Asking an AI to plan a week's groceries from a budget

One of the genuinely useful things AI assistants do is meal planning. Here's how to prompt them well, especially when you have constraints.

The bad prompt

"Plan my week of meals."

You'll get something generic, something expensive, and something that requires ingredients you don't have.

The good prompt

"I have $75/week for groceries for 1 person. I want to hit ~2,000 cal/day, 130g protein, 30g+ fiber. I have a basic kitchen (no instant pot, no air fryer). I cook 4 nights a week and eat leftovers the other 3 days. I don't eat pork. I have soy sauce, olive oil, and basic spices already. Give me: a shopping list with prices, a 7-day plan, and a Sunday prep checklist (60 min max)."

The structure that does the work:

1. Budget. Otherwise you'll get fancy ingredients.
2. Macros. Otherwise it'll be carb-heavy by default.
3. Equipment constraints. Otherwise it'll suggest sous vide.
4. Cooking frequency. Otherwise it'll give you 7 distinct meals you'll never make.
5. Restrictions. Be specific — "no pork" is clearer than "I don't eat much pork."
6. Pantry assumptions. Otherwise the list will redundantly include salt.
7. Format requirements. "Shopping list, plan, Sunday prep" — much more usable than free-form text.

What the output should include

A useful meal plan from a good AI prompt has:

• A shopping list with rough prices. So you can actually buy it.
• A 7-day overview. With a clear pattern (which nights are cooking nights, which are leftover nights).
• Recipes for the cooking nights. With portion sizes that match your macro target.
• A prep checklist. What to do on Sunday so the week works.
• A "what could go wrong" section. This is what separates good prompts from great ones — ask for it.

The follow-up turn

Once you have an initial plan, the second turn is:

"Of those meals, which are the most freezer-stable? And can you give me a back-up 'low-energy night' option for the days I won't want to cook?"

A good AI will answer this well. The contingency planning is what makes the plan actually survive contact with reality.

What this has to do with us

Our own app's meal planning feature uses similar principles internally — budget, macros, restrictions, equipment, prep time, contingencies. We're not better than ChatGPT at meal planning per se; we're just integrated with your tracking, so the meal plan you accept becomes the food log you fill out.

For users who use a general AI for meal planning, that's fine — copy the plan into our app's "save as recipe" feature and you're done.

A meta-observation

AI assistants have made meal planning, which used to be a real chore, into a 2-minute task. The bottleneck is no longer "what should I eat this week"; it's "what should I prompt the AI with this week."

For people who eat well at home, this is a quiet superpower. For people who order DoorDash four nights a week because "I don't know what to make," this is a fixable problem.

The meal plan is the diet. The cooking is just rendering.