The Future of Meal Prep: AI as the Customisation Layer
By Alex Fahey, founder of Chop it. Last updated 3 August 2026.
Short answer: meal prep is about to stop being one-size-fits-all. For twenty years the tools handed every household the same plan, the same recipe database, the same "7-day clean eating" PDF, and left you the hard part: adapting it to your household, your fridge and your tastes. AI moves that work to the machine. Say "five dinners under 30 minutes, no fish, the kids won't touch mushrooms" and get a real plan with a real shopping list back, and customisation turns from a premium feature into the default. That is the layer AI adds to meal prep.
This is an opinion piece about where I think meal prep goes next, written from inside the problem. Read it as a bet rather than a forecast.
Quick answers
- What does AI change about meal prep? The interface. Constraints you used to configure with filters and spreadsheets ("quick, veggie twice, no coriander") become a sentence, and the plan adapts to you instead of the other way round.
- Is this recipe generation? Generated recipes are the easy half. The valuable half is planning against your constraints and consolidating the result into one accurate shop.
- Will AI replace cooking? No. It takes the admin around cooking: deciding, listing, de-duplicating, remembering.
- Can I use any of this today? Yes. Chop it runs inside ChatGPT: find or create recipes, build a weekly menu and turn it into one shopping list. Open a result in the iPhone app when you want to keep and cook it.
- What's still to come? Deeper personalisation: plans that know your pantry, your household's tastes and what you cooked last month, without you filling in a form.
Meal prep 1.0 was built for a fictional household
The first generation of meal-prep tooling had one move: hand you someone else's plan.
The Sunday batch-cook guides, the laminated 7-day plans, the apps with a thousand recipes behind a filter bar. All of it assumed a household that does not exist: one that eats anything, has unlimited time, owns every ingredient and will eat the same turkey and rice five days running.
Real households have constraints. A partner who won't eat fish. A Tuesday that's always a write-off. A child in a beige-food phase. Half a bag of spinach that needs using by Thursday. A budget. The old tools could not see any of that, so you closed the gap between "the plan" and "your week" by hand. Most people close it for about a fortnight and then stop. I did that three or four times over, which is a large part of why Chop it exists.
The most successful meal-prep tool of the last decade was the takeaway app, which at least answered tonight's question with no admin.
The missing layer is customisation
Strip it back and meal prep is a constraint-satisfaction problem. Time, money, tastes, dietary lines, what's already in the fridge, how much variety you can stand. The plan that works is the one that fits your constraints, and the constraints change weekly.
Software has been bad at collecting those constraints. Few people finish a 40-question onboarding about their food preferences, and the answers would be stale by March anyway. So the tools defaulted to the average household and shipped one plan to all of them.
Language models change the economics. A sentence carries a lot of constraint data: "five dinners under 30 minutes, no fish, one veggie, use up the chicken thighs" is six constraints in four seconds, with no forms. The interface for customisation now matches how you think about food.
So I call AI the customisation layer for meal prep. Generating a plausible recipe was never the bottleneck. Fitting the week to the household was.
The conversation is the easy half
There is a catch, the same one I wrote about in why ChatGPT changes home cooking: a general assistant can hold a lovely conversation about dinner and still fall over at the checkout.
Ask a bare LLM for a week of meals and you get prose that looks like a plan. It holds no real quantities, it cannot merge the coriander across three recipes, and its "shopping list" is an approximation that dies on contact with a supermarket. The hard half is structure: real recipes with real quantities, consolidated into one accurate, de-duplicated shop.
The shape of it: assistants hold the conversation, and a structured layer underneath knows recipes as data. The assistant carries your constraints, and the layer turns them into something you can cook and buy.
What this looks like today
The first version of this is already live. Chop it in ChatGPT can find or create recipes from a plain-English request, build a weekly menu and consolidate that menu into one shopping list. When an idea is worth keeping, you can open it in the iPhone app as a structured recipe with metric quantities, clear ingredients and a method you can follow.
The app is the permanent layer. It keeps AI recipes alongside the recipes you rescue from websites, TikTok and Instagram, or scan from a cookbook page. From that one library you can plan the week, build the list and cook. A good dinner idea should outlive the chat that produced it.
That is the customisation layer, version one: constraints in a sentence, structure underneath and your own mixed recipe library in place of someone else's fixed plan.
Where it goes from here
Predictions rather than promises. This is the direction I am building in, and I would be surprised if the industry lands somewhere else.
Plans that know your kitchen. The next obvious step is pantry-aware planning in the conversation: the assistant knows you have half a bag of spinach and two chicken thighs, and the week starts from there. The app does this today with "What's In?", and wiring that awareness into the assistant layer is the step that pays, because the cheapest and least wasteful ingredient is the one you already own.
Taste memory in place of preference forms. Each plan you accept, swap or bin is preference data you did not have to type. Over months the plan should drift toward your household: more of what you cooked, less of what you skipped, with no settings page involved.
The household rather than the user. Food is the most multiplayer decision in the home, and today's tools treat it as single-player. The customisation layer has to hold the whole household's constraints at once, your fibre goal against their fish veto, and find the overlap.
Variety as a first-class constraint. Personalisation narrows. Feed a recommender your favourites and it serves you the same five dinners for a year. Food is the one domain where the system should push the other way, which is why Chop it scores the variety of your week across plants, fibre and protein rather than learning your favourites. The plan I want keeps your comfort food and widens the range around it.
What won't change
The cooking. None of this touches the twenty minutes at the hob, which is the good bit. Your Friday lasagne stays exactly as it is.
Everything around it changes: the deciding, the listing, the de-duplicating, the guilt about the spinach. That is admin, and admin is what software should take. WRAP estimates that an average UK household of four wastes food worth about £1,000 a year. Make the admin easier and more of the food you buy has a chance of becoming dinner.
The bottom line
Meal prep tools spent twenty years handing households the same plan and treating the gap between that plan and your life as your problem. AI closes the gap from the other side: your constraints, spoken like a human, turned into a week you can cook and one shop you can buy.
The assistants will hold the conversation and the structured layer underneath will do the arithmetic, which leaves you with your plan rather than the average household's.
If you want to see how far along it already is, open Chop it in ChatGPT and ask for your week.
Sources: WRAP (UK household food waste cost). Product capabilities described are current as of 3 August 2026. Forward-looking sections are the author's opinion about direction, not announced features.