How to Write a Good AI Prompt: A Step-by-Step Guide

Why writing a good prompt improves your outreach

A prompt is the instruction you give to an AI system such as ChatGPT. For recruiters, this means a clear brief that gets AI to write a message that's personal, relevant and short. For example: "Write an informal LinkedIn invite for a backend developer who's active on GitHub."

Many prompts lack context or clarity. Think of instructions such as "Write a fun follow-up." AI then has no idea what the message should achieve, who the recipient is, or how long it should be. The result is unfocused copy that's too long, too formal or too vague. Effective AI support needs precision.

Good prompts give AI something to work with. You feed in the right goal, the right tone and relevant information. This produces output that performs better and needs less correction. That's why it pays to work from a fixed structure.

Step-by-step plan for writing a good prompt

Use these six steps to consistently build prompts that work. It also helps your team use AI the same way.

1. Goal

Give one instruction. For example: "Write a short, informal invitation." Avoid multiple goals in a single prompt. This keeps the output focused.

2. Context

Who is the message for, and why now? Use profile information as your hook. For example: "The data engineer works at Adyen and posted yesterday about scalable systems."

3. Task

What should AI take into account? Think: ask a question, reference shared history, or give a reason for reaching out. Describe this clearly.

4. Constraints

Set boundaries on style, tone and length. For example: "Use a friendly tone and a maximum of 250 characters. Plain, simple English."

5. Output format

Be specific about what you expect. For example: "Give 2 versions of an invitation and 1 InMail. All in a natural tone and without jargon."

6. Comparing variants

A good prompt leaves room for A/B tests. By asking for multiple options, you can measure what works best and reuse that later.

You can lock in this approach through fixed instructions and templates. That way, everyone on your team uses AI in the same way, with control over tone and format.

Example prompt for invites and InMails

LinkedIn has several types of messages. A connection request needs to stand out within 250 characters. An InMail gives you a little more room for context. Below you'll find an example prompt for each type, together with example copy.

1. Prompt for a connection request

"Write a short connection message to a frontend developer who just posted about UX. Keep it friendly, ask one question, maximum 250 characters. B1 language level."

Example copy

"Great post about UX last week! How do you apply that within your team? Would love to connect."

2. Prompt for an InMail

"Write an InMail in natural English for a data engineer at Adyen. Open with a shared employer. Tone: curious. End with an invitation to talk. Maximum 500 characters."

Example copy

"Noticed we both worked at ING at some point. Curious how you see your growth path. I work with data teams myself now. Fancy a quick call about it?"

As you can see, the tone always matches the channel. A connection request has to land fast. An InMail builds up more calmly towards the ask. Want to use prompts directly inside your workflow? Take a look at how prompts fit into the LinkedIn flow.

Writing a good prompt also means personalising at scale

Anyone producing a lot of AI messages wants to combine speed with personalisation. You can do that with prompt variables such as {role}, {reason} or {question}. This lets you tailor copy without rewriting everything from scratch. For example:

"Write an invitation to a {role} who just posted about {topic}. Keep it personal and add an open question about {hook}."

Fill in these variables each time with profile information. Pay attention to things like recent posts, shared connections or employers. The more concrete the input, the more relevant the message.

You also want the tone to stay consistent. So state whether you're after a friendly, formal or curious tone. That keeps your style recognisable, even with several team members involved. This approach suits anyone who wants to work with AI at scale. Read more about the application for agency recruiters.

Test, improve and save what works

Writing a good prompt is more than a one-off attempt. You improve the process by testing systematically. Ask for several versions of the copy and measure which one performs best. Think of the percentage of connections that get accepted, or the number of InMail replies.

You'll learn, for example, that short opening lines with an open question work better than generalities. Or that an invitation with a profile hook stands out more than one without. Save the top-performing versions as a template and build your own library of effective prompts.

You can see a concrete example in this real-world case study. There you'll discover how small changes to prompts led to more replies and fewer edits afterwards.

Common mistakes when writing a good prompt

Use this checklist to avoid typical mistakes and make your prompts stronger:

• No length limit: messages become too long or too formal.
• Multiple goals at once: AI loses focus in the output.
• No clear message or call to action.
• No example output requested: hard to judge whether it works.
• No tone of voice specified: AI switches between styles.

Start simple and scale up

You don't have to tackle everything at once. Start with a clear prompt for one role or audience. Let AI write two versions: one for an invitation, one for an InMail. Test them for a week. Based on the results, pick the best-performing prompt and build on it further.

This is how you build, step by step, an approach that stays scalable and personal. Want to get started within your own team? Ask for a short demo or ask your question. We're happy to show you how this process makes your outreach smarter.

Frequently asked questions

What makes a good prompt when using AI in recruitment?

A good prompt contains one clear goal, context about the recipient, a concrete task, style and length constraints, and instructions on the desired output format. The more specific the input, the more usable the AI output.

How long should a prompt be for a LinkedIn invitation?

A clear prompt for a LinkedIn connection request is often just one or two lines. Give the type of role, the profile hook, the tone (for example friendly), and cap the length of the final result at 250 characters.

Why is stating tone in a prompt important?

Tone creates consistency. Without instructions, AI can sound formal or distant. By specifying "friendly" or "natural", the output matches your message and audience better.

What can you do with prompts that perform well?

Save top-performing prompts as templates. Label them by audience or message type. That way you can reuse them, share them easily within your team, and adjust faster based on test results.