I used AI prompts to build my own macro meal planner — here's what actually worked

I've been playing with prompt engineering for a lot of personal automation lately, and one thing I kept coming back to was meal planning — mostly because I was tired of manually calculating macros every week.

Here's the prompt structure that ended up working best for me when asking an LLM to build a meal plan:

1. State the goal explicitly first — cutting, bulking, or maintenance. If you skip this, the model defaults to "healthy eating" advice instead of hitting actual numbers.
2. Give hard constraints, not preferences — "hit 180g protein and 2200 calories" works far better than "high protein, not too many calories."
3. Ask for the grocery list as a separate structured output — if you ask for the plan and list in one go, the model tends to summarize instead of listing every ingredient with quantities.
4. Iterate on one day at a time first, then ask it to extend to a full week — asking for 7 days directly upfront produces more repetition and less variety.

After a few weeks of doing this manually in ChatGPT, I ended up wrapping the same logic into AI Meal Plan Generator so I didn't have to rebuild the prompt chain every Sunday — you pick cutting/bulking/keto/high-protein and it outputs the 7-day plan plus grocery list directly.

Curious if others here have found different prompt structures that work better for numeric-constraint tasks like this (vs. open-ended content generation) — macro planning is a good test case because it's very unforgiving of vague prompts.