How do you prepare ChatGPT before asking for anything?

Most disappointing ChatGPT answers start with a cold open: a request typed into an empty chat, with nothing around it. The model then has to guess who is asking, why, and what a good answer would look like, and it guesses toward the average. Preparing the conversation first takes a minute and changes everything that follows, because ChatGPT keeps reading the earlier messages of a chat each time it replies.

A Beginner's Guide to ChatGPT treats this as a habit rather than a trick: set the model's role, describe your own situation, and say what kind of output you expect before you ask for anything. A briefing message does all three at once.

The four parts of a briefing

Who you are. Your job or role, your field, and how much you already know. An answer pitched at a beginner reads very differently from one written for a specialist.

The goal. Not just the task, but what it is for. "A list of activities" and "a list of activities for a rainy day with twelve nine-year-olds" lead to different answers.

Constraints. Budget, time, length, audience, tools you have or don't have, words or topics to avoid. These are the details a model cannot invent correctly.

Format. Lists or paragraphs, tables, headings, word counts, reading level, tone. OpenAI's own prompting advice for developers says the same: be specific about the context, the outcome, the length, the format and the style you want.

Two extra lines make a briefing work better. Ask the model to tell you what is missing before it begins, and ask it to wait for your first real request. That turns a one-way instruction into a short exchange in which gaps come out early.

A worked example: planning a teen book club

Imagine a public librarian launching a monthly book club for teenagers. Here is her first message, before she asks for anything:

I'd like your help over this conversation, so here is some background first.

Who I am: I'm a librarian at a small public library. I run programs for teens but I've never led a book club.

Goal: I'm starting a monthly book club for ages 13 to 16, from September to December. I want it to feel relaxed, not like school, so that the same teens come back each month.

Constraints: Meetings are 60 minutes, after school, in a room for 15 people. My budget is small, so books must be ones our library already holds in several copies or can borrow through interlibrary loan. No food in the room.

Format: Keep answers practical and short. Use bullet points for lists and a table for any schedule. Write anything meant for the teens in a friendly, plain voice.

Before we start, ask me up to three questions about anything that would change your advice. Then wait for my first request.

A good reply asks pointed questions, perhaps about the teens' reading levels, whether parents need to sign up, or how members will get the books. Once she answers, a request as short as "Draft the first meeting plan" lands on target, because all the context is already in the chat.

Where custom instructions fit

Some of a briefing never changes: your job, your field, how you like answers formatted. OpenAI's custom instructions, introduced in July as a beta, are the place for those facts. You fill in two boxes, one about yourself and one about how ChatGPT should respond, and it considers them in every new conversation.

That leaves the briefing for what belongs to this project only: the goal, the specific limits and anything unusual about the task. The librarian might store "I'm a public librarian; keep answers practical and use tables for schedules" in custom instructions, and keep the book-club details for the first message of the book-club chat. She can also keep one chat per project, so the briefing stays at the top of the right conversation.

What should go in your first message?

Put in what the model needs to do the job well that it cannot guess: your role, the goal and what it is for, the limits, and the format. Then add the two working lines, inviting questions and asking it to wait. A useful length is a short paragraph per part, enough to fill a screen at most. OpenAI's prompting guidance also suggests stating what to do rather than only what to avoid, so "use plain language a 14-year-old can follow" works better than "don't be too complicated."

Leave out what doesn't change the answer, such as the backstory of how the project started, and anything private you would not want stored, like names, addresses or account details. If you notice yourself repeating the same opening lines in every new chat, move those lines into custom instructions and keep your first message for the project in front of you.