How to Write a Job Brief for AI Recruiters to Screen Against
AI screening is only as good as the brief behind it. How to turn a vague job description into criteria a model can apply consistently and a hiring manager can defend.
11 November 2026 · 7 min read · XenonLabs
Almost every disappointing result from AI recruiters traces back to the same cause, and it is not the model.
It is the job description. Most job descriptions are written to attract applicants, not to assess them. They list twelve "essential" requirements that are not essential, describe the company in warm generalities, and never say what actually separates someone who thrives in the role from someone who struggles. A human recruiter compensates by asking the hiring manager what they really want. A screening system only has the page.
This guide is about writing the brief that the system — and, frankly, your own recruiters — can work from.
The job description is not the brief
A job description is outward-facing marketing. A screening brief is an internal document that answers a different question: given a stack of applications, how do we decide who moves forward?
If you cannot answer that question in writing, nobody can answer it consistently — not a model, and not two recruiters working the same pile. Writing the brief is therefore useful even before any technology is involved. It forces the conversation with the hiring manager that usually happens only after the first shortlist disappoints.
Separate what is required from what is preferred
Start with the most common failure: everything is labelled "required".
Go through the requirements and sort each into one of three groups.
- Must have. Without this, the person cannot do the job or is not legally permitted to. A licence, a right to work, a specific technical skill with no substitute.
- Strongly preferred. It makes the person substantially more likely to succeed, but a good candidate might lack it and compensate.
- Nice to have. A tiebreaker, nothing more.
Describe evidence, not labels
"Strong communicator" and "team player" tell a model — or a person — nothing. What would you actually see in an application from someone who has them?
Rewrite each criterion as the evidence you would accept:
- Instead of "strong stakeholder management": "has led a project involving several departments with competing priorities, and can describe the outcome".
- Instead of "analytical": "has used data to change a decision, and states what the data showed".
- Instead of "leadership experience": "has managed people directly, or led a team without formal authority, and says how many and for how long".
State what you are not looking for
This is the step most teams skip, and it is where the most important protections live.
Explicitly list what must not influence screening. That normally includes name, age, gender, nationality, photo, marital status and anything else that is irrelevant to doing the job or legally protected in your market. It can also include proxies — a particular university, a gap in employment, a postcode — unless you have a defensible reason for each.
Write it into the brief as a standing rule. It makes your intent clear, it gives the system something to be configured against, and it gives you a reference point when you later audit outcomes.
Add the context a candidate would not guess
The most useful sentences in a brief are often the ones that do not appear in the job description.
- What does this role look like in the first ninety days?
- What has gone wrong with previous hires in the same seat?
- Is the team remote, hybrid, or in one office — and is that negotiable?
- What is the salary range, and is it flexible for the right person?
Decide who decides when AI recruiters screen
No brief is complete until it says what the output is for.
The principle we recommend is simple: the system screens and ranks, with its reasoning attached; a person decides who advances. Your brief should therefore say who that person is, what a reviewer should look at (the score, the evidence behind it, and anything flagged as uncertain), and what happens to borderline applications.
Borderline cases deserve a deliberate rule. Common options are to send them to a human review queue, or to ask the candidate a clarifying question before deciding. What matters is that "close but uncertain" does not silently become "rejected".
Test the brief before you rely on it
Treat the first run as a test of the brief, not of the technology.
- Take a previous hire you know turned out well, and a previous hire who did not. Run their original applications through the brief. Do the scores point in the direction you now know to be right?
- Read the reasoning attached to a dozen scores. Where it is wrong, it will usually be wrong because the brief was ambiguous.
- Ask the hiring manager to shortlist blind from the same pool, and compare. Disagreements are the most valuable information you will get, because they reveal criteria nobody had written down.
- Revise the brief and run it again.
Keep measuring after launch
A good brief at launch can quietly degrade as the role, the market and the applicant pool change.
Track the stages that follow screening: how many shortlisted candidates the hiring manager actually wanted to interview, how many progressed, how many accepted offers, and how the screening outcomes break down across groups. Where a result looks off, go back to the brief. It is almost always the cheapest thing to fix.
Be aware that employment-related AI is subject to evolving rules in many markets, including requirements around disclosure, audits and human oversight. Involve your employment counsel when you design the process rather than after launch.
A checklist you can use today
- Must-haves are short, and genuinely essential.
- Every criterion is phrased as evidence you would accept.
- Preferred and nice-to-have items are separated.
- Non-criteria are listed explicitly.
- The ninety-day picture and past hiring mistakes are written down.
- The decision-maker and the borderline rule are named.
- You have tested the brief against known good and poor hires.
If you would like to see how a brief turns into scored, explained shortlists, book a HirePro demo and bring a live role.