AI sourcing maps: from JD to boolean strings in one run

A sourcing map is everything you need to start hunting for a role, generated in one run: the target companies to raid, the adjacent talent pools most recruiters miss, the search channels worth your time, and ready-to-paste boolean strings for each of them. What used to be an afternoon of tab-hopping becomes the first fifteen minutes of the search.

This post breaks down what a good sourcing map contains, why generating it in one run beats prompting for boolean strings piecemeal, and how to keep the quality consistent across every search your team runs.

The piecemeal problem

Most recruiters already use AI for sourcing — one fragment at a time. “Give me a boolean string for a senior data engineer.” Then, in another chat: “Which companies in Berlin have strong data teams?” Then: “Synonyms for ‘data pipeline’?”

Each answer is fine. The problem is that the fragments don’t talk to each other:

• The boolean string doesn’t know which companies you’re targeting, so it can’t exclude them or tune for their tech stack.
• The company list doesn’t know the JD’s actual must-haves, so it optimizes for fame rather than fit.
• Nothing gets reused: next search, you re-improvise the whole set from scratch, and a teammate running the same role gets entirely different results.

Sourcing quality ends up depending on who did the prompting and how much patience they had that day. That’s the chatting-vs-building gap in its purest form.

What a sourcing map contains

A complete sourcing map, generated from the job description and intake notes in a single run, has four layers:

1. Target companies. Where do people with this exact skill profile work today? Not just the famous names: the unglamorous mid-size companies where the JD’s stack is the daily job. Grouped into direct competitors, adjacent industries, and feeder companies.
2. Talent pools. The non-obvious populations: people with the right skills under different titles (“analytics engineer” hiding inside “BI developer”), bootcamp-and-OSS profiles, relevant communities and conference speaker lists.
3. Channels. Which of LinkedIn, GitHub, niche job boards, or your own ATS history is actually worth searching for this role. Sourcing a COBOL engineer and a design lead are different games.
4. Boolean strings per channel. Ready-to-paste strings tuned to each platform’s quirks (title variants, skill synonyms, exclusions), derived from the same must-haves the rest of the search uses.

The layers reinforce each other, and that’s the point of generating them together: the boolean strings inherit the company targets and the title variants from the talent-pool analysis, instead of being generic guesses.

Why one run beats one afternoon

Beyond speed, the single-run approach fixes three things piecemeal prompting can’t:

• Consistency. Every searcher on the team starts from the same map structure. Search #50 is as thorough as search #1, because thoroughness is encoded in the workflow, not in someone’s mood.
• Provenance. The map derives from the JD and intake notes you attached. When a target company looks odd, you can trace why it’s there — same philosophy as evidence-quoted CV screening: every output should justify itself.
• Iteration. When the client adds a constraint (“no candidates from vendors”), you re-run the map with updated intake notes. The whole set (companies, pools, strings) updates coherently, instead of you patching one boolean string and forgetting the rest.

Garbage in, garbage map

One honest caveat: a sourcing map is only as sharp as the job description feeding it. “Looking for a rockstar engineer with good communication skills” produces a map of everywhere and nowhere. This is why the sourcing map sits second in the search lifecycle, after a JD workflow has turned raw intake notes into explicit must-haves. Fix the JD first; the map gets dramatically better for free.

Running it as a workflow

The Sourcing Map workflow in Calyflow takes the job description and intake notes you’ve attached to a project and produces the full four-layer map in one run: typically a dozen boolean strings across channels, with the run cost shown at the bottom. It runs on your own AI key and your own tools, so the same workflow serves the whole team regardless of which model you prefer, and your search strategy never leaves your control.

The takeaway

Stop prompting for boolean strings one at a time. Generate the whole sourcing map (companies, pools, channels, and strings that share one understanding of the role) in a single run, from the same JD the rest of your search uses. Your first hour on any new role should be spent contacting candidates, not assembling search queries.

Generate your first sourcing map free: create a free account. Your own API key, no credit card.