A ChatGPT meal plan takes about nine seconds to write and about twenty minutes to specify.
The nine seconds are the part everyone talks about. Seven dinners balanced across proteins, a grocery list at the bottom, the whole thing formatted so cleanly it feels like a chore just got deleted. The twenty minutes are what it costs to describe your household to a model that has never met it and will have forgotten it by next Sunday.
That asymmetry is the whole story, and it’s the gap A Better Meal was built into: a plan that’s simply there when you open it, your household described once rather than re-described every week.
Why a ChatGPT meal plan looks like the answer
Because the first one is free, instant, and genuinely impressive.
The demo lands every time. You type a sentence about your family and out comes a week of dinners with more coherence than most of us manage on our own. The nutrition looks reasonable. The recipes are real recipes. There’s a grocery list, and it’s even organised.
For a one-off week, that’s a useful thing to have, and anybody telling you otherwise is selling something. The trouble starts on the second week, and compounds every week after.
What the first ChatGPT meal plan gets right
Worth being precise about, because the failure isn’t in the writing.
- The dinners are sensible and the balance across the week is sane
- It respects a constraint you name: gluten-free, no pork, under thirty minutes
- It costs nothing to try
- It rewrites the whole week if you ask, without complaint
- It’s better at breadth than any of us are at 6pm on a Wednesday
A language model is very good at producing something plausible about dinner. Dinner is not a plausibility problem.
Where the ChatGPT meal plan breaks
It breaks at the blank prompt, every single time.
Everything the model knows about your household, you told it. That’s the arrangement. So every week you re-supply the same context: how many of you there are, who went vegetarian in October, which nights are realistic, that the eight-year-old has decided rice is a texture issue now. Miss a detail and the plan’s confidently wrong. Include them all and you’ve spent twenty minutes writing a brief.
Writing the brief is the planning. The work moved into a chat window and got called automation.
That’s the trick of it. You didn’t hand off the job. You’ve just been given somewhere new to do it.
Memory helps a little, and today’s models do carry preferences between conversations. Preferences aren’t state. Your household’s state is the swimming lesson, the cabbage nobody’s dealt with, and the fact that you cooked twice this week instead of five times because work went sideways and Tuesday disappeared into a meeting that wasn’t on the calendar.
Then there’s the quieter failure. The model doesn’t know what you actually cooked. It’s got no idea the salmon died on Tuesday when nobody got home before 7, so it’ll suggest salmon again next week, cheerfully, because as far as it knows salmon went great.
That’s fine once. It’s wearing by March.
The list itself is the other problem. Ask for one and you’ll get one, but it arrives as a block of text: no aisles, no order, and no way to tell which of those forty items belong to the dinner you’ve already decided to skip. Cross a dinner off and nothing downstream knows. It doesn’t survive a supermarket on a Saturday morning with a trolley and a five-year-old.
The prompt becomes one more thing you maintain
This is the part nobody warns you about.
Households that keep a ChatGPT meal plan running past a month end up with a saved prompt. A long one.
Watch what accumulates in it:
- The roster, and who’s currently off which food
- The dinners that got rejected, so it stops suggesting them
- Which nights are realistically cooking nights this term
- Standing constraints: the nut allergy, the thirty-minute ceiling
- Last week’s plan, pasted in so this week doesn’t repeat it
- A note about the yoghurt, because there’s always too much yoghurt
Some households graduate to a spreadsheet that feeds the prompt. At that point the ChatGPT meal plan has quietly become a small, badly documented piece of software with exactly one maintainer, and the maintainer is a tired parent at 10pm who would like, more than almost anything, to be watching television instead.
Nobody set out to build that. It accretes, one useful note at a time.
Which is a strange place to arrive at, for a tool that was going to take the planning off your plate. The planning didn’t leave. It got a new interface, and the interface has a blinking cursor.
Mark Semmelbeck, who founded A Better Meal, puts the distinction plainly: “We don’t want to be a tool for doing a task. We are a solution to a problem.” A prompt is a tool for doing a task. It’s a very good one. It hands the task straight back to you every Sunday.
What a plan-first app gives a household instead
The relief is that the week is already there when you open the app, and nobody had to describe the household again to get it.
You tell A Better Meal once who you’re feeding, what they steer clear of, any allergies, and how many nights you actually want to cook. The plan is built from that picture. Nobody is reconstructing it at 12:40 on a Tuesday, sandwich going stale at the desk, trying to recall which child is off fish this month.
The grocery list falls straight out of the plan in one action, sorted by aisle. Push a single recipe to it, or several, or the next three days, or the next six. Edit a line and the plan underneath doesn’t break. The recipes you already trust come into the app with you, so a plan that fits your house doesn’t start from a stranger’s idea of Tuesday.
None of that requires you to remember what you told it in March. You’re not the one keeping it alive.
Generic is the problem, and the model is fine
The model is remarkable. General-purpose is the constraint.
A general tool aims at the average of everything ever written about meal planning, and the average household is a fiction. There’s no average number of cooking nights, no average picky eater, no average Tuesday. Every real family is a pile of specific constraints, stacked in an order nobody else would recognise, and reconstructing that pile from scratch in a text box is exactly what a general-purpose assistant asks of you every single time you open it.
That reconstruction is the tax. It’s small on week one, when writing the brief still feels like a novelty. It’s brutal by week six. By March you’ve stopped opening the thing at all, and dinner’s back where it started.
A plan you can lean on when there’s nothing left
Any plan looks fine on the Sunday you made it. The real test comes on the Wednesday you’d forgotten it existed.
At 6:15, with practice having run late and one child negotiating hard about the broccoli, a chat window asks you to type. The plan just tells you what’s for dinner. That’s the whole promise. It’s a small one, which is exactly why it holds on the nights that are going badly, when a bigger promise would have needed something from you that you don’t have left to give.
You keep the last word. Swap Thursday for something faster, leave the week exactly as it is, order in on Friday and start again Saturday. The deciding was handled before you arrived; what you do with it never stopped being yours.
Dinner, handled. Try a week with A Better Meal.
