A recruitment brief for an AI sourcing agent is not a shortened job ad. It is a structured document in five blocks: the context of the assignment, the three criteria that disqualify, the two that make the difference, what the hiring manager rejected last time, and the non-negotiable conditions. With those five blocks, a sourcing agent produces a usable shortlist; without them, it finds people who look like the job title.
A recruiter at an IT services company receives a tender on Monday: a senior data consultant is needed for a client pitch on Thursday. The business manager sends three lines: “Data engineer, 5 years, Lyon, Spark a plus”. She runs the tool, gets sixty profiles, calls four. The manager concludes that AI does not work. The brief was three lines long.
Why is a job ad not enough as a recruitment brief for AI?
A job ad is written to attract. It talks about the company, its values, its perks. A sourcing agent needs the opposite: reasons to exclude. The most active LinkedIn thread of the week on AI sourcing, among trainers and recruiters, converged on one conclusion: if an AI returns fifty wrong profiles, the problem is the brief, not the AI.
Three pieces of information are almost always missing from an ad passed as-is to a tool:
- Implicit criteria: the manager knows she does not want a pure consulting background, or that environments without production are out. She never wrote it down.
- The context of the assignment: which client, which team, how much autonomy is expected from day one.
- The history of rejections: the last three candidates presented were turned down for the same reason. That reason is the most valuable criterion in the brief, and it is nowhere.
An AI-readable job brief is therefore not longer than an ad. It is organised differently.
Writing a recruitment brief in 5 blocks
1. Context in three sentences
Who the client or team is, what the assignment is, what result is expected at six months. No adjectives. A sourcing agent uses this to understand which career paths resemble the role, not to sell it.
2. Three disqualifying criteria, no more
What the candidate will not be interviewed without, whatever else they bring: a certification, verifiable experience in a specific environment, availability. Beyond three, the pool empties and the shortlist fills with profiles that tick every box on paper.
3. Two differentiating criteria
What separates a good candidate from a very good one: having already run a client pitch, having worked on a comparable migration. These criteria rank, they do not exclude. That is the nuance a tool must respect.
4. The manager's last rejection, in one sentence
“The last profile presented was turned down because his Spark experience dated from 2019 and he had never worked in a product team.” That sentence is worth ten lines of ad copy. It gives the agent the real boundary between acceptable and not acceptable.
5. Non-negotiable conditions
Salary or day rate, location, on-site presence, start date. Writing them down avoids contacting candidates who will say no at the first exchange, and protects the reply rate of the account sending the messages.
📄 A brief your tool and your hiring manager read the same way. The AI-readable job brief template: five blocks to fill in, with worked examples for SAP, data and cloud consultants, in Notion and Word.
Download the AI-readable job brief template
The mistakes that produce a bad shortlist
Four mistakes come back in briefs handed to tools, and they explain most disappointing shortlists.
| Mistake | What the tool does with it | Fix |
|---|---|---|
| “5 years of experience” with no ceiling | Surfaces 20-year profiles who will not move | Give a range, not a floor |
| A list of “nice to have” technologies | Treats them as mandatory or ignores them | Separate disqualifying from differentiating |
| A job title with no context | Searches for the title, not the actual work | Describe what the person does day to day |
| No reason recorded on rejections | Produces the same shortlist next week | Write down the why of every rejection |
The last point is the most expensive. A rejection without a reason is information lost. A reasoned rejection is one more criterion.
From job brief to scorecard: who evaluates what
Once the brief is written, it turns almost mechanically into a scorecard: each disqualifying criterion becomes a knock-out line, each differentiating criterion a weighted line. In an IT services company, this grid solves a second problem: alignment between the recruiter, the business manager and the end client, who often each have “their” ideal candidate. With a shared scorecard, the candidate no longer lives through three versions of the role.
The brief then becomes an asset. It is used to search, to explain every proposal criterion by criterion, and to recalibrate the search at every accept or reject. A tool that does not reuse it from one week to the next sends the team back to zero at every tender.
What we've seen
In September we met the recruiting team of a regional construction group in France with 2,000 employees: five recruiters, up to forty open roles each, about a hundred roles in total. Their request was about ranking inbound applications, but the need they voiced first was elsewhere: being able to enrich the brief beyond the written offer, because the criteria that really matter to managers never appear in the ad. A real-estate role published two days earlier had already received fifty resumes, most of them off target. Volume was not the problem. The absence of written implicit criteria was. That is exactly what the five blocks above formalise.
Frequently asked questions
How long should an AI-readable job brief be?
One page. Three sentences of context, three disqualifying criteria, two differentiating ones, one sentence on the last rejection, the conditions. A longer brief dilutes the criteria; a shorter one leaves the tool guessing.
Do you rewrite the brief for every tender?
No, you update it. Context changes, the trade's criteria change little. A brief reused and enriched with previous rejections gives a better shortlist from the second search onwards.
Who should write the brief: the recruiter or the hiring manager?
Both, in fifteen minutes. The manager provides the criteria and the last rejection; the recruiter structures them and translates them into language that can be verified on a profile. A brief written by only one of the two is incomplete.
Where Umamy stands
We start every search with the brief, not with a keyword. The agent asks the missing questions, builds a scorecard the recruiter can edit at any time, then computes for each profile a Dynamic Fit Score readable criterion by criterion. First candidates arrive 10 minutes after the brief, and every reasoned rejection recalibrates the next search: that is context enrichment, not a frozen model. Umamy is based in Paris and currently filing for AI Act and GDPR compliance. The full workflow is described on the AI sourcing agents page.
📄 The five-block brief template, ready to fill in. With a complete example for a data consultant and the mapping to a scorecard.
Download the AI-readable job brief template
Sources: LinkedIn discussion of the week of 26 October 2026 on AI sourcing (Umamy content radar summary, 30 October 2026, no author quoted) ; Sourceurs non peut-être, issue 293, 25 October 2026 on applicant tracking systems that keep events but not the reasons for rejections ; Umamy figures: Umamy Marketing design system, October 2026.
