Custom Instructions in 4 steps

ChatGPT Custom Instructions: the two fields, solved in 4 questions

Answer four things about your work and the app will write the two official custom instruction fields, already measured for ChatGPT to accept them.

No one abandons custom instructions because they don't know what to put: they abandon them in front of an empty box with a red counter.

ChatGPT has two custom instruction fields and both are poorly explained. The first —‘what should ChatGPT know about you’— is not a biography: it's the context you don't want to repeat in every conversation. The second —‘how do you want ChatGPT to respond’— is not an adjective: it's behavioral commands. Writing them by hand almost always ends the same way, in two polite paragraphs that change nothing.

Here you don't write: you answer. The questionnaire has exactly four steps and none ask you to write. Step 1: what you do and your technical level. Step 2: what you work with every day —tools, type of company, for whom— in a free field of up to 900 characters. Step 3: how you prefer the responses (concise, analytical and detailed, or results-oriented). Step 4: what you hate about ChatGPT's responses, in six checkboxes filled with complaints that people repeat in forums: no compliments or beating around the bush, don't agree with me if I'm wrong, complete code without ‘your code here’, no alerts that it's an AI, say when it doesn't have enough data, I prefer a friendly tone.

A single generation returns both blocks at once. The first is in the first person and dense, without filler phrases: profile, tools, context, and objectives. The second is in the second person and is pure command: output format, what it must do, and what it must never do. Each block comes with its own counter and its own copy button, because you will paste them in two different places on the ChatGPT screen, one after the other.

And the part that usually makes you start over: the limit. You state your plan before generating —free, 1,500 characters per field; paid, 5,000— and the character meter is green, amber, or red while the text is in front of you. If a block exceeds the limit, the ‘Adjust to 1,500’ button really trims it down: first without using any AI (removes duplicate phrases, what the other block already says, and filler connectors, always cutting by complete sentence) and only calls the AI if that's not enough. You can undo and return to the long text.

No sign-up

Try Custom Instructions in 4 steps right here

Loading the tool…

Use it free without an account. If you want to save your result, you can sign up afterwards.

Save / Sign up

Benefits of Custom Instructions in 4 steps

Four steps and none ask you to write

profession and level, what you work with, how you prefer the responses, and what bothers you about them. It's all touches and an optional free text field. There's no blank screen at any moment.

Both fields in a single generation

the 'about you' block and the 'how you want me to respond' block come out together, each written in its grammatical person —the first in first person, the second in second person— because ChatGPT reads them with different roles.

The limit is decided before writing, not after

you choose a free plan (1,500 characters per field) or a paid plan (5,000) at the start, and the text is generated already aiming for that length. It's the opposite of writing and then discovering it doesn't fit.

Traffic light meter per block

green below 80% of the limit, amber between 80% and 100%, red if you exceed. Each block has its own, so you know which of the two you need to trim.

"Adjust to the limit" without using up generations

the first level of trimming is deterministic and costs zero —it even works with the daily quota exhausted—. It removes repeated phrases, what is already stated in the other block, and filler connectors, and always cuts at complete sentences, never halfway through a word. Only if that doesn't fit does the AI come in, and you can revert to the long text.

The checkboxes are written with real complaints, not euphemisms

"don't agree with me if I'm wrong", "full code, without your code here", "no notices that it's an AI". What you check goes into block 2 as an explicit prohibition, not as a suggestion.

If you contradict yourself, it tells you on screen

checking both "no flattery or beating around the bush" and "I prefer a cordial tone", or asking for concise and analytical, does not result in an incoherent text. The app resolves the conflict by applying the last thing you checked and informs you of what has been applied.

Start without an account and copy without an account

the questionnaire, the two blocks, the character meters, and the limit adjustment are all complete as a guest, with ten generations per day. The free account is only needed to save multiple instruction profiles and switch between them.

Four answers now, and ChatGPT stops starting from scratch every morning

No account, no card, and no connection needed: the app writes the two blocks and you paste them into the two ChatGPT fields.

In the features that use AI, responses are generated by artificial intelligence and are not a substitute for professional advice.

ChatGPT Custom Instructions: custom instruction generator in 4 steps