AI sourcing prompts that build strategy, not fake candidates.
Use AI to structure the search, expose assumptions, and build better lanes. Do not use it to invent people, fabricate contact data, or verify things it cannot verify.
The safe prompt pattern
Give the model the role context, ask it to separate must-haves from flexible signals, then require output in lanes. Every lane should include evidence terms, false-positive filters, and a question for the hiring manager.
You are a senior technical sourcer. Turn this JD into a source pack. Return: 1. Must-have evidence 2. Flexible evidence 3. Adjacent titles 4. Donor companies 5. Search lanes 6. Boolean strings 7. X-Ray searches 8. False-positive filters 9. HM calibration questions Rules: do not invent candidates, links, contact data, clearance, or verification. Label assumptions.
What good output looks like
A useful AI output gives you a search plan you can inspect. It should not hide uncertainty. It should make the gaps obvious before you start sourcing.
- Strict lane: exact role, core tools, required domain.
- Adjacent lane: title variants and transferable environments.
- Evidence lane: GitHub, publications, packages, registries, or open-web proof.
- Calibration lane: questions that force tradeoff decisions.
Guardrail prompt to append
Do not invent people, companies, jobs, links, profiles, contact data, or verification. Treat public clearance text as an unverified breadcrumb only. Treat open-to-work language as a public signal only. Do not merge identities. Do not draft outreach unless there is specific evidence to reference.
SourcingOS workflow
Use this prompt before Candidate Search. Then run the resulting lanes through BooleanOS, X-Ray Launcher, or Candidate Search so the evidence stays visible and reviewable.