Service "What to cook": how to get a practical recipe from available products
The service "What to cook" is a text recipe generator. You enter ingredients and their quantities, preferences or restrictions, the number of portions and available time, and you get one text result. In this article we will go through how to prepare the request, what to check in the generated recipe and in which cases the tool will not fit. The interface is a web form with text fields, no sliders, no style presets and no change levels. The result can be copied or downloaded as a plain TXT. The service does not create images, audio or video, does not count calories and does not store your data. Model responses may contain errors and availability may change.
Open tool: Recipe generator ↗What exactly the service does and what is left to you
The service "What to cook" accepts four text fields: ingredients and quantities, preferences and restrictions, people or portions, available time. Based on this it offers practical recipes. This is not a step-by-step culinary assistant with timers and not a weekly menu planning system. It does not store your data, does not send notifications and does not publish recipes. Everything you get is one text response that can be copied or downloaded as TXT.
Manual work starts before input: collect an exact list of products and their quantity. For example, "potato — 4 medium tubers, carrot — 1 large, onion — 1 head, boiled lentils — 200 g, tomatoes — 2 medium". The more specific, the more useful the result. After generation, check that all ingredients are accounted for, there are no extras, and proportions match your experience. The service may make mistakes, so the final decision is always yours.
AI handles well combining familiar products and suggesting a basic dish structure. But it does not know your stove, pot size, product freshness and personal preferences for salt and spices. Therefore manual adaptation is mandatory: reduce or increase the amount of liquid, adjust thermal processing time, add spices to taste.
Input example: ingredients — potato 4 pcs., carrot 1 pc., onion 1 pc., lentils 200 g, tomatoes 2 pcs.; restrictions — no gluten, no milk; portions — 2; time — 30 minutes.
How to prepare input data to get a useful recipe
First field — ingredients and quantities. List them via comma or new line. Indicate not only the name but the state: "boiled lentils", "raw tomatoes", "frozen peas". If a product needs prior processing, write it. For example, "dry lentils, soaked overnight" — the service will account that it needs boiling. Maximum length of the first field is 8000 characters, but for a regular home set 200–400 signs is enough.
Second field — preferences and restrictions. This includes dietary requirements (no gluten, no lactose, vegan), disliked products (no mushrooms, no spicy), as well as desired cuisine (Italian, Mediterranean). Do not write too generally: "something useful" — a bad guide. Better: "no sugar, low salt, many vegetables". Third field — number of portions or people. Fourth — available time in minutes. If time is short, specify "20–30 minutes"; the service will offer faster options.
Check for contradictions: for example, "no gluten" and "use wheat flour" at the same time. The service may try to resolve the conflict, but the result will be less reliable. Better to immediately exclude incompatible products or indicate priority: "main thing — no gluten, flour can be replaced". Also do not overload the field with details that do not affect the recipe: "I have a new frying pan" — extra.
Good input: "potato 4 pcs., carrot 1 pc., onion 1 pc., boiled lentils 200 g, tomatoes 2 pcs.; no gluten, no milk; 2 portions; 30 minutes". Bad input: "something from potato and lentils, quickly, tasty".
What to check in the generated recipe
First — compliance with restrictions. If you specified "no gluten", check that there is no wheat flour, soy sauce, breadcrumbs in the recipe. If "no milk" — did not appear butter, cream, cheese. The service may not know that some products contain hidden gluten or lactose. Read the composition carefully and replace ingredients manually if necessary.
Second — proportions and time. AI may suggest boiling potato 10 minutes, though for your tuber size 20 minutes are needed. Or indicate too much liquid for stewing. Evaluate realism: if the recipe implies 30 minutes, and it has three stages of 20 minutes each, this is a mismatch. Check if the indicated amount is enough for the stated portions. Sometimes the service gives excessive amount of oil or salt — this can be corrected to taste.
Third — culinary logic. Does it suggest adding raw lentils to a dish without cooking? Does it require unavailable technique (for example, multicooker, if you only have a stove)? Does not contradict common sense: first fry onion, then add water and cook potato — normal; first cook potato, then fry it with onion — also possible, but clarify what was meant. If something seems strange, rephrase the request and generate again.
Generated recipe may suggest "stew potato with lentil and tomatoes 25 minutes". Check: lentil is already boiled, potato cut into cubes — then 25 minutes is enough. If lentil raw, time needs to be increased.
When the service is not suitable and what to do instead
The service will not suit if you need a photorealistic image of the dish. It does not generate images and does not create audio or video. In the interface there is only text output. If visualization is required, use other tools or search photos in verified sources. The service is also not intended for exact medical or dietary calculations: it does not count calories, proteins, fats and carbohydrates, does not give personal nutrition recommendations at diseases. For therapeutic nutrition consult a doctor-dietitian.
Do not rely on the service if you have allergy to a rare product or severe form of intolerance. AI may not know all hidden sources of allergen. In such cases it is better to use verified recipes from specialized sources and carefully read the composition of each product. The service is not suitable for industrial volumes: it is designed for home kitchen, not for production.
If the service is unavailable due to queue or restrictions, do not try to bypass it. Just wait or use another tool. Do not load confidential data into fields: the service does not require registration, but does not guarantee privacy. Enter only what you are ready to send for processing. Remember that model responses may contain errors and availability may change.
Situation: you have allergy to nuts, and you ask for a recipe without nuts. The service may suggest a dish with nut sauce, if you do not specify the restriction explicitly. Always check the composition.
Practical example: from request to ready dish
Suppose you have potato, carrot, onion, boiled lentil and tomatoes. Need to cook for two people in 30 minutes, no gluten and no milk. You enter: "potato 4 pcs., carrot 1 pc., onion 1 pc., lentils 200 g boiled, tomatoes 2 pcs.; no gluten, no milk; 2 portions; 30 minutes". The service may suggest, for example, stewed vegetables with lentil. This is an example, not a guaranteed result: every time the answer may differ.
Having received the recipe, check: were all ingredients used? If the service added garlic or spices you do not have, replace or exclude. Make sure the cooking time is realistic: chopping vegetables takes 5–7 minutes, frying onion — 3–5, stewing — 15–20. Total about 30 minutes. If the recipe suggests cooking lentil from scratch, the time will exceed the limit — then use canned or already boiled lentil.
After cooking evaluate the result: enough salt, did not burn, liked the texture. Record successful proportions, so next time enter them more precisely. The service does not remember your preferences, so save successful options manually in a text file. Thus you gradually create your own base of recipes, adapted to your products and tastes.
Example output: "Fry onion and carrot 5 minutes. Add chopped potato and tomatoes, pour 200 ml water. Stew 15 minutes. Add lentil, salt, pepper, stew 5 more minutes". This is only an example, the generated text may be different.
How to work with the service without unnecessary delays
Do not expect the service to replace a cookbook or an experienced cook. It is a tool for quick ideas and combining available products. Use it as a draft: get an option, then improve manually. Do not spend time trying to get "perfect" recipe from the first attempt. If the result does not suit, change the input data: add or remove product, clarify time, change restrictions. Usually enough two-three attempts.
Remember interface limitations: there is no settings panel, no style choice, no possibility to compare several variants at once. You get one result at a time. If you need several ideas, generate several times with different formulations. Do not try to upload a document or image — it is not supported. Enter only text. Maximum length of the first field is 8000 characters, second is 1000, but for a recipe this is more than enough.
To save the result use the copy button or download TXT. Do not rely on auto-save: it is not there. If the service is temporarily unavailable, this may be due to queue or load. Just try again later. And main: always apply common sense. If recipe seems dangerous (for example, insufficient thermal processing of meat or eggs), do not follow it. AI does not bear responsibility for your health.
Tip: save successful recipes in a separate text file with date and list of products. Thus you can quickly repeat the dish.
The service "What to cook" is a text assistant for generating ideas based on your products. It does not replace manual checking, does not guarantee exact proportions and is not suitable for medical diets. Best approach: give AI the rough work of combining, and you check constraints, time and logic yourself. Use the service as a source of inspiration, not as a final recipe. And remember: model responses may contain errors and availability may change. Cook with pleasure and common sense.
Open tool: Recipe generator ↗Interface translations are machine-assisted and may need refinement.