How to use AI for recipes — and what to check before you cook
AI is genuinely good at turning what you already have into dinner. It is less good at times, temperatures and food safety. Here's how to use it properly.
The short answer: open a chatbot such as ChatGPT, Gemini or Copilot, tell it what food you already have, who you’re cooking for and how much time you’ve got, and ask it for a recipe. You’ll get one in seconds, written in plain steps, and you can ask it to change anything you don’t like. What you should not do is trust its oven temperatures, cooking times or food safety advice without checking — those are exactly the details it gets confidently wrong.
That’s the whole method. The rest of this article is about doing it well: what to type, what comes back, and which parts to check before the oven goes on.
Why AI is genuinely good at this
Most recipe websites start from the recipe and work backwards to your shopping list. AI works the other way round. You start with what’s actually in the fridge — half a cauliflower, some bacon, an onion that’s seen better days — and it suggests something you can make tonight without going out.
It’s also endlessly patient about adjusting. Too much garlic? Say so. Need it softer because of dentures? Say so. Cooking for one instead of four? It will do the arithmetic on the quantities for you, which is the fiddly bit most of us get wrong. None of this requires you to know anything technical. If you’ve never used one of these tools before, your first conversation with an AI walks through the very first step.
What to type — three examples you can copy
The single biggest improvement you can make is to stop asking for “a chicken recipe” and start describing your actual situation. Compare these.
Using up what you have:
“I have two chicken thighs, a tin of chopped tomatoes, an onion, rice and the usual dried herbs. Two people, about 40 minutes, one pan if possible. What can I make?”
Cooking around a health need:
“My husband has been told to cut down on salt. Give me a simple fish supper for two with plenty of flavour but very little added salt. Nothing fancy — I’m a decent but not adventurous cook.”
Making an old favourite easier:
“I’ve always made a shepherd’s pie from scratch, but standing at the hob tires me now. Can you give me a version with fewer steps and less time standing up, that still tastes proper?”
Notice what all three have in common: ingredients, number of people, time, and any limits. Give it those four things and the answer will be usable. Leave them out and you’ll get something generic.
Then keep talking. “Can I swap the cream for milk?” “What do I serve with it?” “I only have a small oven dish — will this fit?” This back-and-forth is where the real value is. It costs nothing to ask a follow-up, and there’s no such thing as a silly question. Nobody is listening but you.
Where AI recipes go wrong
Understanding this bit is what separates a good tool from a bad dinner. An AI does not cook, taste or measure. It produces text that looks like a recipe, based on the enormous number of recipes it has read. Usually that’s plenty. Sometimes it isn’t. How does AI “know” things? explains why in more detail, but here is what it means at the stove.
Times and temperatures are the weak spot. It will state an oven temperature and a cooking time with total confidence, and it may simply be wrong — especially for large joints of meat, whole birds, or anything where undercooking matters. Treat every number as a suggestion, not an instruction.
Quantities can drift. Ask it to scale a recipe up or down and it usually manages, but it occasionally leaves an ingredient at the original amount. If a quantity looks odd to you, it probably is. Your own judgement, after decades of cooking, is better than the machine’s.
It sometimes invents. Ask for “Delia’s lemon drizzle” or “my grandmother’s Lancashire hotpot” and it will produce something plausible with a confident name on it. That’s not the real recipe; it’s a reconstruction. If you want a specific, named recipe from a specific cook, get it from that cook’s own book or website.
It can’t see or smell. It doesn’t know your oven runs hot, your pan is thin, or that the sauce has already caught. You do.
None of this makes it useless. It makes it a helpful assistant with no senses — which is exactly how you should treat it. There’s more on this in how to tell when AI is wrong.
The parts you must check with a human, not a machine
Some things are not worth a second’s risk.
Food safety. Whether pork or chicken is cooked through, how long leftovers keep, whether it’s safe to reheat rice, how to handle a large bird — check these against your national food standards body’s own website, or a cookbook you already trust. These organisations publish this guidance free, precisely because getting it wrong makes people ill.
Allergies. If someone at the table has a food allergy, read the actual packet. An AI can suggest a nut-free dish, but it cannot see what’s in the jar in your cupboard, and manufacturers change recipes.
Diets given to you by a doctor. If you’ve been told to manage diabetes, kidney disease, blood thinners and vitamin K, or anything similar, use AI to get ideas — then run the plan past your GP, practice nurse or dietitian. They know your case; the chatbot knows the average of everything ever written.
Preserving, canning and bottling. Home preserving has real safety rules. Use a tested source, not a generated one.
A routine that works
- Say what you have and who you’re feeding. Ingredients, number of people, time, limits.
- Ask for the recipe, then ask for a shopping list of anything missing. It’ll produce one.
- Read it through before you start. Does the order of steps make sense? Does anything look off?
- Check the numbers that matter — meat temperatures, cooking times for anything that must be cooked through — against a source you trust.
- Cook it your way. Your eyes, nose and experience overrule the text every time.
- Tell it what happened. “Too dry — what would you change?” It’ll adjust, and the next version is usually better.
The short version
- Describe your ingredients, your people and your time; don’t ask for a recipe in the abstract.
- Keep asking follow-ups — swaps, portions, what to serve alongside. That’s where it shines.
- Treat times, temperatures and scaled quantities as suggestions to check, not facts.
- Food safety, allergies and doctor’s diets go to a trusted source or a real person, always.
- Your own experience in the kitchen beats the machine’s confidence. Trust it.
If you’d like to see where else this kind of help fits into ordinary life, ten everyday things AI can genuinely help you with covers the rest — and if you’re still unsure what these tools actually are, start with what AI actually is.
