What Should a Candidate Sourcing Agent Return?

A candidate sourcing agent should return more than a list of names. It should hand a hiring team a reviewable research packet: who merits attention first, what supports that priority, what remains unknown, where each signal came from, and what to do next.

That output contract keeps discovery separate from selection. A shortlist is an investigation queue, not a promise that the people at the top are complete matches, objectively ranked, or likely to be hired. The team still verifies claims, chooses outreach, runs interviews and makes the employment decision.

The output contract at a glance

Output	What it should contain	Review question
Shortlist	Ordered candidates and the role criteria used	Why is this person in this position?
Evidence	Supporting profile, project or career signals	What visible fact supports each relevant requirement?
Uncertainty	Unknown, adjacent or conflicting information	What should a human verify?
Source trace	Source name, link, date and search context	Can a reviewer retrace the signal?
Feedback	Corrections, preferences and false-positive notes	What should change in the next search?
Shareability	A readable report with context and permissions	Can another decision-maker review it without a call?
Next action	A named owner and disposition	Is the next step verify, progress, hold or close?

1. A shortlist that explains priority

The shortlist should be ordered for attention, not presented as a verdict. Put the role brief beside each candidate: must-haves, nice-to-haves, boundaries and approved adjacent backgrounds. For each result, explain in plain English what visible work, scope or context made it worth reviewing. State the search scope so nobody mistakes it for a complete market census.

2. Evidence attached to each requirement

Evidence should be requirement-by-requirement. Show which public signal supports a must-have, which is adjacent and which preference is present. Separate a title from the work behind it and a technology keyword from a project using it. Reviewers should be able to inspect the source; if a profile says "led migration" but omits scale, record the claim and missing context rather than inventing a number.

3. Uncertainty made visible

Every result needs an uncertainty field. Useful labels include supported, adjacent, unknown and contradicted. Unknown means the public material did not establish a fact; contradicted means it conflicts with an agreed requirement. Missing information is not automatically negative, so add a verification question for material unknowns. Do not hide uncertainty inside a score or describe a score as a probability of job performance.

4. A source trace a colleague can follow

Source trace turns a result into an auditable hand-off. Record the source name, direct URL where appropriate, date checked and the profile section supporting the explanation. Note the search or role-brief context too. Distinguish public evidence from interpretation, never imply access to private information, and verify material details before outreach because public information can be incomplete or stale.

5. Feedback that changes the next pass

Feedback is part of the output, not an afterthought. Let reviewers mark false positives, useful adjacent backgrounds, missing evidence and over-weighted criteria. Capture corrections in the brief's language, such as "treat payments infrastructure as relevant adjacent experience" rather than "show fewer like this". Change one important input at a time, record the approved change and compare the next search without pretending it is automatically accurate.

6. Shareability with enough context

A shareable report should stand alone. Include the role brief version, search date, criteria, explanations, uncertainty, sources and reviewer notes. Send the smallest useful report to people who need it, avoid sensitive or speculative details, and let recipients add a disposition or question without changing the original evidence.

7. A named next action

End every candidate record with a disposition and owner. Progress means the evidence merits a human conversation; verify means an important question remains; hold means timing or a second opinion is needed; close means the profile conflicts with an approved requirement. These are workflow states, not predictions. The report should say whether to verify a small set, revise the brief, run another search or hand the shortlist to the hiring manager.

Where Talent Summoner fits

Talent Summoner is our product for AI candidate sourcing, CV ranking and outreach. It sends outreach from your own connected Gmail, Outlook or LinkedIn account, only after you review each message and confirm the send. Its candidate sourcing page, checked 3 October 2026, says a role brief is searched across LinkedIn, GitHub and other public sources across 200M+ profiles, returning a ranked shortlist with must-haves, nice-to-haves and plain-English reasoning. It also describes feedback-driven sourcing; use the contract above to review evidence, uncertainty and source trace.

For an existing CV pool, the candidate-ranking tool accepts up to 50 PDF, DOCX, MD or TXT files and creates a shareable report, according to the live page checked the same date. Your first Role is free, with no credit card needed, and your second is US$4, an intro price shown at checkout, not on the pricing page. After that, the pricing page currently lists US$29 for one Role, US$79 for three Roles and US$59 per month for 10 Roles; pack Roles remain valid for 24 months and unused monthly Roles do not carry over. Recheck terms before purchase.

Talent Summoner is a sourcing and ranking product, not an ATS. Your team owns verification, the review and send decision for each outreach message, interviews, accommodations, selection and communication.

Bottom line

Specify the output before judging the agent: shortlist, evidence, uncertainty, source trace, feedback, shareability and next action. That contract makes a sourcing result useful even when the available information is incomplete. Start with a candidate sourcing brief, then ask the role owner to inspect the evidence and assign the next action. For a wider test that also covers outreach and setup, see our guide to AI recruiting agents.

FAQ

Should a sourcing agent return a score?

It may return an ordering or fit explanation, but a score should not be treated as proof of accuracy, fairness or job performance. The useful output is the evidence and uncertainty behind the priority.

What is source trace in candidate sourcing?

Source trace records where a signal came from, when it was checked and how it relates to the role brief. It lets another reviewer retrace the reasoning and identify stale or incomplete information.

How should feedback affect a new search?

Feedback should identify a concrete correction, such as an over-weighted title, a missed adjacent background or an unclear must-have. Apply the approved change, record it and compare the next result set with the previous one.

Can a shortlist be shared with a hiring manager?

Yes, if the report includes the criteria, reasoning, uncertainty, sources and a way to record the next action. Talent Summoner's candidate-ranking report supports shareable reports; limit access to the people who need the information.

Does a candidate sourcing agent replace the hiring decision?

No. It can organise discovery and prioritise review. Humans verify material claims, decide whether to contact someone, conduct the assessment and make the hiring decision.

Talent Summoner product facts read from our live candidate sourcing, candidate ranking and pricing pages, verified 3 October 2026. We re-verify this page quarterly — tell us if something changed.