How to Write a Role Brief an AI Recruiter Can Work From

Why the brief decides whether the agent is any good

When a pilot disappoints, the shortlist is usually not the problem. The brief is. An AI recruiting agent turns your brief into structured criteria, then uses those criteria for every search, every score and every outreach message it writes. A vague input produces a vague search, which produces a plausible looking shortlist of people who are roughly in the right industry and wrong for the job.

Human recruiters mask bad briefs. Give an experienced recruiter three sentences and they will fill the gaps from context, ask a hiring manager two clarifying questions in the hallway, and quietly correct your unrealistic seniority band. An agent does not do that. It takes what you wrote literally and executes at volume, which means an ambiguity that a human would have smoothed over gets multiplied across two hundred candidates. That is a real limitation, and the fix is cheap: be specific once, at the start.

What goes in a role brief

Six parts, in this order. The whole thing should fit on one screen.

Write must haves that actually disqualify

The single most common failure is a requirements list where nothing is truly required. If your must haves are eight years of experience, a computer science degree, Kubernetes, Go, distributed systems, fintech background, startup experience and strong communication, you have described a person who mostly does not exist. Worse, you have told the agent that all eight matter equally, so a candidate missing the one thing you truly cannot compromise on scores about the same as a candidate missing something you would happily trade away.

The discipline is to ask, for each line: would I reject an otherwise outstanding candidate for missing only this? If the answer is no, it is not a must have, it is a strong signal, and it belongs in section three where it can shape the ranking instead of gating the pool. Most teams find they have two or three genuine must haves. That is normal and it produces far better shortlists than a wish list does.

Be specific about the shape of experience too. "Five years of Python" is a weak criterion because it counts time. "Has owned a production service with real users, in any backend language" is strong because it describes a capability, and it lets the agent surface a strong Go engineer you would have been glad to see and your keyword filter would have dropped.

Use candidate language, not internal language

Search works on the words people use to describe themselves. If your title is "Growth Engineer II" and the market calls that role a full stack engineer or a growth marketer depending on the company, write both. If your industry has jargon that only shows up on your own careers page, translate it. The agent will search on the vocabulary you give it, and a brief written in internal shorthand narrows the pool to people who already worked somewhere like you.

The same applies to seniority. Titles inflate at different rates across company sizes, so a Staff Engineer at a two hundred person startup and at a large enterprise are frequently not the same person. Describe the scope instead: how many people they will work with, what they will own, whether they will be the first hire in the area or the fifth. Scope translates across companies in a way that titles do not.

Keep the criteria defensible

Because the brief becomes the scoring criteria, it also becomes the thing you would show if a candidate or a regulator asked how the process worked. That is a good reason to write it carefully, quite apart from shortlist quality. New York City Local Law 144 requires a bias audit and candidate notice for automated employment decision tools, Illinois regulates AI in video interviews, and Colorado has broader obligations for employers using AI in hiring.

In practice this means two habits. Every criterion should be job related and observable in someone's work history, so "has shipped a mobile app to the App Store" rather than "seems like a self starter". And avoid criteria that act as proxies for protected characteristics: graduation year stands in for age, a specific set of universities stands in for background, and "culture fit" as a scored criterion is both unfair and unhelpful, because it is not a property of the candidate. Teams in regulated industries usually already track which obligations apply to which process, and hiring criteria belong in that inventory alongside everything else.

Calibration examples are the highest leverage ten minutes

If you do one thing from this article, do this. Name two or three real people who would be an obvious yes for this role, and one who would be a near miss. They can be past hires, people you interviewed and lost, or public profiles of people at other companies. Then say in one line why each is a yes and why the near miss is a near miss.

This works because it converts implicit judgment into something the agent can score against. Your written criteria capture maybe seventy percent of what you actually mean by a good candidate. The examples capture much of the rest, especially the pattern preferences you would struggle to articulate: the kind of company someone thrived at, the trajectory that suggests they will keep growing, the combination of things that reads as strong to you.

The near miss matters as much as the yes examples. Anyone can describe an ideal candidate. Telling the agent where the line is, and that this specific person falls just outside it and why, is what makes the ranking sharp rather than generous.

How do I know the brief is working?

Run it, then check three things on the first shortlist. First, hit rate: of the top ten, how many would you actually screen? Below four and the brief is too loose, above eight and it may be too narrow to be finding you anything you would not have found yourself. Second, disagreement: pick the candidate you disagree with most and read the evidence behind their score. Nine times out of ten you will find the exact sentence in the brief that produced them. Third, sameness: if all ten look like the same person from the same three companies, one of your criteria is doing too much work.

Then edit and rerun. The brief is a working document, not a form you fill in once, and two iterations in the first week is normal. Once it is calibrated it becomes reusable: teams that hire the same profile repeatedly keep a brief per role family and adjust it slightly per opening, which is how high volume recruiting stays consistent instead of drifting with whoever wrote the last requisition.

One last thing worth saying plainly. A good brief does not remove your judgment from hiring, it moves it earlier. You are deciding what good looks like before you see candidates, instead of rationalizing it afterwards from whoever happened to apply. That order is better for consistency, better for fairness and better for the hire, and a person still makes every final call.