The pitch for an AI meal planner is almost too good: describe your week, get a plan, go shopping. No Sunday-night stare into the fridge, no scrolling forty recipes to land on the same stir-fry you always make. We were sceptical enough about that pitch to spend a month actually testing it — planning our own dinners with an AI planner instead of doing it by hand, and writing down what happened.
Some of it genuinely surprised us. Some of it was worse than doing it ourselves. Here's the honest version, including the bits that don't flatter the technology, because if you're weighing up whether an AI meal planner is worth your time, the marketing copy isn't going to tell you where it falls over.
The setup: what we actually asked for
We're two people who cook most nights, own more recipes than we can remember, and lose about twenty minutes every Sunday deciding what to eat. That's the problem to beat — not "we don't know how to cook", but "we can't be bothered choosing".
So the brief we gave the planner each week was deliberately vague, the way you'd tell a friend: four dinners, quick midweek, something vegetarian in there, use up whatever's in the fridge. Then we let it choose. The rule for the month was that we'd cook what it picked unless there was a real reason not to — no quietly overriding it back to our usual five dinners.
Week one: it beat us at remembering our own library
The first thing an AI planner does well is embarrassingly simple: it reads your whole recipe library, and you don't. We have well over a hundred saved recipes and we cook maybe fifteen of them on rotation. The rest are recipes we liked enough to save and then completely forgot existed.
Week one's plan had three of those forgotten ones in it. Not because the AI is a genius, but because it has no habits. It isn't tired at 6pm, it doesn't default to the thing it can cook without reading, and it has no opinion about the recipe we had a bad experience with in 2024. That alone justified the first week — the same effect you get from keeping all your recipes in one place, except something is actively reading the pile back to you.
The second thing that landed immediately was the shape of the week. It didn't try to fill seven nights. Ask for four dinners and you get four, with the rest of the week left alone — which is exactly the argument we've made about how many meals to actually plan per week. A planner that fills every slot is a planner you'll abandon by Wednesday.
Where it fell flat: the ambitious Wednesday
Now the bad part. AI planners have no idea how tired you are.
Ask for "quick midweek" and you'll get recipes that claim thirty minutes, which is true if you're a food photographer with prepped mise en place and false if you got home at 7:10pm and the onion isn't chopped. Twice in the first fortnight we ended up staring at a Wednesday recipe that was, technically, thirty minutes, and practically, an hour with washing up. We cooked eggs instead.
The fix wasn't cleverer prompting. It was telling the planner which nights are actually busy — once we started marking the genuinely flat-out nights, and letting takeaway and leftovers be legitimate answers instead of gaps to fill, the plans stopped being aspirational. That's a real limitation worth knowing before you start: an AI meal planner optimises for the week you describe, and most people describe the week they wish they were having.
The repetition problem nobody warns you about
By week three a pattern showed up. Left to its own devices, the planner kept reaching for the same tone of dinner — a lot of one-pan, a lot of tray bakes, a lot of things that are easy to describe. Variety within a week was fine. Variety across weeks drifted.
This is the part where you have to actually steer it. Saying "less pasta this week" or "we had a tray bake on Monday, do something else" changed the output immediately and for the better. Which is the honest headline for the whole month: an AI planner is very good at the first draft and mediocre at the second one unless you tell it what was wrong with the first.
If you've ever split the mental load of dinner with a partner, this will feel familiar — it's the same conversation, just faster. Worth reading alongside our take on splitting cooking responsibilities, because the planner doesn't remove that negotiation, it just gives you something concrete to negotiate about.
The web-discovery experiment
Halfway through we turned on the part that goes looking beyond our own library — genuinely new recipes from the web rather than the ones we'd already saved.
This was the biggest swing in either direction. When it worked, it was the best thing about the month: real recipes from sites we already trust, checked against what we eat, dropped into the week alongside our own. We found two dinners we'd never have gone looking for and have cooked both again since.
When it didn't work, it produced recipes that read beautifully and cook awkwardly — the sort of thing that photographs well and turns out to need an ingredient nobody in this hemisphere stocks in July. Roughly one in five, at a guess. Not a disaster, because you see the recipe before you commit to it, but not the effortless magic the category likes to advertise either.
What actually changed about how we plan
Four things stuck after the month ended.
- Planning got shorter, not automatic. Twenty minutes of deciding became about five minutes of reviewing. That's the real saving — not "no effort", just much less of it.
- The library finally earned its keep. Saving recipes only pays off if something reads them back to you. This does.
- We stopped over-planning. Being handed a four-dinner week, repeatedly, broke the habit of optimistically scheduling seven.
- The grocery list stopped being a separate job. Because the plan feeds it directly, the shop was already written by the time we'd agreed on the week — the same automatic grocery list workflow, minus the planning step we used to do by hand.
Where an AI meal planner still can't help
It can't tell you what you feel like eating. It can guess from what you've saved and what you've said, and the guess is decent, but the thing that makes a dinner right on a particular night — the weather, the day you've had, the half a cabbage that needs using — is still yours to supply. The best weeks were the ones where we told it those things. The worst were the ones where we expected it to know.
It also can't fix a thin library. If you've saved twelve recipes and eight are from the same site, the plan will look like that. Give it a proper spread to work from and it gets noticeably better, which is an argument for spending a fortnight saving recipes from anywhere you cook from before you judge the planning.
So is an AI meal planner worth it?
For us, yes — with a caveat that matters. It's worth it if your bottleneck is deciding. If you own plenty of recipes and still end up ordering in because nobody wants to choose, this genuinely removes that stall, and it does it in the time it takes to boil the kettle.
It's not worth it if you love the choosing. Some people's Sunday ritual is a coffee and a browse and a slow argument about Thursday, and handing that to a machine takes the good part away. There's no version of this that's better than a couple who enjoy planning together — the tool is for the weeks you don't.
And it's not worth it if you're expecting to stop thinking altogether. A month in, our honest description is that it's a fast, forgetful, slightly over-optimistic sous chef who's read everything you own. Useful. Not autonomous.
How we'd tell someone else to start
If you want to try this properly rather than getting one underwhelming plan and giving up:
- Save thirty recipes first. Anything less and you're judging the planner on a library that can't support it.
- Ask for fewer dinners than you think. Three or four. Let leftovers and a night off do the rest.
- Mark the busy nights honestly. The plan is only as realistic as the week you describe.
- Push back on the first draft. "Less pasta." "Nothing that needs the oven." One sentence, much better plan.
- Judge it on week three, not week one. The first plan is a novelty. The third one tells you whether it fits how you two actually eat.
That's the workflow slrp's AI meal planner is built around — it plans from the recipes you and your partner have already saved, finds new ones when your library's gone stale, and hands the result to the shared grocery list without you touching a thing. Three plans are on us before there's anything to pay, which is roughly enough to reach the week-three verdict for yourself.
If you'd rather see the unassisted version first, our complete guide to meal planning for couples covers the whole loop by hand — and honestly, that's still where we'd tell most people to start. The AI is a shortcut through a process you should understand first.
