The 5 Prompts Every Recruiter Should Have

Your Input Determines the Output

Think about the worst briefing call you've ever had. A client who sent you a two-line job spec and expected a shortlist of six by Friday.

You know what happens when you build a search from that. You end up guessing. You optimise for availability rather than fit. You send candidates you're not confident in and hope the client fills in the blanks.

That's exactly what happens when you under-brief AI.

The model has no knowledge of your desk, your candidate relationships, your client culture, or what separates a strong hire from a safe one in that sector. It fills every gap you leave with the most average, most common version of what you asked for.

Most people treat it like a search engine. Type a question, get an answer. But it's closer to a very fast, very literal junior who has read everything and knows nothing about your specific situation. The quality of what comes back is determined entirely by what you put in.

The recruiters getting useful output aren't more technical. They brief well. They already do it every day on client calls, intake meetings, candidate debriefs. They just applied the same discipline here.

How to Write a Prompt That Produces Something Usable

Before the five use cases, it's worth understanding the structure behind a prompt that works. Because the structure matters more than the specific words.

Give it a role before you give it a task

Start every prompt by telling the AI who it is very specifically. What you do is, putting the AI in a certain role to perform better. 

Here is an example: 

"Act as an experienced executive search consultant specialising in CFO placements for PE-backed businesses" will produce something fundamentally different from "act as a recruiter." The more precisely you define the lens, the more calibrated the output.

AI models adjust their framing, vocabulary, and level of sophistication based on the role you assign. 

Front-load the context

The model has no memory between sessions. No knowledge of your client, your candidate pool, your market, or what made the last three hires fail. 

Unless you put it in the prompt, it doesn't exist.

Context means: what is the role, what level, what sector, why does this company exist, what kind of person has actually thrived there, who are you trying to reach, what do they care about, what tone does the output need to carry.

The same way you'd brief a headhunter before a client meeting. You'd give them everything relevant, not just the job title. Do that here.

Be precise about what you want and what you don't

"Write me a job description" is a topic. A prompt is a specification.

Tell it the word count, the tone, what to lead with, what to avoid. "Write a job description of around 400 words. Lead with what makes this opportunity interesting, not with responsibilities. Tone should be direct and credible. Don't use phrases like 'fast-paced environment' or 'passionate self-starter.'"

Constraints are instructions. Leave them out and the model fills the gaps with assumptions. Most of those assumptions will be wrong for your specific situation.

Tell it what format you want back

Do you want a block of text? Sections with headers? Three options to choose from? A structured table?

If you don't say, you'll get whatever default format the model defaults to. For something like a scorecard or interview guide, format is half the value. Specify it explicitly.

Break complex tasks into steps

For anything multi-layered, don't ask for everything in one prompt. 

Instead, build a sequence:

"First, pull out the key themes from these interview notes. Then organise them under these four headings. Then write a two-paragraph candidate summary in a professional but readable tone."

One instruction at a time produces cleaner output than one instruction covering five things at once. Same logic applies when you're briefing a person.

Ask it to review its own output

This is the step most people skip. After the model generates something, follow it with a second prompt.

"Review what you just wrote. Flag anything that sounds generic or could apply to any company in any sector. Then rewrite those sections to be more specific to the context I gave you."

You can also pressure-test it against your own criteria. "Does this job description say anything that a competitor posting for the same role wouldn't say? If not, what's missing?"

Almost all AI output is a first draft. Treating it as a finished product is where quality falls apart. A self-review prompt gets you meaningfully closer to something usable without any extra effort on your end.

The Mistake That Quietly Kills the Quality

Most recruiters improve their prompts fast once they understand the structure. That part is fixable.

The mistake that's harder to catch is trusting the output without interrogating it. Pasting it into a document and sending it because it looked plausible.

Three specific habits that erode results and, in some cases, create real professional risk:

Pasting CVs and candidate data into public LLMs

ChatGPT's free tier uses your conversations to train its models by default. Candidate names, contact details, employment history — that's personally identifiable information covered by GDPR. Pasting it into a public tool is not a grey area.

Work with anonymised information. Describe the candidate's background, seniority level, and key competencies without naming them. You'll get equally useful output and you won't be creating a compliance problem.

Publishing without reading it properly

AI doesn't know your client. It doesn't know the company went through a restructure last year, that the last hire left after six months, or that the hiring manager means something very specific when they say "commercial." You do.

The model gives you a draft. Your judgment is what makes it accurate. Read everything before it goes anywhere. If a sentence could apply to any company, it probably will, and it probably shouldn't.

Writing thin prompts and blaming the tool

"Write me an InMail for a finance director" will produce an InMail that reads like every other InMail a finance director has ever ignored.

The tool isn't underperforming. The brief is underspecified. When you give it context, constraints, and a clear output format, the gap between what comes back and what you'd write yourself gets much smaller.

The bar for a prompt is roughly the same as the bar for a solid intake call. You already clear it every day.

The 5 Use Cases Worth Building Prompts For

These are the tasks where a well-structured prompt makes the biggest difference between output you can use and output that wastes your time.

Job Description

Job descriptions are where most recruiters first try AI and where most get burned. The output is always competent. It's rarely true.

The issue is almost always context. A prompt that only gives the AI a job title and a list of responsibilities will produce a JD that sounds like every other JD for that role. It needs to know why the position exists, what the company is actually like to work for, what kind of person has done well there, and what tone should come through.

Load that in and the output starts to sound like a real opportunity rather than a template someone filled in at speed.

InMail / Candidate Outreach

A message that reads like it was written by a tool is a message that gets ignored. The irony is that using a tool properly is how you avoid that.

A good outreach prompt tells the AI who you're writing to specifically: their background, their current position, what they're likely working on, and what angle on your opportunity is most likely to be relevant to them. Not just the title. The reason this particular person might care.

The output still needs your review before it goes anywhere. But it will be much closer to personalised than anything written from a blank page.

Scorecard and Interview Guide

Structure is where AI earns its keep. Scorecards and interview guides have a defined shape, and the model is good at building to a spec when you give it one.

Tell it the role, the competencies you're assessing, the format you want (weighted scoring, rating scales, behavioural question sets), and the seniority level. Specify what you want in each section. It builds the scaffold. You bring the judgment about what actually matters for that hire.

Candidate Summary

This is probably the most underused application on the list.

If you take any notes during a candidate conversation, even rough ones, AI can turn them into a structured, readable summary in under a minute. The prompt needs to specify the format, the headings, what to prioritise, and the tone you're writing for (client-facing and professional, not a sales pitch).

The model handles the structure and the prose. You check that it's accurate and that nothing got lost in translation. It takes real time off every write-up without changing what the output sounds like.

Personalised Outreach

When you have both a candidate's profile and a scorecard for the role you're presenting them for, you can prompt the AI to identify exactly where they align and write a pitch that leads with those specifics.

This is where a well-structured prompt produces something that doesn't read like outreach at all. It reads like an argument. Why this person, for this role, based on what you actually know about both.

Give it both inputs. The candidate context and the role criteria. Without both, it defaults to generic. With both, it has something to work with.

One Last Thing

You've been doing the hard part of this for years.

Every intake call where you pushed back on a vague brief.

Every time you asked a candidate what they actually want rather than accepting the first answer.

Every shortlist where you wrote a clear rationale instead of just attaching CVs.

That's the skill. Writing a good prompt is just applying it to a different audience. The recruiters who get the most out of AI aren't the ones who spent the most time learning the tools. They're the ones who already knew how to be precise about what they needed.

You're probably closer than you think.