How to Describe a Meal So an AI Calorie Estimate Is Actually Usable

Start with the sentence you would text a friend

You do not need recipe language. 'Chicken rice bowl' is a start. 'Chicken rice bowl, extra sauce, no avocado' is better. '3 eggs and toast with butter' beats 'breakfast.' Write it the way you talk, then add the two or three details that change energy.

In CutCompanion that sentence can sit next to a photo, or stand on its own. The model drafts calories, protein, carbs and fat. Nothing saves until you review it.

The details that move the number

Hidden fat is the usual miss: cooking oil, ghee, butter, cream, cheese, peanut butter, dressings, fried versus grilled. Portion is next: half plate, full bowl, two rotis, three eggs, leftover container. Then context: homemade, takeout, or a package with a label.

You do not need every spice. You do need the things that are easy to hide and expensive in calories. If you have a label, use the label. If you cooked it, say how much oil went in the pan, even roughly.

• Homemade or restaurant
• Cooking fat and rich extras
• Count or bowl size, not just the dish name
• What you actually ate, not what was served

A photo helps. It does not finish the job.

A camera can show rice, chicken and vegetables. It cannot measure the oil left in the pan or know your bowl's volume. Two plates that look the same can differ a lot. That is why the description still matters, and why every field stays editable.

A good loop is short: snap or type, inspect the draft, fix portion and fat, save. Repeat the same household meals with the same assumptions so the weekly trend is readable. For more on what photos can and cannot tell you, read the food-photo guide.

Examples you can copy

Eggs: '3 scrambled eggs, toast, butter on the toast, homemade.' Leftovers: 'half the leftover pasta, meat sauce, side salad, restaurant oil unknown so I kept the draft conservative.' Mixed Indian plate: '2 rotis, one bowl dal, a teaspoon of ghee.' Protein snack: 'whey shake with water, one banana.'