12 Claude Prompts for In-House Recruiters
In-house TA teams burn most of their week on three things: building longlists, chasing champions whose jobs changed, and writing the first DM. None of those are creative work. They're pattern-matching jobs, and a Claude + LinkFetch stack does them in a fraction of the time a sourcer with a Recruiter Corporate seat does. A single Recruiter Corporate seat now runs $10,800 per year in 2026, and a five-person team pays $54,000 to $75,000 before InMail overages source. The 12 prompts below replace most of what you're paying that bill for.
How to use this library
Every prompt assumes you've installed LinkFetch's MCP server and connected it to Claude Desktop or Claude Code. Once it's wired, you paste a prompt, hand Claude the role spec or the candidate list, and let it run. No Boolean strings. No pivoting between five tabs. Each prompt has a credit estimate so you can budget before you queue a batch.
LinkFetch is a compliance-first LinkedIn data API that runs through a passive-first browser extension you keep signed in to your own LinkedIn account. The user is the principal, the data is yours. That matters when you're reading 200 profiles a day for a single role.
A note on what this library is not. These are not reusable templates you should send unchanged. The whole point of agentic sourcing is that the model writes a different message to every candidate, grounded in things it actually saw on their profile. The prompts produce drafts. You read them, edit two or three, and send.
Why prompts beat Boolean strings now
The shift this year is from filter-based sourcing to natural-language sourcing. AI-powered sourcing reduces sourcing time by over 70% compared to manual Boolean searches in Recruiter source. The reason isn't that AI is smarter than a senior sourcer (it isn't). The reason is that you can express a fuzzy intent ("people who'd respect this hiring manager") in one sentence, and let the model unpack it into a structured query against a real dataset.
The single biggest variable across every category of sourcing tool is personalization quality, not volume. A recruiter who sends 20 genuinely personalized messages outperforms one who sends 200 templates, every time source. The prompts below are written for that 20-message recruiter, not for the 200-message one.
The library
1. Build a passive candidate longlist from a single role spec
You paste a role spec; you get back a ranked longlist of 60 names from companies that match your competitor set. The 18-to-48-month tenure window is the sweet spot for openness to a conversation. Anyone shorter is still settling in. Anyone longer is either rooted or actively job-hunting (which means they're already in 30 other recruiters' inboxes).
2. Score every name against the must-haves
Longlists are noise without scoring. This prompt runs the must-haves over the full longlist, scores each candidate 0-10, and surfaces only the top 25. The "single biggest disqualifier" line saves you from finding it the hard way on the call.
3. Detect a champion's job change in your talent CRM
Champion job changes are the single most actionable signal a recruiter has. A new VP of marketing at a mid-market company has a 70% chance of changing at least one tool in their stack within the first 90 days source; the equivalent for a recruiter is that someone you placed three years ago who just took a new VP role is going to want to hire 5 people in the next quarter. This prompt flags the move on the Monday it happens.
4. Map a target company's relevant team in 90 seconds
This is the prompt that replaces 40 minutes of clicking through profiles trying to figure out who'd be the right hire's future peers. Hand the model a company and a role, get back a small org chart with the function mapped out and the people you should actually be talking to.
5. Find alumni from feeder companies
Every TA team has 5 to 10 companies whose alumni convert at 3 to 5x baseline. This prompt watches those companies for departures and surfaces the destination. The "openness band" is rough but useful: someone who moved 4 months ago is still in the honeymoon, somone who moved 10 months ago is starting to look around again.
6. Surface boomerang candidates from former employees
Boomerangs close at twice the rate of cold candidates because trust and tooling familiarity are already baked in. The 12-month threshold is the median "I'd consider a move" window for ex-employees of mid-sized SaaS companies. This is a quarterly job, not a weekly one.
7. Draft a personalized first-touch DM
The DM prompt is the one most teams underuse. The trick is the constraint list: no emoji, no "hope this finds you well", no opening lines about coming across the profile. What you get is a 4-sentence message anchored on something the candidate actually shared, which moves response rates from the standard 10-15% baseline into the 35-40% band source.
8. Hiring-manager intake brief
The intake call is where most pipelines go sideways. Hiring managers describe the role they think they want; recruiters spend 6 weeks building it; then it turns out the manager would have hired a different shape of human entirely. This prompt builds the intake brief before the call, so you walk in with three questions sharp enough to catch that misalignment.
9. Convert a Boolean string into a structured shortlist
For teams migrating off Recruiter, the Boolean library is the asset that's hardest to throw away. This prompt translates an existing Boolean directly into a structured LinkFetch query, runs it, and tells you which conditions had zero matches (which is usually where the over-specification lives).
10. Daily candidate triage queue
Most candidate pipelines die from triage rot, not from a lack of names. The 09:00 triage prompt sorts new pipeline adds into "call today / nudge this week / move to network" so the queue doesn't accumulate. Run it for 8 weeks and the average days-to-first-call drops from 11 to 4.
11. Build a diverse slate before the intake call
The "tell me which constraint is binding" line is the important part. If you can't hit a 50% women or non-binary slate without dropping the must-haves, the model tells you which must-have is the blocker. That's the conversation to have with the hiring manager, not after the slate has already shipped.
12. Post-rejection talent network nurture
Final-round rejects are the most underused asset in any TA function. They've been through your pipeline, they liked you enough to come back twice, and they're 12+ months into the role they took instead. This prompt is the one that turns the talent network from a graveyard into a pipeline.
Credit budgeting
A typical in-house TA team running 5 open roles uses LinkFetch like this: ~600 credits a week on longlist + scoring (prompts 1, 2, 9), ~200 a week on triage and DMs (prompts 7, 10), ~100 a month on alumni and boomerang scans (prompts 5, 6, 12). On the LinkFetch Team plan, that's well under the monthly allotment, with credits left over for the on-demand prompts. A five-recruiter team replaces one Recruiter Corporate seat at roughly a tenth of the cost.
FAQ
Do I need a LinkedIn Recruiter seat to run these prompts?
No. The whole point of the library is that you don't. LinkFetch's data layer runs through a Chrome extension that observes your own LinkedIn account, so you keep the same access you'd have as a regular signed-in user. You lose the InMail surface, which most teams replace with regular connection requests plus a follow-up note. Connection-plus-note response rates beat InMail in most A/B tests in 2026.
How is this different from sending the same prompts to ChatGPT or Gemini?
ChatGPT and Gemini don't have a tool layer that can read live LinkedIn data. They can write good-sounding messages, but they're guessing at the candidate's last role and last project. Claude with the LinkFetch MCP server reads the actual profile before writing the message, which is why prompt 7 produces drafts that don't read as templated.
Will sending DMs at this volume get my account flagged?
Volume is the wrong axis. LinkedIn's classifier looks at the ratio of outbound to inbound, not the absolute number. The library is built around personalization, which raises reply rates, which protects the ratio. The DM prompt is also designed to produce messages worth replying to, not messages that look like a sequence step.
What happens if a candidate is private or has a missing profile?
The model is told to say so explicitly and return only the firmographic data it could pull. You'll see "profile is private, here's the company context only" rather than a fabricated summary. That matters more than it sounds. Most AI sourcing tools will hallucinate a tenure rather than admit a gap.
Can I share prompts across the team?
Yes. Most teams keep them in a shared Notion page with the role-spec template at the top. The prompts are the playbook; the role specs are what change.
Last updated 2026-05-04 by the LinkFetch team.