Creamy Wild Mushroom & Mascarpone Orzo
Wild mushroom orzo55 min
Learn

Can AI Create Recipes? What Works, What to Check

Read
2 min
Updated
By
Alex Fahey

Pictured: Creamy Wild Mushroom & Mascarpone Orzo

Short answer

Yes. A large language model such as ChatGPT can write a workable recipe in seconds from a plain request, and for familiar weeknight dishes it's usually sound. It follows the patterns of the cooking text it learned from rather than testing anything, so treat it as a first draft and check quantities, times and temperatures before you shop.

Contents

What AI recipe generation is good at

Familiar dishes. Ask for a chicken curry, a tomato pasta or a traybake and the model draws on thousands of published versions, so the technique is usually right.

Constraints. A recipe site filters by "vegetarian". A model takes "vegetarian, one pan, under 30 minutes, no aubergine, and use up the feta" and gives you one answer.

Adapting. Ask for a recipe dairy-free, cheaper or doubled and it rewrites it. In Attest's 2026 survey, 75.9% of UK respondents were comfortable with AI recipe recommendations, against 49.1% for automated online grocery shopping.

The two failure modes to check

Quantities. A generated amount can look right and be wrong. The model copies the shape of a recipe rather than calculating one, so a soup can get double the sensible salt or a tenth of the stock. Check the amounts before you shop.

Times and temperatures. Generated times average every pan, oven and cut the model has read about, so they fit none in particular. Cook until the food is done, not until the timer goes. For meat and fish, check against the Food Standards Agency, not the model.

Baking runs on ratios, and a wrong flour-to-liquid ratio fails in a way a curry never will. For bread and cakes, start from a tested recipe and let AI change the flavourings.

How to test a generated recipe safely

  1. Put all your real constraints in one message.
  2. Read it through. Are the quantities sane? Is a step missing, such as when the onions go in?
  3. Ask it to check itself: "Are these quantities right for four portions?" It catches a fair share of its own slips.
  4. Cook it once as written and note what you changed. After that it's your recipe.

Keeping the recipes worth repeating

A good recipe in a chat log gets buried under next week's chats, so move the keepers somewhere you'll find them. Our guide to saving recipes from any source covers how, and ChatGPT meal planning covers planning whole weeks in a chat.

I built Chop it to plan the week and shop it once, so weigh this accordingly. In the app, Make me a recipe turns a request into one finished recipe for 1 credit. Recipes you import are rewritten in British English and UK metric and credited "Inspired by", with a link to the source. In Chop it in ChatGPT, connect your account and a recipe you make there saves to the library you plan and shop from.

Quick answers

  • Can I trust an AI recipe if someone has a food allergy?
    Don’t rely on it. A model can miss an allergen hidden in a stock cube, a sauce or a spice mix, and it doesn’t know what’s in the brand you buy. Check every ingredient label yourself, as you would with any recipe.
  • Is an AI recipe copied from somewhere?
    Usually not word for word. A model writes a new recipe from patterns across many it has read, so a well-known dish will look like the versions it learned from. If you want a particular cook’s recipe, go to their book or site.
  • Can AI make a recipe from what is in my fridge?
    Yes, and it’s one of the things it does best. List what you have, say which items need using first and what you’re short of, then check the quantities against what’s actually in the fridge.
  • How much does Chop it’s Make me a recipe cost?
    1 credit a recipe. Every account gets 3 free credits a week, which don’t build up, and Pro includes 50 a month. Planning the week is free.

Keep reading