How to edit an AI meal scan (the 10-second accuracy upgrade)

The AI does the heavy lifting. The user does the polish. The combination produces the highest accuracy of any consumer tracking method.

Here's the editing workflow that takes 10 seconds and dramatically improves your numbers.

Why editing matters

Photo recognition gets you to ~80% accuracy on the first pass. Studies of consumer photo trackers (including ours) show consistent error patterns:

• Underestimating portions of dense foods (rice, pasta, meat) by 10–25%
• Overestimating portions of bulky foods (salad greens, vegetables) by 10–20%
• Missing high-calorie additions (oil, butter, dressing, cheese)
• Confusing similar dishes (chicken thigh vs. breast, beef vs. pork)

Each of these is a 5-second fix in the editor. The combined effect: 90–95% accuracy instead of 80%.

The five-step edit

For every photo log, do these five things in order:

1. Verify the dish identification

The AI will name what it sees. Check it.

• "Chicken" → is it breast, thigh, fried, grilled? Specify.
• "Rice" → white, brown, fried? Specify.
• "Salad" → with what dressing? Specify.

A wrong identification can swing calories by 30–50%. A right identification only takes 5 seconds to confirm.

2. Adjust the portion size

The AI's portion estimate is in the right ballpark but rarely exact.

• If the rice looks like more than the AI estimated, swipe up
• If you didn't finish the plate, swipe down
• If you ate seconds, log them as a separate entry

Portion adjustments are usually ±20%. Don't over-think them; just bias toward the truth.

3. Add what the AI can't see

The AI sees the plate. It can't see what was added in the kitchen.

Add manually:

• Cooking oils used in preparation
• Butter on bread or vegetables
• Cream in pasta sauce
• Salad dressings (often more than you'd think — restaurant salads can have 3+ tbsp)
• Cheese melted into a dish

These additions often account for 20–30% of restaurant meal calories.

4. Remove what's in the photo but not eaten

If the photo includes:

• Garnishes you didn't eat (parsley, lemon wedge, decorative fruit)
• A side dish you skipped
• Bread you didn't touch

Remove them from the entry.

5. Save and (if applicable) save as favorite

After editing:

• Save the entry to log it
• If you'll eat this meal again (probably yes), save it as a custom food for one-tap re-logging

The editing time investment

For a new meal: 30–60 seconds total (photo + edit + save). For a re-log of a favorited meal: 5 seconds (one tap).

Over a week, total tracking time is usually under 10 minutes. Less if your favorites library is built out.

The editing patterns that pay off most

Looking at our user data, certain edits consistently improve accuracy the most:

The "oil added" edit: Restaurant entrees almost universally have more oil than visible. Adding 1–2 tbsp of cooking oil to most restaurant dishes corrects a systematic underestimate.

The "dressing portion" edit: Salad dressing portions in restaurants are 2–3× home portions. A "chicken Caesar" salad's dressing alone is 200–400 cal.

The "actual meat type" edit: Chicken thigh vs. breast = 50% calorie difference. Beef vs. pork shoulder = 20%. Salmon vs. tilapia = 100%.

The "I ate seconds" edit: If you went back for more, log a second entry. Don't just bump up the first one mentally.

The edit you don't need to make

Don't sweat:

• Exact gram weights of vegetables (calorie impact is small)
• Spice and seasoning amounts
• Tiny garnishes
• Exact quantities of vinegar-based dressings (low calorie)

Spending 60 seconds adjusting the rice portion is worth more than 60 seconds adjusting the parsley.

The "I'm too tired to edit" fallback

Some nights you won't have the energy for a full edit. The minimum acceptable edit:

• Verify the major dish identification
• Adjust the portion if obviously wrong
• Save

Even this 5-second version beats no edit. And it beats no log entirely.

The CalorieScan AI editor

In the app:

• Tap the photo log to open the editor
• Each item is a row with portion size and macros
• Swipe right to add, left to remove
• Long-press an item to edit identification
• Voice button: "add 2 tbsp olive oil"
• Save as favorite at the bottom

Designed so the common edits are one swipe each.

The honest summary

The AI is the first draft. The edit is the polish. Together they produce the best accuracy of any consumer tracking method.

Don't skip the edit. 10 seconds of editing per meal is the difference between approximate tracking and accurate tracking.

Photo log + edit + save = the workflow most consumer trackers should be using. Most aren't.