How to track your macros with ChatGPT (and where it breaks)

How do I set up ChatGPT to estimate macros?

Give the assistant a format instruction once at the top of the chat (or in its memory or custom instructions). Ask for six values, a per-ingredient breakdown, the assumptions it made, and a calorie range. That single rule turns a vague "roughly 600 calories" into something you can check.

From now on, when I tell you what I ate, reply with:
1. Calories, protein, carbs, fat, fiber and sugar for the whole meal.
2. One row per ingredient with quantity and the same six values.
3. A short list of the assumptions you made (portion sizes, cooking method, brands).
4. A plus-or-minus range on the calories.
Do not add advice. Ask me one question if a quantity is missing.

How do I describe a meal so the estimate is useful?

Name every food and every quantity you know. The assistant fills the gaps with assumptions; the fewer gaps, the tighter the range. Cooking method and fats matter most: "grilled chicken" and "chicken fried in oil" differ by a few hundred calories.

Lunch: 180 g grilled chicken breast, about a cup of cooked white rice,
half an avocado, a handful of cherry tomatoes, olive oil drizzle.

A good answer lists the rice at roughly 200 kcal, the avocado at roughly 120, the oil as an assumption ("drizzle counted as 1 tsp"), and a calorie range such as 620 ±60.

How do I correct an estimate in the chat?

Reply with one sentence about the assumption that was wrong and ask for updated totals: "the oil was closer to a tablespoon" or "make the rice a double portion". Assistants handle this well inside one conversation. Keep an eye on the assumptions list every time; it is where the error hides.

The oil was about a tablespoon and the chicken was 250 g. Update the totals.

Can ChatGPT keep a running daily total?

Inside one conversation, sometimes. Ask "add this to today's total" after each meal and the assistant will sum what it remembers. It is unreliable across long chats and it stops working the moment you open a new conversation, because the previous meals are not there. This is the first place a raw chat log breaks.

Add this meal to today's running total and show me the day so far.

Where does a ChatGPT food log break?

Four places, all structural rather than a matter of prompting:

• No memory across days. A new chat does not know yesterday's meals. Weekly averages, the number that actually matters for a target, never exist.
• No totals you can trust. Summing a day depends on the assistant re-reading the whole chat; long conversations drop meals.
• Drifting portions. The same "bowl of oatmeal" gets a different estimate on different days, and nothing records what was assumed last time.
• No correction trail. When you fix an estimate, the old number is gone and there is no editable history to check later.

How do I turn the chat into a real diary?

Connect the assistant to a diary that stores every meal. With Macropass, you connect Claude, ChatGPT, or any custom MCP client; describe what you ate in the chat and your assistant logs calories, protein, carbs, fat, fiber, sugar and the full ingredient breakdown into your Macropass diary. It can also read today's totals, search and edit your meal history, and log weigh-ins. The chat keeps doing the estimating; Macropass keeps the running totals, weekly averages, editable history and trends. Setup takes about three minutes: connect Claude or ChatGPT to Macropass.

Meals you log inside the app get the Macropass draft on top: the words you used, every ingredient, the stated assumptions, a calorie band, and a one-sentence correction before anything counts.

When is a database tracker the better tool?

If you weigh your food and verify against labels, use MacroFactor or Cronometer. If you quit trackers because logging is a chore, or you already ask your AI what is in your food, Macropass is built for you. An AI estimate, in a chat or in Macropass, is a stated-assumption estimate with a range, not a label-verified number.

Questions about tracking macros with an AI chat