What Is a Prompt? How to Write Better AI Prompts

A prompt is the instruction, question, or task you give an AI system. To write a useful prompt, explain the task, provide relevant context, define the output you want, and add constraints such as audience, length, or format. Then review the response instead of assuming it is correct.

Key takeaways

• A prompt can be a question, command, example, or combination of instructions and context.
• Good prompts make the task, audience, desired output, and boundaries clear.
• A reliable five-part prompt includes the role, task, context, output format, and constraints.
• The best way to improve a weak answer is to diagnose what was missing and revise one part of the prompt at a time.
• AI output is a draft until you verify important facts, calculations, citations, and recommendations.

What Is a Prompt in AI?

A prompt is the input you provide to an AI tool to guide its response. It might be a short question such as What is a prompt?, a detailed set of instructions for drafting a report, or a request to analyze information you paste into the conversation.

Prompts are not limited to typed questions. An uploaded document, spreadsheet instruction, image-editing request, voice command, or request to rewrite text can also function as a prompt. In each case, you are telling the AI what to do with the information available to it.

Most generative AI tools produce new text, images, code, summaries, or other content from your instructions. If you are new to the broader technology, this plain-English guide to generative AI explains how these systems create responses and why their output needs review.

The quality of a prompt does not guarantee a correct answer. AI systems can misunderstand vague instructions, overlook relevant context, make unsupported claims, or follow the wrong interpretation of an ambiguous request. A good prompt reduces avoidable confusion; it does not replace human judgment.

For example, compare these two requests:

• Write an email about the meeting.
• Draft a concise email to five project stakeholders explaining that Thursday's launch meeting is moving to Friday at 10 a.m. Include the reason, a request to confirm attendance, and a professional but friendly tone.

The second prompt gives the AI a defined task, audience, facts to include, tone, and expected content. That gives it a much narrower target.

What Makes a Good Prompt?

A good prompt is clear enough that another person could understand the task without asking several follow-up questions. It also gives the AI enough information to produce a useful first draft.

Clarity

Use direct language and put the main task near the beginning. Replace broad instructions such as make this better with an observable action such as rewrite this introduction for a nontechnical audience in 120 words.

Context

Context explains why the task matters and who will use the result. Include the relevant facts, source text, definitions, or business situation. Without context, an AI tool may fill gaps with assumptions.

Specific output

Tell the tool what the answer should look like. You can request a table, checklist, email, outline, JSON object, set of headings, or numbered recommendations. Specify the level of detail and whether the response should use plain language.

Useful constraints

Constraints help the AI stay within practical boundaries. They can include word count, reading level, tone, deadline, allowed sources, formatting rules, or things to avoid. Add constraints that matter, but do not overload a simple request with unnecessary instructions.

A clear success test

Ask for an outcome you can evaluate. For example, a useful success test might be Every recommendation must include an owner and a next action or Separate confirmed facts from assumptions. This is more useful than saying give me a high-quality answer.

There is no magic phrase that makes every response better. Prompting is closer to writing a clear brief than finding a secret command. For more practical advice, the AI tips and best practices collection covers related habits for everyday use.

The Five Parts of an Effective Prompt

A simple framework for how to write AI prompts is to include five parts. You do not need every part for every request, but the framework gives you a reliable starting point.

1. Role or perspective

Tell the AI what perspective would be useful. Examples include act as a copy editor, act as a project coordinator, or explain this like a patient tutor.

A role can influence vocabulary and priorities, but it does not give the AI real-world authority or access to private knowledge. Do not treat a role instruction as proof that the response is expert-reviewed.

2. Task

State exactly what you want done. Use an action verb such as summarize, compare, classify, rewrite, extract, brainstorm, plan, or explain.

If there are multiple tasks, number them. This makes it easier to check whether the AI completed each one.

3. Context and source material

Provide the details the AI needs. You might include the intended audience, business goal, relevant constraints, existing draft, customer segment, or pasted data.

If you want the AI to use only information you provide, say so explicitly. You can also ask it to identify missing information instead of guessing.

4. Output format

Define the shape of the answer. Examples include:

• A table with columns for issue, evidence, owner, and next step
• Five bullet points followed by a 100-word summary
• A professional email with a subject line
• A checklist ordered from highest to lowest priority
• A response in valid JSON with named fields

5. Constraints and quality checks

Finish with boundaries and a way to test the result. You might specify length, tone, reading level, prohibited claims, required sections, or a request to flag uncertainty.

Here is a reusable template:

Act as a [useful role or perspective].

Task: [describe the specific action you want].

Context: [explain the audience, goal, situation, and relevant source material].

Output: [describe the format, sections, length, and tone].

Constraints and checks: [state what to include, avoid, verify, or flag as uncertain].

You can often write a good prompt in one paragraph. The five parts are a checklist, not a requirement to make every prompt long.

Before-and-After Prompt Examples

The fastest way to understand what makes a good prompt is to compare a vague request with a more useful one.

Example 1: Turning notes into a project update

Weak prompt:

Write a project update from these notes.

This does not identify the audience, desired tone, structure, or what to do with missing information.

Stronger prompt:

Act as a project coordinator. Turn the notes below into a weekly update for senior stakeholders.

Include four sections: overall status, completed work, risks and decisions needed, and next week's priorities. Use concise bullet points and a neutral, direct tone. Do not invent dates, owners, or progress. If a detail is missing, mark it as [needs confirmation]. Keep the update under 250 words.

Notes:
[paste notes here]

The improved version gives the AI a role, task, audience, format, length, and accuracy boundary.

Example 2: Rewriting customer-facing copy

Weak prompt:

Make this sound better.

Stronger prompt:

Rewrite the product description below for first-time customers who are not technical. Keep the central claims unchanged, remove jargon, and use a helpful, confident tone without exaggeration. Return two versions: one under 50 words and one under 100 words. List any claim that needs factual or legal review.

Original description:
[paste copy here]

This is a practical example of how to write a good prompt: define the audience, preserve important meaning, set limits, and ask for a review flag.

How to Improve a Prompt That Gets a Poor Answer

A poor response usually points to a missing instruction, missing context, or unclear success criteria. Instead of starting over randomly, diagnose the failure.

A five-step troubleshooting process

1. Describe the problem. Is the answer too general, too long, inaccurate, poorly formatted, or aimed at the wrong audience?
2. Find the missing information. Add the audience, purpose, source material, examples, or definitions the AI lacked.
3. Make the output testable. Specify sections, length, number of options, required fields, or ranking criteria.
4. Add a boundary against guessing. Ask the AI to identify uncertainty, list assumptions, or request missing details.
5. Run a smaller revision. Ask it to revise the specific section that failed instead of regenerating everything without explanation.

For example, if an AI gives generic marketing ideas, do not simply say be more specific. Add the customer, product, channel, budget or resource limits, desired outcome, and evaluation criteria.

You can also ask the AI to improve the prompt itself:

Review the prompt below before answering it. Identify up to three ambiguities or missing details that could lead to a poor result. Then provide a revised prompt and explain what you changed. Do not complete the underlying task yet.

Prompt:
[paste your draft prompt here]

Be careful with long conversations. Earlier instructions and assumptions may affect later responses. If the task changes substantially, restate the goal and relevant context in a fresh prompt or conversation.

How to Write Prompts for Different Work Tasks

The core method works across jobs, but each task benefits from different details.

Work task	Include in the prompt	Useful output request
Summarizing	Audience, source, purpose, what to omit	Executive summary plus key decisions and open questions
Drafting	Audience, goal, tone, facts, call to action	Draft with subject line, headings, or alternatives
Analysis	Dataset or source, definitions, comparison criteria	Findings separated from assumptions and limitations
Brainstorming	Objective, audience, constraints, evaluation criteria	Grouped ideas ranked by feasibility or impact
Planning	Desired outcome, deadline, dependencies, resources	Ordered steps with owners, risks, and next actions
Editing	Intended reader, house style, claims to preserve	Revised copy plus a short list of major changes

Summarizing and extracting

Tell the AI what matters, not just that you want a summary. Ask it to separate decisions, facts, risks, and questions. If the source is long, identify the sections or pages that matter most and ask it to flag information it cannot locate.

Drafting emails and documents

State who will read the draft and what you want them to do. Include facts that must remain accurate, the desired tone, and a length limit. Ask for a subject line or a version suitable for a particular channel if needed.

Brainstorming

Good brainstorming prompts define the problem and the boundaries. Ask for a specific number of ideas, group them by approach, and request a short reason or trade-off for each. If you need originality, explain what common ideas to avoid, but still review the suggestions for feasibility.

Working with data

Describe column meanings, units, date ranges, and the decision you are trying to make. Ask the AI to distinguish calculations from interpretation. Never paste confidential, personal, or restricted data into a tool unless your organization permits it and the tool is approved for that information.

Creating repeatable workflows

Turn a successful prompt into a template with replaceable fields such as audience, date range, product, and source text. Test the template on several realistic examples before making it part of a team process. If your organization is rolling out AI broadly, AI training for teams can help establish shared practices and review standards.

How to Check AI Output for Accuracy

Treat AI output as a draft, even when it sounds confident and polished. Verification is especially important for medical, legal, financial, employment, safety, security, customer, and policy-related work.

Use this review checklist:

• Check factual claims: Verify names, dates, figures, definitions, quotations, and references against reliable source material.
• Check calculations: Recalculate important numbers with an approved tool or the original spreadsheet.
• Check completeness: Compare the response with the prompt and confirm that every requested section is present.
• Check assumptions: Look for details the AI inferred rather than received from you.
• Check tone and audience: Remove language that could be misleading, insensitive, overconfident, or inappropriate.
• Check confidential information: Make sure the output does not expose personal, proprietary, or restricted content.
• Check policy requirements: Follow your company's AI, privacy, records, and review policies.

A useful verification prompt is:

Audit the draft below against the source material that follows. Create a table with three columns: claim or statement, supporting source, and status. Mark each item as supported, unsupported, contradicted, or unclear. Do not silently correct the draft; list recommended corrections separately.

Draft:
[paste draft]

Source material:
[paste approved source]

The audit itself also needs review. AI can miss errors or incorrectly label a claim as supported. For high-stakes work, have a qualified person check the source and final decision. For more guidance on responsible use, see the AI Risks: Use AI Safely at Work course.

Prompting becomes easier when you treat each request as a small specification: define the task, supply context, describe the output, set boundaries, and check the result. Start with the shortest prompt that contains the necessary information, then improve it based on the response.

Next step

Build these habits through practice in the ChatGPT course, where you can apply the five-part prompt method to common work tasks.