Your First Prompt

Your first prompt should feel less like a search query and more like a short request to a capable assistant. If you asked a coworker "help with this" and gave them no document, audience, deadline, or desired format, they would have to guess. AI assistants have the same problem, except they can guess fluently.

The beginner move is to make the invisible parts visible. Say what you want done. Say why you need it. Paste the source material if the answer depends on it. Then describe the answer you want back: a draft, checklist, table, outline, summary, critique, plan, or set of questions.

The five parts of a useful first prompt

1. Task: the verb. Summarize, rewrite, compare, plan, explain, critique, extract, brainstorm, or turn this into something else.
2. Context: the situation around the task. Who is the audience? What is the goal? What happened before this request?
3. Source: the text, notes, document, data, links, or facts the model should use. If you do not provide it, the model may rely on general knowledge.
4. Constraints: the boundaries. Length, tone, format, reading level, tools to avoid, assumptions to avoid, and what not to change.
5. Output shape: what the response should look like. A table, three bullets, a draft email, a step-by-step plan, or questions first.

Before and after examples

Make this email better.

Rewrite the email below for a busy customer who is frustrated about a delayed shipment.

Goal: sound calm, specific, and accountable.
Constraints: under 150 words; do not overpromise a refund.
Output: final email only.

Email:
[paste draft]

Explain APIs.

Explain APIs to someone who knows spreadsheets but has never coded.

Use a simple analogy first, then a concrete example involving a weather app.
Avoid jargon until the final section.
Output: 5 short sections with one quiz question at the end.

Plan my launch.

Help me plan a launch checklist for a small newsletter.

Context: I have 2 weeks, no paid ads, and an audience of 900 subscribers.
Goal: launch a paid tier without annoying free readers.
Output: table with owner, task, deadline, and risk.
Ask me up to 3 questions first if the plan depends on missing details.

Do not start with a persona unless it helps

"Act as a marketing expert" can help when the role changes the answer. But beginners often use role prompts as a substitute for details. A role without a task is vague. A task with context is useful. If you use a role, pair it with the real goal: "Act as a skeptical editor and identify unsupported claims in this draft."

Ask for questions before answers

A reliable beginner habit is to let the model stop before it guesses. Add: "If anything important is missing, ask me up to three questions before answering." This works especially well for plans, strategy, medical-adjacent research, legal-adjacent research, financial decisions, public claims, and anything involving private details you have not yet provided.

When facts matter, give the facts

AI tools can produce confident mistakes. OpenAI explicitly tells users to verify important information from reliable sources. If you are asking about your own project, document, school assignment, company policy, product, or local situation, paste the relevant material or link it if the tool can browse. Then ask the model to separate what is in the source from what it is inferring.