- Step 1: launder the job title before you search
- Step 2: build the search inside the tool that pulls the most signal
- Step 3: triage the long list before you read a single profile
- The counterintuitive part: search the people who left, not the people who are there
- The failure mode to watch for
- What you save, where it goes
The candidates you cannot find on LinkedIn are usually not hiding. They have a different job title, work at a company you have not searched, or describe their work in language that does not map cleanly onto your requisition. Boolean strings are good at finding people who already match a search you can articulate. Hidden talent, by definition, is the talent your search cannot articulate yet. That is the gap a sourcing workflow with AI is built to close.
LinkedIn’s Future of Recruiting 2025 reports 37% of recruiting professionals are now actively using or experimenting with generative AI, with sourcing and outreach as the top use cases. The recruiters getting real lift are not the ones running prettier searches. They are the ones running searches their competition has not thought of.
Step 1: launder the job title before you search
A “Senior Product Manager” job is also titled, in different organisations, Lead PM, Group PM, Principal Product Manager, Staff PM, PM II, Product Lead, Director of Product, Head of Product, Senior Owner, Squad Lead, and a half-dozen company-specific variants. Your search misses all of them by default.
Open ChatGPT or Claude and run a title-laundering prompt before you touch a sourcing tool:
You are helping me build a sourcing search for a [ROLE TITLE] at a [COMPANY SIZE + INDUSTRY]. List 20 job titles this person might currently hold across: large enterprises, mid-market companies, scale-ups, agencies, consultancies, and adjacent industries. Then list 10 phrases this person is likely to use in their LinkedIn “About” or headline that would not appear in their job title. Then list 5 GitHub, Substack, or community signals (open source repos, conference talks, newsletter authorship, Slack community moderation) that would suggest they have done this work even if their title doesn’t say it. Output as three labelled tables.
The third table is the one your competitors are not running. A “principal data scientist” who maintains a popular open-source library has signalled their level publicly in a way that is invisible to a job-title search.
Step 2: build the search inside the tool that pulls the most signal
For most agency and in-house teams, the sourcing tools doing real work right now are hireEZ, SeekOut, Eightfold, and Gem. Each pulls from a different mix of public profiles, GitHub activity, paper authorship, conference attendance, and patent filings. The trick is using the title-laundering output as the input to the search, not retyping the requisition.
Paste the table from step one into hireEZ or SeekOut as your title list. Layer your “About” phrases as keyword boosts, not as required terms. Layer the community signals as filters. You will pull a candidate set that looks 30 to 50% different from a Boolean-only search of the same role.
Step 3: triage the long list before you read a single profile
This is where most sourcing time disappears. A 200-name long list takes hours to read. AI shrinks the read to minutes if you feed it the right artefact. Export the long list as a CSV with profile text, then:
I am sourcing for [ROLE], reporting to [MANAGER TITLE], at [COMPANY]. The non-negotiables are: [LIST 3 TO 5]. The disqualifiers are: [LIST 2 TO 3]. The “interesting but not required” signals are: [LIST 3 TO 5]. I am pasting candidate profile text below. For each candidate, return: a one-line summary of why they are or are not a fit, a “fit score” out of 5 against the non-negotiables, the single strongest signal in their profile, and any disqualifier that fired. Sort the output strongest to weakest.
You will end up with a ranked shortlist of 20 to 30 names with the reason they are interesting attached, which is what you actually need to write outreach.
The counterintuitive part: search the people who left, not the people who are there
Sourcers spend most of their time on people currently doing the role at competitor companies. The richer pool, often, is the people who used to do the role and moved adjacent. A senior engineer who became a founder and is winding down a startup. A lead designer who took a sabbatical and is two months from re-entering. A product head who joined an early-stage company that just shipped its Series B and might be ready for the next thing in 18 months.
A prompt that finds them:
List 15 hypothesis profiles for someone who would be unusually good at [ROLE] but is not currently in that title. Examples might be a founder whose company is in a quiet phase, a senior IC who took an executive route and wants to go back to making, a person at a stagnating company in the same domain, or a domain expert in an adjacent industry who has the underlying skill. For each hypothesis, give me 3 LinkedIn search filters or GitHub queries that would surface that profile. Avoid generic suggestions.
Run those queries through hireEZ or SeekOut. The hit rate is lower per profile, but the candidates that come back are usually warm, available, and not on every other recruiter’s list.
The failure mode to watch for
The first time most recruiters run an AI-augmented sourcing flow, they end up with 200 names that all look plausible and a list of reasons that all sound right. AI is excellent at producing confident summaries from thin signal. If you are not careful, you spend the time you saved on triage emailing 200 people who don’t actually meet your bar.
Two safeguards. First, make the prompt require a disqualifier check, not just a fit score. Most candidates have something that should knock them out, and if AI is not finding any, your criteria are too soft. Second, sample-check the top 10 profiles by hand. If 7 of them are actually a fit, the ranking is working. If 3 of them are, your prompt is hallucinating signal and you should rewrite the non-negotiables more sharply.
What you save, where it goes
The honest gain on this workflow, once you have run it three or four times, is somewhere between 40 and 60% of the hours that used to go into long-listing and triage. That is meaningful but it is not the whole win. The bigger win is that the candidates you reach out to are stronger on average, because the search pulled from a wider pool and the triage was more disciplined. Outreach reply rates, downstream, look better. The shortlist for the hiring manager is shorter and tighter.
The hours you reclaim should not go into running more searches. They should go into the candidate conversations the searches surface, which is the part of the job AI cannot do for you.