How to meal plan with ChatGPT (what works, what breaks)

ChatGPT is genuinely good at three parts of meal planning: turning your constraints into a short list of dinners, suggesting substitutions when you are missing something, and rewriting a recipe for a different number of people. It is weak at the parts that decide whether a plan survives the week. It does not hold the recipes your household actually cooks, it does not know what is already in your kitchen, and the shopping list it writes stops being true the moment you are standing in the store.

So use it for the thinking, not as the place your plan lives. Below are the prompts that reliably produce something usable, the three failure points to expect, and the jobs where a chatbot beats any planning app outright.

Prompt with constraints, not with “make me a meal plan”

“Make me a 7 day meal plan” gets you the internet’s average week: a quinoa bowl, a salmon fillet, a stir fry, and a sheet-pan chicken. It is not wrong, it is just nobody’s actual week. The quality of the output tracks almost perfectly with how many constraints you put in.

Plan 5 weeknight dinners for 2 adults and a 6-year-old. Two of them need to be on the table in 25 minutes. One vegetarian. We do not eat fish. Normal supermarket ingredients only. For each dinner give me the name, one line on why it fits, the ingredients with quantities, and the hands-on time. No breakfasts or lunches.

Five things are doing the work there. The number of nights (five, not seven, because a real family week eats at least one dinner), who is eating, a time budget, the hard exclusions, and the output format. Ask for seven and you will cook four and bin the produce for the other three.

Then keep going. The first answer is a draft, and the repair prompts are where a chatbot earns its keep:

Swap Wednesday. We had something almost identical last week. Keep the same ingredients where you can.

Three of these need cilantro and nothing else uses it. Rework the week so the fresh ingredients overlap and nothing gets bought for a single dish.

That second prompt is the one most people never think to send, and it is the one that cuts the shop. Overlap is what stops half a bunch of herbs rotting in the drawer.

The four prompts worth saving

Start from the fridge. The single highest-value use of a chatbot in a kitchen. Type what is actually in there and let it do the combinatorics.

I have half a cabbage, three eggs, a jar of harissa, cooked rice, and a bag of carrots. Give me three dinners that use as much of this as possible, listing only what I would need to buy.

That is the fridge-first method with a machine doing the idea generation, and it is the version least likely to send you shopping for things you already own.

Scale a recipe and get the shopping delta. Paste the recipe in, then say what you already have.

Scale this for 5 people. Then give me the shopping list, but leave out olive oil, salt, garlic, and canned tomatoes, which I have.

Read a messy recipe. Paste in a wall of blog prose or a screenshot and ask for the ingredients and steps only. This is the thing large language models are honestly best at in cooking, and it is the same job an AI recipe importer does behind the scenes.

Ask for the boring version. Chatbots drift toward novelty because novelty reads as helpful. Tell it not to.

These are too ambitious for a Tuesday. Give me five dinners a tired person will actually cook, no more than 8 ingredients each, and at least two of them one pan.

Where a chatbot plan falls apart

Three specific places, and they are structural, not a prompting problem.

1. It does not hold the recipes you actually cook. Your household has maybe fifteen dinners in real rotation, and the value of a plan is mostly in choosing well among those, not in inventing new ones. ChatGPT can remember things you tell it, and you can keep a chat or a project where your recipes are pasted in, but that is a transcript, not a library. The ingredients are not structured, you cannot search it the way you would search a recipe box, and nothing connects a chosen dinner to a night on a calendar. Practically, most people start over from a blank prompt every week, which means the plan never gets better at being theirs.

2. It does not know your kitchen. Every recommendation is made blind to the half bag of lentils, the yogurt going over on Thursday, and the second jar of cumin you already bought by mistake. You can type an inventory into the chat, and it works well the first time. Nobody types their pantry into a chat window every week for a year. The moment you stop, the plan starts buying what you own and ignoring what needs using, which is precisely the mechanism behind most household food waste: WRAP’s UK research finds that nearly 40% of the edible food households throw out is binned simply because it was not used in time (WRAP, 2022).

3. The list stops being true. A shopping list in a chat transcript is a snapshot. It does not merge the three onions across three recipes into one line unless you ask, it cannot be ticked off in the store in any way the other person sees, and it does not change when you swap Wednesday’s dinner. Real lists get edited, in the aisle, by whoever is closest to the store. That is the gap between a written list and a working one, and it is the whole argument for a shared grocery list.

There is a fourth, smaller one worth naming: models occasionally get a quantity or a cook time wrong with total confidence. For a shopping list that is an annoyance. For anything involving raw meat or an allergy, check it yourself.

What ChatGPT beats a planning app at

Being fair about this matters, because the honest split tells you what to use when.

• Substitutions on the spot. No buttermilk, no shallots, no tamarind. A chatbot answers that better and faster than any app, and it explains the trade-off.
• “What can I make with these five things.” Open-ended combinatorics is exactly what it is good at.
• Explaining why. Why the pan has to be that hot, why the dough needs to rest, why the sauce split. A recipe app stores instructions. A chatbot answers questions about them.
• Cooking outside your repertoire. A cuisine you have never cooked, a technique you have never tried, a rough plan for a dinner party. Idea generation is its strong suit.
• Cutting a recipe down for one. Awkward fractions, one egg problems, and pan sizes, worked through conversationally.

Use it as an advisor standing next to the stove. Do not ask it to be the system of record.

The workflow that uses both

1. Save the recipes you actually cook into one place, so the pool you plan from is dishes your household has eaten and liked, not a stranger’s average week.
2. Use the chatbot for the thinking. Ask it to combine what needs using, to suggest an overlap, to swap a dish, to tell you what to do with the fennel.
3. Put the chosen dinners on nights somewhere that turns them into a list automatically, so nothing gets retyped.
4. Shop from a live list, shared with whoever else is going, with what you already own left off.

The chatbot does the parts that need language and imagination. The tool does the parts that need memory and arithmetic. Trying to make either do the other’s job is where the frustration comes from. That split is the same one the AI meal planner guide lays out in more detail.

Where Mealhive fits

Mealhive is the memory half of that workflow. You save recipes from a link, a photo, or pasted text and they land as clean, searchable recipes with real ingredient lists, so your collection is the dishes you actually cook rather than a chat transcript. Drop a few onto the week and the grocery list builds itself: quantities added up across recipes, grouped by aisle, with staples you always have left off the list. The list is shared live with your household, so it can be ticked off in the store by whoever is there. AI does the reading and the matching, plain rules do the adding and the sorting, and you can keep using ChatGPT for the “what do I do with this fennel” questions it is genuinely good at. If that division of labor sounds right, get on the waitlist.

FAQ

Can ChatGPT make a meal plan?

Yes, and it does it well if you give it constraints instead of asking for a generic week. Tell it how many nights, how many people, the time you have on a weeknight, what your household will not eat, and the format you want back. Ask for fewer dinners than there are nights, because a plan for all seven is a plan you will abandon partway and throw produce away over.

What is a good ChatGPT prompt for meal planning?

A good prompt names the number of dinners, the eaters and their ages, a hands-on time limit, hard exclusions like allergies or foods nobody eats, the kind of store you shop at, and the exact output you want (ingredients with quantities, hands-on time). Then follow up to fix it: ask it to swap a dish, to reuse fresh ingredients across recipes so nothing is bought for one meal, or to make the whole week simpler.

Can ChatGPT make a grocery list from my recipes?

It can, if you paste the recipes in and ask it to merge duplicate ingredients, add up quantities, and group the result by aisle. Two limits are worth knowing. It will not subtract what is in your kitchen unless you type that out every time, and the finished list lives in a chat transcript, so it cannot be ticked off in the store or seen by whoever else is shopping.

Does ChatGPT remember my recipes and food preferences?

It can retain things you tell it, and you can keep your recipes pasted into a project or a long-running chat, which helps. But that is a transcript rather than a structured recipe collection: the ingredients are not stored as data you can search, scale, or turn into a shopping list, and nothing links a dish to a night on your calendar. For preferences it works reasonably well. For a recipe library it is a workaround.

Is ChatGPT better than a meal planning app?

They are good at opposite things. ChatGPT is better at ideas, substitutions, explaining technique, and figuring out what to cook from what is in the fridge. An app is better at holding your actual recipes, remembering what you own, building a list that adds up across recipes, and sharing that list with the person doing the shopping. Most people get the best result using the chatbot for thinking and an app for the record keeping.