Prompting an AI assistant to plan dinner from what's in your fridge

One of the most underrated uses of a general AI assistant is "I have these things in my fridge — what should I make?" Done well, this can replace a recipe search and a meal plan in 30 seconds. Done badly, you get an unhelpful list of cuisines and a recipe that requires fish sauce you don't have.

This is a short guide to prompting it well, plus how it pairs with a calorie-tracking app like ours.

The bad prompt

"I have chicken, spinach, rice. What should I make?"

You'll get a generic chicken-and-spinach-and-rice idea, possibly with ingredients you don't have, no portion sizing, no calorie awareness, no acknowledgment of what you're trying to accomplish nutritionally.

The better prompt

"I have these ingredients on hand: chicken thighs (about 1 lb), baby spinach (a 5oz container), white rice, garlic, olive oil, soy sauce, lemon, parmesan, eggs, and a half-block of feta. I want a 600-calorie dinner, around 45g of protein, and prep should take under 25 minutes. Suggest two options with rough macros and step-by-step instructions. Assume I have basic pantry items (salt, pepper, common spices)."

You'll get useful output. The structure that does the work:

1. Constraint your inventory. List actual amounts when you can.
2. Constraint your calories. Otherwise you'll get cream-and-butter-heavy "American restaurant style" output.
3. Constraint your time. This rules out braises, marinades, complex multi-step recipes.
4. Constraint your protein target. This is the single most useful nutrition number to give it.
5. Mention your pantry assumptions. Otherwise it'll suggest fish sauce.
6. Ask for two options. Comparison output is almost always better than single output.

The next-level prompt

"Same constraints. Also: I have 1,400 calories left in my day after my lunch, and I've already hit 80g of protein. I have a hard workout in the morning so I'd like 50–60g of carbs. My partner is vegetarian, so include a parallel adaptation that swaps the chicken for something."

This is what a good calorie tracker enables: you walk into a cooking decision with a budget and a set of targets, not a vague desire. The AI can plan against the budget if you tell it the budget.

How this fits with our app

CalorieScan AI has a "what should I cook?" feature in beta that does roughly this, with two advantages:

• It already knows your remaining calorie/macro budget for the day
• It knows what's commonly in your fridge from your past logs (and asks if you want to add/remove items)
• It logs the meal automatically when you confirm

For users who already use ChatGPT or Claude for cooking, the workflow is the same — we just removed the prompting overhead.

A working prompt template

Here's one you can paste into ChatGPT, Claude, or Gemini and adapt:

I want a meal idea using only what I have. Here's what I have: [list]. I need: [calories] calories, [protein] g protein, ready in [minutes] minutes. I [do/don't] eat [meat/dairy/gluten/etc]. Suggest 2 options. Include macros and a 6-line instruction set per option. Be specific about portions.

That prompt, refined to your situation, will outperform 90% of recipe searches for "what's for dinner."

A few prompts that don't work

• "Be creative." AI assistants interpret this as "use unusual ingredients" or "reach for fusion." You'll get a recipe with miso paste and you don't have miso paste.
• "Make it healthy." "Healthy" is undefined. Replace with specific macro/calorie constraints.
• "Make it fancy." This produces 12-step recipes. If you want fancy, ask explicitly: "I have 75 minutes and an oven."

Why we encourage this

People underestimate how much of their food choice is driven by indecision. "I don't know what to make" is the leading cause of "I'll just order DoorDash again." Eliminating decision friction with a 30-second AI prompt is one of the cheapest, highest-leverage habits for both budget and health.

The future of nutrition apps is fewer searches, more conversations.