How to Write a Good Prompt

The short version

The difference between a mediocre AI response and a great one usually comes down to how you ask. This article is a hands-on guide to writing better prompts, with five real before-and-after rewrites you can learn from and adapt.

If you are new to prompts, read What Are Prompts and How Do They Work? first. This article builds on those foundations with practical techniques you can use immediately.

The four principles of a good prompt

Every effective prompt shares four qualities: it is specific, it provides context, it sets constraints, and it includes an example when the desired output format is not obvious. You do not need all four every time, but the more you include, the better your results will be.

Be specific

Tell the AI exactly what you need. Include the format, the length, the audience, and the goal. The more detail you give, the less the AI has to guess.

Give context

The AI does not know your situation unless you explain it. Tell it who you are, what you are working on, and why you need this particular output.

Set constraints

Tell the AI what to avoid. Constraints prevent the most common problems: responses that are too long, too technical, off-topic, or filled with unnecessary disclaimers.

Show an example

If you have a specific format in mind, show the AI what good output looks like. One example is often worth more than a paragraph of instructions.

Five before-and-after prompt rewrites

The best way to learn prompting is to see real improvements side by side. Here are five common scenarios with weak prompts rewritten into strong ones.

Rewrite 1: Writing a work email

Before:

Write an email about the project delay.

After:

Write a short email (under 150 words) from a project manager to a client named Tanya. The project delivery date is moving from March 1st to March 15th because the third-party payment integration took longer than expected. Tone: professional, apologetic but not grovelling. Include a revised timeline and one sentence about what we are doing to prevent further delays.

Why it works: The rewritten prompt specifies the audience, the reason, the tone, the length, and the structure. The AI has almost no room to go off track.

Rewrite 2: Summarising a long document

Before:

Summarise this document.

After:

Summarise the following document in exactly five bullet points. Each bullet should be one sentence. Focus on decisions made and action items assigned. Ignore background context and introductions. Write for a busy executive who has 30 seconds to read this.

Why it works: "Summarise this document" is one of the most common AI prompts, and one of the least effective. The rewrite tells the AI the format (five bullets, one sentence each), the focus (decisions and action items), what to skip (background), and the audience (busy executive).

Rewrite 3: Explaining a technical concept

Before:

Explain Kubernetes.

After:

Explain what Kubernetes does in three paragraphs for someone who manages a software team but does not write code. Use a simple analogy in the first paragraph. In the second paragraph, explain the main problem it solves. In the third, give two real situations where a team would benefit from using it. Avoid jargon. If you must use a technical term, define it in parentheses.

Why it works: The rewrite defines the audience (non-technical manager), the structure (three paragraphs with specific purposes), and the language rules (no jargon, or define it). The AI produces a response that the reader can actually use.

Rewrite 4: Generating ideas

Before:

Give me some marketing ideas.

After:

I run a 10-person B2B software company that sells an AI security product to engineering teams. Our budget for the next quarter is $5000. Suggest five low-cost marketing ideas that do not require a dedicated marketing hire. For each idea, include: what it is, estimated cost, time to execute, and one example of a similar company that has done it successfully.

Why it works: Without context, "marketing ideas" could mean anything from a Super Bowl ad to a hand-written letter. The rewrite provides the company size, industry, budget, headcount constraints, and a structured format for each suggestion.

Rewrite 5: Reviewing code

Before:

Review this code.

After:

Review the following TypeScript function for: (1) correctness: does it handle edge cases like empty arrays and null values? (2) readability: are variable names clear and is the logic easy to follow? (3) performance: are there obvious inefficiencies? Format your review as a numbered list of specific issues. For each issue, quote the line, explain the problem, and suggest a fix. If there are no issues in a category, say so explicitly.

Why it works: "Review this code" invites a generic response. The rewrite tells the AI exactly what to evaluate, how to structure the feedback, and what to do when there are no issues. This produces actionable, specific feedback instead of vague observations.

Common patterns that work

Beyond the four principles, a few prompt patterns consistently produce better results:

"Act as..." Giving the AI a role shapes its tone, vocabulary, and depth. "Act as a senior product manager reviewing this feature spec" produces very different output from "Review this document."

"Here is an example of what I want..." Showing one good example of the output format (sometimes called one-shot prompting) is faster and more reliable than describing the format in words.

"Do not include..." Explicitly excluding things you do not want (disclaimers, introductions, bullet points, analogies) prevents the AI from padding its response.

"Step by step..." Asking the AI to work through a problem step by step often produces more accurate results for tasks that require reasoning or calculation.

"Before answering, list your assumptions..." This forces the AI to surface what it is guessing about, giving you a chance to correct misunderstandings before it writes the full response.

When good prompting is not enough

Better prompts get you better responses, but they do not solve every problem. AI models can still produce incorrect information (see What Are AI Hallucinations and Why Do They Happen?), and they cannot take real-world actions on their own, unless they are connected to tools as AI agents.

As you start using AI not just for generating text but for taking actions like sending emails, making purchases, and managing data, the stakes go up. The next article explains how the major AI models compare so you can choose the right one for your needs.

Key takeaways

• Good prompts are specific, provide context, set constraints, and include examples.
• Small changes in how you ask can dramatically change the quality of what you get back.
• Use the before-and-after rewrites as templates for your own prompts.
• Patterns like "Act as...", "Do not include...", and "Step by step..." consistently improve results.
• Always review AI output, even from well-crafted prompts. Better prompts reduce errors but do not eliminate them.

Next up: ChatGPT vs Claude vs Gemini: What Is the Difference?