doesn't look right.

Updated on: 23/09/2026