Picture this: you open the fridge and find half a zucchini, two eggs, a bit of leftover rice, and some cheese. You have no idea what to make. This is arguably the most rewarding domestic use of a chatbot today—snap a photo or dictate a list of ingredients, and within seconds you get three meal ideas. The real challenge isn’t generating ideas; it’s that the chatbot knows nothing about your kitchen setup or who will be sitting at the table.
For anyone looking to use ChatGPT, Gemini, or another assistant to decide what to cook, plan the week, and avoid forgetting anything at the supermarket, these tools work well for inspiration and organization. But they don’t replace a food label when allergies are involved, they don’t always remind you of proper storage rules, and they can get quantities wrong. Food safety standards still come from authorities like AESAN, not from a chatbot.
- Ingredient-based recipes: Describe or photograph what you have and ask for options—but verify what the AI actually recognized before cooking.
- Weekly planning: AI excels at generating menus and shopping lists, but always double-check quantities and cross-reference the list with the menu.
- Allergy safety: Never rely solely on a chatbot for allergen information; the product label is the only reliable source.
Case 1: ‘What can I make with this?’
This is the simplest use case: describe or photograph what you have and ask for options. Current assistants accept images; according to Gemini’s help documentation, as of November 2026 it allows adding photos and up to 10 files per request (subject to availability), with a maximum of 100 MB per non-video file. If you want to understand what a model actually sees in a photo and where it gets confused, we’ve covered that in our explainer on multimodal AI models.
Three tips to make the suggestions useful:
- State what isn’t visible. The photo doesn’t show your pantry, how much time you have, or which appliances you own. Saying ‘I have 20 minutes, an air fryer, and no oven’ changes the answer. If you’re unsure between devices, we’ve compared the real energy consumption of air fryers, ovens, and microwaves.
- Ask it to use only what’s there plus basic staples you specify (oil, salt, spices). Otherwise, it will tend to add ingredients you don’t have.
- Check that it identified the photo correctly. A chatbot can confuse one cheese with another or miss an expiration date. Ask it which ingredients it recognized before you start cooking.
Case 2: Weekly menu and shopping list
This is where AI saves the most time. Give it the number of people, the days, preferences, and what you already have, then ask for a five-day dinner menu with a shopping list grouped by supermarket section. You can then ask it to reuse ingredients across dishes to avoid food waste.
Two things to always review:
- Quantities. A chatbot can make mistakes when scaling portions. If a recipe serves four and you’re two, double-check the conversion, and be suspicious of odd amounts (‘350 g of egg’).
- That the list matches the menu. Ask it to mark, for each product on the list, which dish it’s used in. That way you can spot what’s extra and what’s missing.
A clarification: this article does not provide dietary or nutritional guidance. If you need a menu for health reasons, that’s a job for a healthcare professional, not a chatbot.
Can I trust AI if someone at home has allergies?
Not as a sole source. A chatbot can help you find recipes without a specific ingredient, but reliable information about whether a particular product contains it is on the label, not in the conversation. A ‘nut-free’ recipe is useless if the pesto you buy contains them.
European regulations are clear on this. Regulation (EU) 1169/2011 lists in Annex II 14 substances or products that cause allergies or intolerances: cereals containing gluten, crustaceans, eggs, fish, peanuts, soy, milk, tree nuts, celery, mustard, sesame, sulfur dioxide and sulfites (above 10 mg/kg or 10 mg/l), lupin, and mollusks. According to AESAN, in packaged foods these must be emphasized typographically in the ingredient list, and Article 44 of the regulation requires that they also be declared for non-packaged foods.
AESAN also reminds that the mention ‘gluten-free’ can only be used if the food does not exceed 20 mg/kg of gluten, and ‘very low gluten’ if it does not exceed 100 mg/kg. A chatbot has no way to verify that for you.
If there are allergies at home, a practical option is to set them in the custom instructions of ChatGPT, Gemini, or Claude so the assistant takes them into account in every recipe. It helps you not forget, but it doesn’t eliminate the need to read labels. And keep in mind that you’re sharing health data with a company.
What the chatbot won’t remind you: food safety
A chatbot can give you a recipe for marinated anchovies or rice for three days without mentioning any risk unless you ask. These are AESAN rules worth keeping in mind when following any recipe, wherever it comes from:
- Raw or undercooked fish. To prevent anisakis at home, AESAN recommends freezing fish at –20 °C or lower for at least five days, and only in freezers rated three stars or higher: lower-rated freezers don’t reach that temperature.
- Rice, pasta, and leftovers. The bacterium Bacillus cereus is especially associated with cooked rice. AESAN recommends refrigerating cooked food promptly if not eaten, not leaving it more than 2 hours at room temperature, not stacking containers so they cool properly, and reheating to 74 °C or higher for at least 15 seconds. Note: one of its toxins survives up to 121 °C, so reheating doesn’t fix badly stored rice.
- Homemade preserves. Among the foods most associated with botulism, AESAN’s fact sheet cites homemade preserves, especially vegetable ones like chilies in oil, and considers that, whenever feasible, it’s better to discourage them. If an AI suggests canning something, look up the procedure in an official source before doing it.
If in doubt, ask the chatbot itself about the risk and request its source, then verify it. Its answers can sound confident without being correct.
My take: AI in the kitchen is one of the most useful and least controversial applications an assistant has today, precisely because the cost of a bad idea is usually just a mediocre dinner. But there are three areas where the cost of a mistake is different: allergies, storage, and raw fish. There, the chatbot should be the one proposing and you the one verifying, with the label in one hand and AESAN guidance in the other.