Can ChatGPT plan your meals?
Yes, and better than most people expect. A general assistant will produce a week of dinners with your constraints, adapt it when you object, and cost nothing. Here is exactly where that stops working, without any claims about the assistant that you cannot check yourself in five minutes.
What a general assistant genuinely does well
It understands a messy description of a household on the first try. It handles contradictions gracefully — no dairy but yes to butter sometimes — in a way that rigid forms cannot. It explains its choices, suggests substitutions, and rewrites the whole week when you say the last one was too heavy. For a one-off week, or for inspiration when you have run out of ideas, it is a good tool and it is free.
Anyone telling you it cannot read your constraints or cannot write a coherent plan is selling you something. It can.
Four structural differences
Not opinions about output quality — differences in what the system is able to guarantee.
| A general assistant | Repast | |
|---|---|---|
| Where dishes come from | Composed. The recipe is generated text, which may or may not correspond to something anyone has cooked. | Selected from 10,000+ published recipes with their own steps, yields, prep times and photography. It cannot compose one. |
| Quantities | Computed in prose, in the same pass that chose the dishes. | Computed on the device by a deterministic engine against a 33-unit table, separately from the model that chose the dishes. |
| Constraints | Held in the conversation. They persist as long as the context does, and a long thread can drift. | Held as structured settings. Ten allergens are re-applied to every proposal and re-validate an existing week when you add one. |
| Where it works | Wherever you have signal and can copy the answer somewhere useful. | The confirmed week is downloaded in full, so the list and the steps work in a store with no signal and in airplane mode. |
The difference that shows up in week three
A single generated week is not where the gap appears. It appears when Wednesday changes. In a chat window, changing one night means asking for a revision and reading a new plan to work out what else moved, then reconciling that against a shopping list you already used. In Repast, moving Thursday onto Friday is a drag, the ticks you have already made are quantities rather than checkboxes so they survive the change, and the undo is always there.
The same is true of adding a constraint late. Telling an assistant in message fourteen that someone has developed a nut allergy gets you a revised plan. Telling Repast re-validates the week you are already in, pinned dishes included, and pulls out what now conflicts rather than quietly replacing it.
What you get that a chat window cannot give you
A shopping list you can tick off in the aisle with no signal. A plan that is still there on Wednesday, on your phone, without you having to find the right message. Rules that hold on the tenth change as firmly as on the first. And a catalog of published recipes with photography, real yields and real prep times behind every night of the week.
Questions about AI meal planning
Can ChatGPT plan my meals?
Yes. Ask a general assistant for a week of dinners with your constraints and it will produce one, often a good one, and it will happily rewrite it when you push back. The limits are not about the quality of the writing: a general assistant composes recipes rather than selecting published ones, does its arithmetic in prose, keeps your constraints only as long as the conversation holds them, and is not in your pocket in a grocery store with no signal.
Is Repast built on ChatGPT?
Repast uses a server-side language model to choose the week. It is not built on ChatGPT specifically, and the model is not what the product is. What surrounds the model is: a fixed catalog it must choose from, a deterministic on-device engine that computes the quantities, allergen rules it cannot override, and the whole week stored locally so it survives losing signal.
Which one gives a more accurate grocery list?
A deterministic engine and a language model fail differently, and that is the whole point. Repast totals ingredients against a table of 33 units that each declare a dimension and a conversion factor, so the same input always gives the same total and you can re-derive it. A model adding half a cup to three tablespoons in prose is usually right; when it is not, nothing in the output tells you which line to check.
Will a chatbot give me the same meal plan twice?
Generally not, and often that is the appeal. It becomes a problem when a plan has to survive editing: change one night and you can get a different interpretation of a constraint you set three messages ago. Repast holds constraints as structured settings rather than as conversation history, so the tenth change is checked against the same rules as the first.
Do I actually need a meal planning app?
If you cook a couple of times a month, or you enjoy the back-and-forth of getting a plan out of an assistant, then no — the free option is genuinely enough and this page will not argue otherwise. The case for a dedicated app is a household running the same loop every week with fixed constraints, where the value is not in generating a plan but in shopping it, cooking from it and changing it without redoing the whole thing.
The model is the easy part
A fixed catalog, arithmetic you can check, rules that persist, and a week that still works with no signal — that is the part a chat window does not have.
No account to create. iOS 17 or later.