Greenhouse AI review (Oct 2026): Talent Matching, Greenhouse Notetaker, and fraud checks

What this Greenhouse AI page covers

This page is for the recruiting lead or talent-ops admin at a company that runs (or is choosing) Greenhouse Recruiting, and for the people-ops generalist who inherited it. The question it answers: which Greenhouse AI features are worth switching on, what does each one actually do with candidate data, and where must a recruiter still read and decide?

Greenhouse’s AI work is spread across a few areas. Talent Matching sorts new applicants against a weighted calibration you define, and a newer Talent Matching assistant drafts that calibration from plain language. Greenhouse Notetaker records consented video interviews and drafts notes mapped to each interviewer’s feedback-form questions. The Candidate Insights Assistant answers questions about one candidate. The Real Talent suite adds fraud detection, a spam blocklist, and identity checks through CLEAR. And Greenhouse MCP lets approved AI tools work with Greenhouse data under each user’s permissions.

Greenhouse frames all of this as “keeping humans in control of every decision.” Everything below comes from Greenhouse’s pricing page, product pages, support articles, and the August and September 2026 release notes as we read them on 6 October 2026. Several features are still in open beta. This is product education, not legal advice — hiring AI rules differ by location.

Major features (with how someone uses them)

Examples at a glance: each feature has a concrete example from a real job plus a short Try this you can run this week.

1. Talent Matching with calibrations (primary)

What it is: Before Talent Matching does anything, a recruiter builds a calibration for the job: a weighted list of job-related skills, titles, industries, and minimum experience. Greenhouse then parses each resume and gives every candidate in the first Application Review stage a match label — Strong, Good, Partial, or Limited match — or Needs manual review when there is no readable resume or the candidate opted out of AI review. Candidates are grouped by label and then ordered by application date. Click a candidate and you see matched and missing skills, a short reason, and the matched terms highlighted on the resume. Both Talent Matching and resume parsing must be on, and editing a calibration recalculates labels for current candidates.

If you enable the candidate AI opt-out, applicants see an AI disclaimer on the job post with a link to manage their preference. Greenhouse says opted-out applications are labeled Needs manual review, are not evaluated, and any existing match result for that application is deleted. Greenhouse’s readiness guide asks you to write a policy for who creates calibrations, how labels inform action, and when people override them.

Concrete example (recruiting lead) A recruiting lead opens a senior accountant role and builds a calibration: CPA (high weight), NetSuite (medium), revenue recognition under ASC 606 (high), 5+ years. Of 210 applicants, 31 come back Strong or Good match. Before touching the rest, she opens five Limited match candidates. One wrote “rev rec for SaaS contracts” instead of naming the standard — the skill is there, the wording is not. She adds that phrasing to the calibration instead of advancing one person by hand, and the labels update for everyone still in review.

Try this

Primary workflow:

Build a Talent Matching calibration for one role. Open five candidates labeled Limited match and read their resumes yourself. If one should advance, fix the calibration rather than the candidate, then check how many other labels changed.

2. Talent Matching assistant (open beta)

What it is: Released in open beta in August 2026, the Talent Matching assistant is a chat panel on the Talent Matching page (“Configure with assistant”). You describe the person you are looking for, and it starts from the job post and the interview-kit attributes already set up for the job to draft two things: a suggested calibration (skills with importance levels, titles, minimum experience, and preferences such as “include similar roles”) and a set of suggested filters (location, keywords, application date range, and some advanced filters). You edit both, then click Apply setup. The difference matters: a calibration changes the order, while filters are hard constraints on who appears in the list at all.

Concrete example (talent partner) A talent partner types: “Mid-level backend engineer, Go or Rust, payments experience a plus, remote in US time zones.” The assistant drafts a calibration and adds a location filter limited to three states it inferred from the job post. Because filters hide people completely, she removes the location filter — the role is open to any US time zone — and keeps “payments” as a preference in the calibration rather than a requirement.

Try this

Filter check:

Ask the Talent Matching assistant to set up one role. Before you click Apply setup, delete every suggested filter you would not reject a candidate for on its own, and move nice-to-haves into the calibration as low-weight skills.

3. Greenhouse Notetaker

What it is: Greenhouse Notetaker became available on the Core, Plus, and Pro plans in August 2026. With candidate consent it joins Zoom, Google Meet, or Microsoft Teams interviews, records and transcribes them, and drafts notes and a summary mapped to the interviewer’s feedback-form questions, with clickable timestamps back to the recording. Interviewers add generated notes to their feedback with one click, then write their own assessment in Key Takeaways. Candidates give consent through self-schedule or availability requests and can withdraw it. Site Admins switch it on (it is off by default), control who can view recordings, and set retention up to 60 days. Since September 2026, transcripts and recording links are also available through the Harvest v3 API and Greenhouse MCP.

Concrete example (interviewer) An engineering manager runs a system-design interview with Notetaker on. Afterwards, the drafted notes map the candidate’s answers to her four interview-kit questions. One note says the candidate “proposed sharding by user ID”; the timestamp shows the candidate actually said sharding by region, then changed to user ID when she pushed back. She edits the note to capture both, then writes her own Key Takeaways — the part the AI never writes for her.

Try this

Timestamp check:

After your next interview with Notetaker on, click the timestamp behind three drafted notes. Fix any note the recording does not support before adding it to your feedback, and write Key Takeaways yourself.

5. Real Talent: fraud detection, spam blocklist, and CLEAR identity checks

What it is: Greenhouse groups its trust features under Real Talent. Fraud detection screens applications using objective digital signals to flag possible identity misrepresentation. The spam blocklist is a control you configure to block known bad IP addresses, emails, or domains before applications reach recruiters. Identity verification runs through CLEAR: candidates confirm who they are with a selfie matched to a government ID, inside their MyGreenhouse profile. Admins choose which offices and departments require CLEAR. Greenhouse’s pricing page lists fraud and spam detection on the Plus plan, and support docs describe Real Talent as an add-on for the Core, Plus, and Pro tiers.

Concrete example (talent-ops admin) A remote-first company keeps seeing the same applicant face show up under different names for engineering roles. The talent-ops admin turns on fraud detection for engineering and requires CLEAR verification only at the final-interview stage, so most candidates never face the extra step. She also writes down what happens to a flagged application — a recruiter reviews it within two days — so a flag never becomes a silent rejection.

Try this

Flag policy:

Write a one-paragraph policy before turning on fraud detection: who reviews a flagged application, how fast, and what evidence is needed to reject. Then turn it on for one department only and review every flag for a month.

Pricing context (October 2026)

Greenhouse does not publish prices. Its pricing page (checked 6 October 2026) describes three plans — Core (bring structure to hiring), Plus (optimize your operation), and Pro (manage hiring complexity) — and says pricing is customized by plan, hiring volume, organizational complexity, and the features you need. You get a number from a Greenhouse sales conversation.

On the plan table, Core lists sourcing and CRM, structured interview kits, the AI Notetaker, scheduling, reporting, SSO, and Talent Matching under Real Talent. Plus adds AI-powered report filters, the Business Intelligence Connector, sourcing automation, texting, and Real Talent fraud and spam detection. Pro adds unlimited CRM events, audit log, enterprise data configuration, and developer sandbox. Greenhouse’s support docs, however, describe Real Talent (including Talent Matching) as an add-on for these tiers — so ask your rep exactly which AI features your quote includes.

Uncertainty: We could not find any public Greenhouse price, so we quote none. Plan contents, whether Real Talent is bundled or an add-on for your contract, and which features are still beta (the Talent Matching assistant, Candidate Insights Assistant, MCP, and Analytics were open beta when we checked) all need confirming with Greenhouse before you budget.

When it’s useful (honest take)

Useful when you already run structured hiring in Greenhouse — interview kits, defined stages, consistent feedback — and inbound volume or note-taking is slowing the team down. Then Talent Matching gives recruiters a sensible reading order, Notetaker frees interviewers to listen, and the trust features help remote roles where fraud is a real problem.

Less useful when no one will own calibrations and read low matches, because a sort you never check can quietly push qualified people down the list. Less useful when you hire a few people a year — the plan and add-on cost will be hard to justify. And if your candidates are mostly hourly or high-volume, a conversational hiring tool built for that job may fit better than a structured-hiring ATS.

Related options (educational)

FAQ

What do Greenhouse Talent Matching labels mean?

Each new applicant in the first Application Review stage gets Strong, Good, Partial, or Limited match against your weighted calibration, or Needs manual review if there is no readable resume or the candidate opted out. A recruiter still reads and decides.

How much does Greenhouse cost in October 2026?

Greenhouse does not publish prices. Its pricing page describes Core, Plus, and Pro plans and says pricing depends on plan, hiring volume, complexity, and features. Ask sales, and confirm whether AI features are bundled or add-ons. See greenhouse.com/pricing.

Is Greenhouse Notetaker on by default?

No. Site Admins enable it under Configure, AI Tools. Candidates give consent through self-schedule or availability requests and can withdraw it, and admins can set retention up to 60 days.

Can candidates opt out of AI review?

Yes, if you enable the opt-out. Candidates see an AI disclaimer on the job post with a link to manage preferences. Opted-out applications are labeled Needs manual review and are not evaluated by Talent Matching.

What is the difference between a calibration and a filter?

A calibration is a weighted list of criteria that changes the order and labels. A filter is a hard constraint that hides candidates who do not match. Keep filters for true requirements only.

Optional next step (if seats overlap)

If you pay for Greenhouse Notetaker, a separate interview notetaker, and a company-wide meeting notetaker that can all join the same interview, that overlap is worth a look. Map the overlap with the $19 AI toolkit audit or the $9 AI toolkit spreadsheet. Email Audit $19 or Toolkit spreadsheet $9.

Checkout is live via Stripe. After you pay, we email delivery (spreadsheet) or intake next steps (audit).

Related: Ashby AI · Metaview · Meeting notes seats · Stack map for people ops · Workday AI · How-to: Get interviewers to file Greenhouse feedback on time without AI writing it · ops hub