Prompting basics that work with any AI assistant

What a prompt is

A prompt is simply the message you give an AI assistant. There is no magic phrase. What works across nearly every assistant is the same thing that works when briefing a person: be clear about what you need, give the background, and say what done looks like.

A handy checklist has six parts: role, task, context, format, examples, and checking. You will not need all six every time, but knowing them lets you fix a weak prompt quickly.

Role and task

The role tells the assistant what perspective to take. 'Act as a patient tutor for a beginner' or 'act as a careful proofreader' nudges its style and priorities. A role is a steer, not a credential. An assistant playing a tutor is not a real teacher, so it does not make the facts any more reliable.

The task is the specific thing you want done, stated as a plain instruction: summarize, compare, draft, outline, critique, translate. Use one clear verb. If you have three tasks, say so and number them, or do them one at a time.

Context and format

Context is everything the assistant would otherwise have to guess: who the audience is, what you already know, what you have tried, and what limits apply. More relevant context usually means a more useful answer. Irrelevant context just adds noise, so include what matters.

Format is the shape of the reply. Ask for exactly what you will use: 'five bullet points, each under fifteen words,' 'a table with three columns,' or 'a short email with a subject line.' If you plan to paste the result somewhere, say so.

Examples

Showing is often stronger than describing. If you want a certain style, paste one or two short examples of what you like and say 'match this tone.' If you want a particular structure, give a small sample of it.

Use examples you are allowed to share, and keep them short. One good example usually does more than a paragraph of explanation, but the assistant may copy it too closely, so say which qualities to borrow.

Checking the result

Prompting does not end when the reply arrives. Read it against your original request. Did it follow the format? Did it invent anything? Are the facts ones you can confirm elsewhere?

You can build checking into the prompt itself: 'If you are not sure about something, say so instead of guessing,' or 'list any assumptions you made.' These lines do not make errors impossible, but they make them easier to spot.

Finally, refine in rounds. Change one thing at a time so you can tell what made the difference. Keep a short note of prompts that worked well for you, because you will reuse them.

Different assistants respond a little differently to the same words, so a prompt that works well in one may need a small adjustment in another. Treat your saved prompts as starting points and expect to tune them.

Practice

1. Take a prompt you have used before and rewrite it with all six parts. Mark which parts were missing originally.
2. Run the old and new versions in the same assistant and compare. Write one sentence about what changed.
3. Add the line 'If you are unsure, say so and list your assumptions' to a new prompt and see how the reply differs.

Check yourself