AI Candidate Sourcing: Define Criteria and Review the Results

AI-assisted candidate sourcing uses AI tools to speed up the search for qualified candidates: generating search strings, reviewing profiles against criteria, and drafting personalized outreach messages. The process is faster than manual sourcing, but it does not remove the need for a recruiter to define what good looks like, review the results, and check that the criteria are legally defensible. Tools like fenbs help recruiting teams keep the approved sourcing rules and process lessons in one place without storing any candidate information there.

The most important part of AI-assisted sourcing happens before any search is run: defining the criteria. An AI that searches on vague criteria will produce a noisy candidate pool. An AI that searches on criteria that have disparate impact will surface a pool that creates compliance risk. Getting the criteria right is the recruiter's responsibility.

Defining searchable criteria before you start

Good sourcing criteria describe the skills, experience, and context that actually predict success in the role — not a copy of the job description. The distinction matters because job descriptions often include requirements that are organizational habits rather than genuine predictors, and because AI will return candidates who match the text without evaluating whether the text is right.

• Skills and experience that directly relate to the core job tasks, not the full range of nice-to-haves.
• Signals of relevant experience that can be inferred from a profile — not explicit credentials that exclude candidates who have the skills but not the title.
• Out-of-scope criteria: what you are not looking for, so the AI does not surface it.
• Compliance constraints: criteria that are legally impermissible under EEOC guidance must be excluded. Age, race, national origin, religion, sex, disability, and protected veteran status cannot be sourcing criteria.

The Equal Employment Opportunity Commission publishes guidance on employment selection procedures at eeoc.gov↗ (opens in a new tab). Any criteria that might have disparate impact on a protected class should be reviewed before they are used in sourcing.

Running AI-assisted searches

With clear criteria, AI can help generate Boolean search strings for LinkedIn Recruiter, Indeed, or other sourcing platforms. It can also help review a list of profiles against the criteria to narrow a large initial set to a manageable shortlist for human review.

Treat AI profile screening as a filtering step, not a ranking step. An AI that ranks candidates implies a precision in the comparison that the source data does not support. Use AI to remove clear mismatches; use human judgment to compare the candidates who make the shortlist.

Check the demographic profile of your shortlist as you build it. If the criteria are producing a pool that is less diverse than the available talent pool, that is a signal to review the criteria before proceeding. This check belongs with the recruiter, not with the AI.

Drafting personalized outreach

Outreach messages that reference a candidate's specific experience get better response rates than generic messages. AI can help draft personalized outreach at scale: give it the candidate profile, the role details, and the value proposition for this role, and it produces a draft that references relevant experience.

Review every outreach message before it is sent. AI drafts can misread a profile, reference experience that is not actually relevant, or produce language that is enthusiastic in ways that do not match the candidate's actual background. A brief review of each draft is faster than dealing with the confusion a bad outreach message creates.

Keeping lessons from each sourcing run

Every sourcing effort teaches something. A search string that returns strong candidates for a software engineering role can be reused or adapted. A criterion that seemed important turned out to exclude the strongest candidates in the pool. An outreach approach that got a good response rate is worth repeating.

Recording these lessons improves the next sourcing effort. Most recruiting teams learn these things and then lose them when the recruiter who ran the search moves on or the project ends. Writing down what worked and what did not — without including any candidate information — builds a sourcing knowledge base that outlasts any individual project.

How to do this with fenbs

1. Create a free account. Add a project for the role or the recruiting program.
2. Record the approved sourcing criteria as Decisions and rules: the skills and experience that predict success, the out-of-scope criteria, and the EEOC compliance constraints.
3. After each sourcing run, add a lesson: what search approach worked well, what criteria to adjust, what outreach language got better responses.
4. Use tasks for open items: search strings to test, outreach sequences to refine, criteria reviews to run with legal or HR.
5. Connect your AI tools at /integrations. Each sourcing session reads the approved criteria and lessons before starting, so it uses current guidance rather than starting fresh.

Candidate profiles, names, contact information, and application details never go into fenbs. The project record holds only the process knowledge — the criteria, the rules, and the lessons — that helps the next sourcing effort start from a better position.