AI sourcing stack for passive-candidate hiring at scale

This is the stack for the sourcing function, not the recruiting function. It assumes you already have an ATS and recruiters, and that your constraint is upstream of both: the people you need do not apply. Passive-candidate sourcing at scale is a recall problem — every profile index has holes, and the roles that stay open longest are the ones whose candidates sit in somebody else’s hole. The stack closes that gap with two search indexes, one CRM in front of the ATS, and an attribution loop that tells you which index paid.

If you want the full hiring loop — screening, scheduling, interview intelligence, debriefs — that is the AI-augmented recruiting stack. This one goes deeper on one job.

How the pieces fit

• SeekOut is the depth index. It searches a public-profile pool SeekOut puts at 1B+ across seven verticals, enriched with GitHub activity, academic papers, patents, and clearance and work-authorization data — the signals that surface senior engineers, researchers, and cleared candidates that title-keyword search misses. Six agents (Rubric Creator, Candidate Finder, Profile Analyst, Outreach Author, Market Researcher, Engagement Tracker) turn a req into a scored slate rather than a result list. Note where the tier boundary sits: ATS rediscovery and silver-medalist search are in Sourcing + Integration, two steps above the published Core plan.
• hireEZ is the breadth index — and, since August 2026, something larger than that. It aggregates a billion-plus profiles from 45+ platforms across the open web, and remains the stronger of the two for Mandarin, Japanese, and Korean sourcing, where LinkedIn coverage thins out. It is also the second opinion on contact data: when SeekOut returns a profile with no verified email, hireEZ resolves a usable one on a meaningful share of them. What changed is the packaging. hireEZ relaunched on 2026-08-18 and reached general availability on 2026-08-27 as an agentic talent-acquisition layer — seven named agents covering sourcing, applicant review, voice and chat screening, scheduling, talent CRM, career sites, and hiring intelligence, sold at a price configured against the tools it consolidates. In this stack you buy the index and decline the CRM half. Reasons below.
• Gem is the CRM and the single front door to the ATS. Every profile from either index lands in a Gem project, gets deduped against Greenhouse before anyone spends an outreach touch, and enters a sequence. Gem’s rediscovery agent searches your CRM and ATS together and surfaces silver medalists and past applicants who match a new req — the cheapest candidates in the building, because you already paid to find them. One detail decides the quote: Gem splits seats into Standard and AI-Powered, and the rediscovery and ranking agents ride the AI-Powered class. A quote that looks cheap per head is worth checking for the mix.
• LinkedIn Recruiter is the verification and InMail channel. It is the profile source of truth the aggregated indexes are copies of, and the only place to reach someone whose email you cannot resolve. Recruiter Lite covers occasional hirers with 20 filters, third-degree reach and 30 InMails a month; full Recruiter opens 40+ filters and the whole member base, with Hiring Assistant as a paid per-seat add-on. Treat it as a channel and a checker, not as your search layer — that is the seat you are buying two indexes to stop over-paying for.
• Greenhouse is the ATS and the terminal record. Sourced prospects convert into Greenhouse candidates with the source field set, stage transitions record against them, and source-of-hire reporting closes the loop back to the two indexes. Greenhouse rebuilt its plan ladder in 2026 — Essential / Core / Advanced / Expert is gone, replaced by Core, Plus, Pro — and 500+ pre-built integrations still make it the ATS with the most room for four upstream tools all writing to it.

The handoffs that make it a stack

The order is fixed and each arrow is an event, not a habit.

A new req opens in Greenhouse. Before any search runs, Gem’s rediscovery agent scans the CRM and ATS for prior applicants and past prospects who match, and returns a rediscovery list — talent rediscovery first, because it costs nothing per profile. What rediscovery does not fill goes to search: SeekOut for the technical or credential-heavy slice of the req, hireEZ for geographic and open-web breadth. Both push their matches into the same Gem project, where dedupe against Greenhouse drops anyone already in process. Gem enriches contact data, and profiles that come back without a verified email route to LinkedIn Recruiter for an InMail instead of being dropped. Sequences run from Gem. A reply flips the prospect to a Greenhouse candidate with the source attributed to the index that found them. When the req closes, Greenhouse source-of-hire reporting and Gem’s funnel analytics answer the only question that matters at renewal: which index produced hires, not which produced profiles.

Why two indexes instead of one

Because sourcing at this level fails on recall, not on interface. Each vendor’s index is assembled from different crawls — SeekOut leans on technical and credential signals, hireEZ on open-web breadth and non-English coverage — and the overlap is nowhere near total. On a hard req, the second index is the difference between a slate of 12 and a slate of 30.

The rule that keeps that from being expensive duplication: two indexes only pay if there is exactly one CRM between them and the ATS. Run SeekOut and hireEZ each with its own sequencing, and you get the same person messaged twice in a week by two sourcers, which is the fastest way to burn a passive candidate. Gem exists here to be the choke point — one dedupe, one sequence, one reply inbox, one attribution model. That is why the ATS-versus-recruiting-CRM split is load-bearing rather than a nice-to-have.

hireEZ’s August relaunch is a direct argument against that rule, and it is worth taking seriously rather than dismissing. hireEZ now ships its own talent CRM agent, its own sequencing, its own screener and scheduler, and prices the whole thing against the point tools you cancel. It does not replace the ATS — Greenhouse stays the system of record either way — but it does replace the layer Gem occupies. So the two products are no longer complements here; they are two answers to the same question, and you should buy one of them. Buying both and using half of each is the version that costs the most and reports the worst.

The agent layer, and the one leg that doesn’t have it

Three of the five tools now expose an MCP server, which was true of none of them when this stack was first written. SeekOut runs one with 14 guided workflows wired to your existing license, workspaces, filters and credits. Gem hosts GeMCP at mcp.gem.com/mcp, scoped to whatever permissions the user already holds. Greenhouse shipped a first-party MCP server on 2026-05-07 that reached open beta on 2026-07-06 across Core, Plus and Pro, with a Site Admin setting scope ceilings under Dev Center.

That combination means a sourcer can ask Claude “who did we source for this req last year and never contact, and who in SeekOut matches the same rubric” and get an answer without opening either product. It is the first genuinely useful thing MCP does for a sourcing function, because rediscovery is a query problem and queries are what assistants are for.

Two guards before you build on it. Greenhouse MCP is open beta, not GA, so tool surfaces can move without a deprecation window — keep it off hiring-manager-facing surfaces and pin an integration test to the schemas. And changing MCP scopes invalidates every active token in the Greenhouse org at once, so batch scope changes into one announced window instead of tuning them request by request.

hireEZ is the gap. It exposes an API key and no MCP server, so anything agent-side against the breadth index is build work you own. If your sourcing team is already working inside an assistant, price that build into the hireEZ evaluation and re-check MCP status at renewal.

What it costs

Budget $70K-$170K a year for a 4-seat sourcing pod, before the ATS, with a median-of-medians build landing near $100K. The line items:

• SeekOut publishes Recruit Core at $149/mo billed annually ($1,788/yr) including 3 seats, 500 contact credits and 1,000 exports a month. Everything above Core is quoted, and the tier carrying ATS rediscovery is two steps up — check which tier your integration sits in before signing. Third-party trackers put real deals in a $10-30K/year band with a median near $20K, plus $2-10K implementation; cap the 5-7% annual escalator at signing. Triangulated estimate, not a rate card.
• hireEZ publishes one figure, $494/mo for the Solo Recruiter plan with a 7-day trial, and states plainly that enterprise is priced against the stack it replaces rather than a flat per-seat rate. Vendr’s de-identified transaction data, last updated February 2026, puts the median contract at $13,000/year across a $7,000-$25,000 range (estimate).
• Gem prices on your company’s FTE count, not on sourcer seats — the detail that decides the quote before you negotiate it. The startup ladder runs $130/mo billed yearly for 1-10 FTE against a $270 list rate, free for six months under 30 employees, with reduced published rates from 30-100. Above 100 FTE it is quoted; Vendr reports a $25,700 median across 233 purchases, a $7,000-$73,184 spread, and about 20% average negotiated savings (estimate).
• LinkedIn Recruiter is usually the largest single line, and LinkedIn publishes no US price. The one public per-seat ladder is the rate card LinkedIn Ireland files on the UK Government’s Digital Marketplace under G-Cloud 14: £8,925 per seat per year at 1-2 licences, £8,575 at 3-10, £8,150 at 11-30, bottoming out at £6,350 at 251+. Four seats sit in the 3-10 band at £34,300/year, roughly $43,000 at 1.27 USD/GBP. Hiring Assistant is a second per-seat line — £2,079 at 1-2 licences falling to £1,575 at 101-250 — and cannot exceed your Recruiter licence count. The discount curve is shallow: 29% from the first rung to the last, arriving at a licence count a sourcing pod never reaches.
• Greenhouse is quoted and sits outside this budget — you have it already or this is the wrong stack. For reference, Vendr’s 873-purchase dataset puts the median at $26,587/year.

There is a real floor version: SeekOut Core, the Gem startup rate, and Recruiter Lite at a buyer-reported ~$1,680/year runs just over $5,000 a year for one or two sourcers. And there is a break-even test. Three contingency placements at 20-25% of a $150K first-year salary — $30,000-$37,500 each — covers the $100K median build. The top of the band needs four to five. Below three passive-sourced hires a year, buy the agency.

Common variations

• hireEZ as the platform, not the second index. Drop Gem, the screener and the scheduler, and run hireEZ’s seven agents on top of Greenhouse. This is the strongest new alternative to the stack as written, and the trade is explicit: you give up the two-index recall argument and Gem’s reporting, and you get one contract and one reply inbox. Take it when the pain is tool count and screening volume rather than search recall, and when nobody on the exec team reads a Gem funnel dashboard. Do not take it while a build-heavy agent integration is on your roadmap — hireEZ ships no MCP server.
• One index instead of two. If fewer than 30% of your reqs are deep-technical, cleared, or non-English, the second index is duplicated spend. Keep hireEZ if the pain is breadth and contact data; keep SeekOut if the pain is technical depth and credential filters.
• Drop Gem, sequence from hireEZ. The narrower version of the consolidation play: keep both indexes but move sequencing into hireEZ. This works at one or two sourcers with no CRM analytics requirement. Swap back the moment a second sourcer double-touches a candidate or a hiring manager asks for source-of-hire by channel.
• Ashby instead of Greenhouse. Ashby’s native analytics cover much of what you would otherwise buy Gem’s reporting tier for, so the Gem quote should come down. Choose it when you are picking an ATS now rather than replacing one.
• Enterprise ATS. On Workday or iCIMS, the CRM layer stops being optional — their native prospect handling will not carry a sourcing pod, so Gem or Beamery becomes mandatory rather than a choice.
• Buy the outcome instead of the seats. When the constraint is sourcer hours rather than search quality, SeekOut Spot and similar managed services sell interview-ready slates per role instead of licenses. That is a different purchase and a different budget line — compare it against the agency fee above, not against this stack.

When this stack is the right pick — and when it isn’t

Pick it when at least 40% of your hires come from outbound, the roles are senior, technical, or specialized, you have 2-8 people whose actual job is sourcing, and you make 30 or more passive-sourced hires a year. That combination is what amortizes two search contracts.

Skip it in four cases. If hiring is hourly or frontline, this is the wrong category — go to Fountain-class tooling. Under 15 hires a year, the contingency-fee math beats the stack outright. If nobody’s job title is sourcer, the tools go unused: recruiters running full loops will not operate two search indexes, and you should buy one plus a managed service. And if your ATS data is unmaintained — duplicate records, empty source fields, stale dispositions — rediscovery returns noise and attribution lies, so fix the data before you buy the stack that depends on it.

What this stack does NOT replace

• A recruiter. This stack produces engaged, replying candidates. Screening, selling the role, and closing stay human and stay downstream.
• Inbound. Recruitment marketing and employer brand drive the applications this stack never has to source. Outbound-only pipelines cost more per hire, permanently.
• The interview loop. No scheduling, no structured interviewing, no assessment, no interview intelligence in the stack as written — that sits downstream in the ATS or in separate tooling. The hireEZ consolidation variation is the exception, and it pulls screening and scheduling into scope along with the compliance duties below.
• Your data-protection obligations. Sourcing profiles of people who never applied puts you in scope for GDPR lawful-basis and notification duties in the EU and UK, and for CCPA/CPRA rights in California. Buying the tools creates the obligation; it does not discharge it. Agent access through MCP does not change the analysis — the same records, read by a different client.
• A bias audit. Agent-scored slates rank people, and voice or chat screening decides who advances. Where AI-assisted ranking or screening feeds hiring decisions, disclosure and audit rules — NYC LL 144 and the EU AI Act among them — attach to the employer, not the vendor. Two-party-consent states add a recording-consent duty on top. Keep a human approval gate on every rejection.