How we use AI

AI helps you find and understand things on this site. It does not decide what a recipe contains. Ingredients, allergen tags, nutrition figures and recipe steps all come from our own catalogue, and no AI writes its way into it. AI is also confidently wrong sometimes, so treat its answers as a knowledgeable friend rather than a reference.

Chef. Ask a cooking question and Chef answers using our own recipes, ingredient notes and techniques as its source material. It is grounded in the catalogue on purpose: when the answer is not in there, it should tell you that rather than invent something. It has no access to your account and cannot change your pantry or plans.

Fridge photo scanning. Take a photo of your fridge or shelf and we try to read the ingredients in it. The list it produces is a suggestion. Nothing goes into your pantry until you confirm it, and low-confidence guesses are flagged so you check them. It reads shapes and labels, not use-by dates, and it cannot tell you whether the thing in the photo is still good to eat.

Recipe adaptation. Ask for a version of a recipe with fewer ingredients, a different diet, or a different number of servings, and we generate a variation. It is a suggested rewrite of a recipe, saved to your device as your own copy. It is not published to the catalogue and nobody else sees it.

Your questions to Chef and your saved adaptations are kept on your device, in the same browser storage as your pantry — see what gets stored on your device.

Fridge photos are processed to work out what is in them and then discarded. We do not store the image, it never appears anywhere public, and it is not attached to your profile. What can remain, on your device only, is the resulting list of ingredients you confirmed.

Where a third-party model is used — currently OpenAI, when the site is configured with it — the text of your question or the photo being scanned is sent to that provider to process the request. We do not send your name, email or account details with it. Some features run on our own retrieval and rules rather than an outside model, so not every request leaves the site at all.

• It does not publish recipes. No AI-generated recipe is added to the catalogue automatically. A person reviews anything that gets published.
• It does not set allergen tags or nutrition. Those come from our ingredient data. They have their own accuracy limits, explained in food, allergens and nutrition, but they are not a model's guess.
• It does not invent food-safety times. Cook Mode timers come from the recipe. If you ask Chef how long leftovers keep or whether something has gone off, it should stay cautious and point you to a food-safety authority. That is the correct answer, not a cop-out: nobody can judge your fridge from here.
• It does not give medical or dietary advice. If you are managing an allergy, a medical condition or a pregnancy, talk to a professional who knows your situation.

AI systems produce fluent answers whether or not they are correct, and they can misread a photo, miss an ingredient behind a jar, or repeat something confidently that is not true. Assume a mistake is possible, especially where it matters: allergens, raw ingredients, cooking temperatures and anything you are relying on for someone else's health.

You are responsible for what you do in your kitchen. To the extent the law allows, we are not responsible for loss caused by relying on an AI answer — read that alongside the Australian Consumer Law section of our terms, which sets out rights that cannot be excluded.

If Chef gives you a bad answer, mark it unhelpful, or email hello@iwantsomefood.com. We read those.

AI requests cost real money to run, and this site has no revenue, so there are daily limits on how many Chef questions, photo scans and adaptations one person can make. If you hit a limit, everything else keeps working and the allowance resets the next day. Please do not try to get around the limits with scripts or multiple accounts.

We do not sell your questions, photos or cooking history, and we do not hand them over to be used for training someone's model. If that ever changed, it would be opt-in and we would say so here first rather than update this page quietly.

The models behind these features may change as better or cheaper options appear, and some features fall back to a simple non-AI mode when no provider is configured — in that case the site tells you it is a practice run rather than pretending. If we add a new AI feature or a new provider that receives your data, we will update this page and the date at the top.