Greenhouse

Greenhouse runs hiring from job post to offer, with AI you can switch off feature by feature. An outside firm audits two of its AI features each month. Its AI terms still put the oversight duty on you, and Voice AI sits outside its uptime and security promises. Greenhouse sets the price in a quote. The contract renews by itself, and leaving runs on a 30-day clock and a 90-day clock.

By Ashish Manchanda Published 4 Nov 2026 Last updated 4 Nov 2026

Greenhouse runs hiring end to end, and its AI terms need a careful read

In one line

Greenhouse runs hiring from job post to offer. Its AI is never the final decision-maker, and a person decides. Its AI addendum leaves the oversight duty with you.

Buy it if

You want one system from job post to offer, with interview plans your hiring managers follow. You can run a quote process. You will name an owner for the AI settings and for human review of every AI result.

Do not buy it if

You need a price you can read before a sales call. You need Greenhouse to answer for AI mistakes, because its AI addendum puts the oversight duty on you. You need Voice AI under the standard uptime promise, because the Voice AI Terms exclude it.

Price

Greenhouse sells three plans called Core, Plus and Pro. Its pricing drivers include hiring volume. Its agreement bases fees on your employee headcount unless the order form provides otherwise. Get one basis in writing before you compare quotes.

What you get from job post to offer, and which AI parts cost extra

Before you read about the AI, you need to know what Greenhouse is underneath it. Greenhouse is an applicant tracking system, which means one place that runs hiring from job post to offer. It posts jobs, collects applications in one place and moves each candidate through stages that you define. Interviewers use interview kits and scoring forms, so a team judges candidates against the same criteria. Scheduling tools protect interviewer time. A recruiting CRM, a store of people you may want to hire later, sits inside the system. A prospect becomes a candidate with no import and no second login. Greenhouse also has 40+ ready-made reports and e-signature forms for onboarding.

After the hire, the details have to reach your HR system. How much of that Greenhouse does for you depends on which HR system you run. The deepest link is HRIS Link. It imports jobs, offices, departments and user accounts from your HR system. It also exports new hires. HRIS Link is available for Plus and Pro with the HRIS Link add-on. The UKG Pro version is also an add-on, and it connects to both UKG Core and UKG Onboarding. The ADP Workforce Now version is partial. It imports users and exports new hires. Workday, BambooHR, Rippling and Personio each get hire details from Greenhouse. BambooHR gets them when a candidate is marked as hired, and Rippling gets them in one click. Ask which link your HR system uses and whether your plan includes it.

The AI is where the buying decision gets harder. Greenhouse’s AI comes in three layers, and each layer has its own plan rules and its own terms. The first layer is in every tier, and it covers the routine work around a hire. It drafts a job post description from a title, qualifications and key responsibilities. It summarises interview scoring forms into agreements, disagreements and key quotes. It filters application results by keyword. It also forecasts when an offer might be accepted and when a new hire’s start date might begin. Greenhouse uses off-the-shelf public AI models and its own models, which it calls Greenhouse AI. The price section covers two more first-layer features, where Greenhouse’s sources disagree on the tier.

Greenhouse Notetaker is part of the first layer. It joins video interviews automatically and transcribes them. It drafts notes tied to the scoring questions. The Core plan includes it. It does not assign ratings, recommend scores or make subjective inferences. So your interviewers still do the scoring. Transcription currently supports English only, so ask about interviews in other languages. Candidates can opt out of being recorded at scheduling, or at any time until their application is closed.

The second layer is Real Talent. It combines fraud and spam detection, identity checks with CLEAR and AI-assisted Talent Matching. Candidates can decline the CLEAR identity check, and declining is not treated as a failure. Talent Matching compares each CV with match criteria that you define and weight. It sorts candidates into Strong, Good, Partial, Limited or Needs manual review. In Greenhouse’s AI principles, it does not assign a single numerical score to rank candidates. In its Talent Matching article, the feature provides match scores. Ask which one a recruiter sees. It only works on candidates in the first Application Review stage. It does not automatically advance or reject anyone, including candidates in the lower categories. Recruiters and hiring managers stay responsible for every hiring decision. To make the match, Greenhouse sends parsed details from each CV to OpenAI. Those are skills, titles, years of experience, employment dates and the companies a candidate worked for. That is the CV data that leaves Greenhouse when Talent Matching runs.

The third layer is Voice AI. It came from Greenhouse’s purchase of Ezra AI Labs, which Greenhouse announced as complete on 27 May 2026. Voice AI runs a two-way voice interview that a candidate completes on their own schedule. It asks follow-up questions and works in 11 languages. It records the audio and video of the answers. It turns the audio into a transcript and compares that transcript with scoring criteria that your recruiter sets. Each question gets a recommended score from 1 to 4. The recordings are not used to produce the score.

A person makes the decision, and Greenhouse’s terms make that a rule. Recruiters and hiring managers can override any Voice AI score and write down why. The score is a recommendation and not a decision. Candidates can ask for a human review of their score. The Voice AI Terms bar you from using the score as the sole basis for any decision. That covers decisions to advance, reject, rank, screen out, schedule or make an offer. Voice AI can be made optional for candidates. A candidate who opts out goes to an alternative process that you define, and is not penalised.

Voice AI is sold as its own annual subscription. Its price depends on company or team headcount. Greenhouse sets a “fair use” allowance of interviews, and there is no price per seat. Voice AI works with 60+ ATS platforms and goes deepest inside Greenhouse Recruiting. Ezra stays available as a standalone product for Greenhouse customers and for companies on other platforms. So you can buy Voice AI without moving your hiring system.

Greenhouse also offers a connection called MCP. It lets AI assistants such as Claude or ChatGPT read and change Greenhouse data. Site Admins decide what their team can reach, and users cannot give themselves more access than the admin allowed. On 22 October 2026 Greenhouse launched a Claude plugin called Greenhouse in Claude. Recruiters and hiring managers use it to work with candidates, interviews and open roles in Claude.

Our guide to applicant tracking systems covers how the rest of this market handles the same decision.

You can switch its AI off, and an outside firm tests it each month

An AI feature inside a hiring system is a risk for you before it is a benefit. So check three things first. These are who can switch it off, who tests it and where your candidates’ data goes. Connections and data rules matter whatever you decide about the AI, because your team uses them every day.

You can switch its AI off, feature by feature

AI in a hiring system touches real candidates, so you want a way to turn it off. Site Admins can switch each AI feature off across the organisation. New AI features are enabled by default, so a named admin needs to check the settings. Pro customers can control how new AI features are rolled out. Enterprise customers can make features opt-in by default. Talent Matching can be switched on or off by office, department or job. Where you enable it, candidates can ask for a review without AI on specific jobs. Greenhouse also provides AI notice templates, so you can tell applicants how AI is used in hiring.

An outside firm audits two AI features for bias every month

A bias check is only worth what the checker can see. Greenhouse engaged Warden AI, an independent third party. Warden audits Talent Matching and the AI Interviewer each month across ten protected classes. The latest Talent Matching audit is dated 14 Oct 2026 with a sample of 91,812. The latest AI Interviewer audit is dated 8 Oct 2026 with a sample of 65,880. Every new version of Talent Matching must pass a bias audit before release. Warden’s New York City report for Talent Matching used its own data set of real candidate profiles. Warden lacked access to historical data. The report reflects controlled test conditions. It is not a formal certification or compliance determination.

Greenhouse does not train models on the personal data it stores for you

Your candidates’ data is the most sensitive thing in a hiring system, so training use matters. Greenhouse does not use the personal data it stores or processes for customers to train its own or third-party models. Its own models are trained on customer data only in anonymised, de-identified form. For Talent Matching and Voice AI, Greenhouse does not train on customer data. For Notetaker, it does not currently use recordings, transcripts or AI notes for model training. Under the customer agreement, it may use anonymised data generated through Notetaker. The AI addendum is wider, and the limits section covers it. Greenhouse holds ISO 27001 (security), ISO 27701 (privacy) and ISO 42001 (AI) certifications.

It has 500+ integrations, including Slack and Microsoft Teams

A hiring system helps most when it reaches the tools people already have open. Greenhouse has 500+ integrations. The Slack integration sends Greenhouse Recruiting notifications to a Slack channel. The Microsoft Teams integration creates a Teams link when you schedule an interview with Outlook 365. A mobile app for Android and Apple handles essential tasks. Approvers can handle offers from an Approvals panel in that app. So a team can follow hiring notifications in a Slack channel and approve offers from a phone.

It has tools for GDPR, US equal opportunity reporting and audit

Data rules are easier to meet when the tool does part of the work. GDPR features come with Greenhouse Recruiting for every customer. You control how long candidate data is kept and what is removed when you anonymise a candidate. You can anonymise candidate data on request, after hire or after rejection. The US equal opportunity report is built on federally approved questions about race, gender, veteran status and disability status. It counts total and hired candidates in each category and does not tie answers to a person. A reader outside the US should check what their own law asks for. The audit log is a Pro feature, and its API covers the previous 30 days.

Greenhouse’s customers run from about 25 people to over 10,000

Whether a hiring system fits depends on your size and on how your hiring runs. More than 7,500 companies use Greenhouse. The results below are Greenhouse’s own reports, and each belongs to the company named. Treat them as claims to test in a reference call. The companies run from about 25 people to over 10,000.

Blissway, a team of about 25, uses Voice AI as its first interview

Blissway runs a team of around 25 people, nearly all engineers. Blissway uses a 20-minute Voice AI interview as the first step in hiring. So Voice AI is not only for large employers. Blissway reports average time to hire down 27%, from 88 days in 2025 to 64 days in 2026.

At about 1,000 people, Planet reports faster hiring and CarGurus reports fuller scoring

Planet has 920 full-time employees across 28 countries. Greenhouse reports that Planet cut time to hire by 47% and raised team capacity by 30%. That makes Planet a reference for multi-country hiring. CarGurus has over 1,000 employees. Greenhouse reports that CarGurus cut time to fill by 18%. It also raised interview scoring form completion from 40% to 98%. That is the result to ask about if you want hiring managers to follow a structured process.

Kaizen Gaming, PetVet Care Centers and the NFL report faster hiring

Kaizen Gaming operates in 20 markets and employs approximately 3,000 people. It reports time to hire down 33%, from 54 days to 36. PetVet Care Centers has 10,000+ team members. Greenhouse reports a 22% shorter time to hire there. It also reports 67% less manager time on recruiting tasks. That is the figure to test if your hiring managers are stretched. The NFL reports time to fill cut 24%, from 63 days to 48.

Two customers report how candidates took to AI interviews

If you worry that candidates will refuse an AI interview, these two results are the evidence to ask about. Railsware reports that 99 of 112 registered candidates finished a Voice AI interview. That is an 88% completion rate. Zapier reports 273 conversational applications across 75 days on senior roles, with a 6% opt-out rate.

The AI terms leave the oversight duty with you, and Voice AI sits outside the standard promises

A hiring system touches real people. When it makes a mistake, the mistake lands on a real candidate, and the contract decides who answers for it. Most of this section comes from Greenhouse’s own terms, and the AI terms matter most.

Under the AI addendum, you carry the oversight duty and the AI risk

Under Greenhouse’s AI addendum, you are solely responsible for human oversight of the AI features. That oversight must be enough to avoid discriminatory or unlawful behaviour. It must also avoid automated decisions that are not allowed. Greenhouse is not responsible for inaccuracies in AI output. It is also not responsible for bias or unfairness in the underlying algorithms or data. The addendum provides the AI features as is and gives no warranty for the output or its accuracy. So you need a named person who reviews AI results before anyone acts on them. The addendum also bars you from putting health, genetic or biometric data into the AI features. The same bar covers government ID numbers and personal financial information.

The AI addendum may let Greenhouse use your data and charge extra

Under the AI addendum, Greenhouse may use your data and your feedback to develop and improve its AI features. If it trains models on your data, it will use commercially reasonable efforts to aggregate or anonymise it. The addendum also lets Greenhouse require extra fees for any or all AI features, under a signed order form. For two features, Greenhouse’s support articles read differently. They say it does not train on customer data for Talent Matching and Voice AI. Ask which wording governs your contract.

Voice AI sits outside the standard uptime and security promises

Under Greenhouse’s Voice AI Terms, the service level agreement does not apply to Voice AI. The Security Addendum does not apply to it either. Voice AI customers typically start with a contract of 12 months or longer. An optional pilot can be set as an opt-out window on that annual agreement. If your hiring depends on Voice AI being up, ask for uptime and security terms in the order form.

Greenhouse’s pages differ on what Voice AI scores

What an AI interview scores decides what each candidate is judged on, so you need one clear answer. Greenhouse’s pages give two. On its Voice AI page the AI interviewer never scores accent, tone, fluency, pace, stutters, pauses, personality or confidence. In its guide to reviewing results, a communication analysis gives an overall score across six dimensions. They are Clarity, Fluency, Structure, Confidence, Engagement and Vocabulary. So fluency and confidence sit on both sides. By default the analysis does not affect the overall score. You can weight it up to 20% in interview settings. Under the Voice AI Terms, Voice AI does not include any “Communications Insights”. Ask Greenhouse which of these your plan switches on and what a candidate would see.

A rule can reject applicants, and new-application alerts skip them

Auto-reject is a rule you set. An applicant’s answer to a question rejects them automatically. The rule can assign a reason and send an email. It is available on Plus and Pro. People set up to get notifications about new applications are not notified of auto-rejected applicants. Auto-advance works the same way in the other direction. It moves an applicant to a stated stage of the interview plan. Your team writes these rules, so your team needs to review what they reject and how often.

Fraud flags can be wrong, and the terms bar acting on a flag alone

Under the Real Talent Terms, the fraud detection feature may incorrectly flag legitimate applications as fraudulent. The same terms bar you from taking action against applicants solely based on that feature. The feature reads objective application data such as IP address, email and phone. It uses a third-party vendor, IPQualityScore, which returns 26 signals. A flag is a reason to look at an application. On its own, it is not a reason to reject one.

Greenhouse expects Voice AI to fall under the EU AI Act’s high-risk rules, unless an exemption applies

Greenhouse expects Voice AI to be high risk under the EU AI Act, unless an exemption applies. It expects this when Voice AI is used to evaluate candidates. The Act’s Annex III is its list of high-risk uses, and recruiting is on that list. Article 6(3) exempts a listed system that poses no significant risk of harm to fundamental rights. That exemption stops applying when the system profiles people. Greenhouse notes that the enforcement date for Annex III high-risk systems has been deferred to 2 December 2027. That is subject to publication in the Official Journal. Under its Real Talent Terms, you acknowledge that Talent Matching is not intended for any Annex III high-risk purposes. Ask your own counsel which applies to your hiring.

Check where candidate data goes before you switch on Voice AI, Notetaker or Talent Matching

You will need to tell candidates, and your own privacy team, where candidate data goes. It leaves Greenhouse for outside providers in several ways. Voice AI is available in the US today, with more regions planned. It sends audio to third-party AI providers. Examples are Deepgram for speech to text and OpenAI, Google or both for analysis. Greenhouse’s service providers also include Anthropic, which hosts large language models for its AI features in the US. Notetaker data is hosted on Recall.ai’s US servers, including for customers on what Greenhouse calls its “EU silo”. For Talent Matching, transfers from the EEA or UK are subject to the EU-U.S. Data Privacy Framework or its UK Extension. Those are the schemes for sending personal data to certified US companies. Greenhouse also relies on Standard Contractual Clauses as a second basis. These are standard legal terms for sending personal data out of the EU or UK.

Several AI parts are still in beta or not yet released

Sourcing recommendations are in closed beta with select Greenhouse Recruiting customers. The Candidate Insights Assistant is in open beta for Core, Plus and Pro customers. Limits on “fair use” may apply, and Greenhouse will communicate them at general availability. Greenhouse MCP is in open beta, and access and pricing go through the account team. The Greenhouse Recruiting Agent is coming soon. Sourcing recommendations reach only select customers, and the Recruiting Agent is not out yet. Keep any part that is not released out of your price comparison.

A security incident at a vendor reached Greenhouse’s business systems

You are handing Greenhouse sensitive data about real people, so its security record matters. Greenhouse reports a security incident at a third-party vendor, Klue. Attackers obtained credentials that Klue used to connect to certain Greenhouse business systems. Ask Greenhouse whether any of those systems held candidate data.

The price comes from a quote, the contract renews by itself, and leaving has a clock

You get Greenhouse’s price from a quote. Greenhouse customises its pricing to each company’s hiring needs. Plan, hiring volume, organisational complexity and required features all drive the cost. Its Master Subscription Agreement bases fees on your employee headcount unless the order form provides otherwise. That headcount includes employees of affiliates that you give access. Hiring volume and headcount can give different totals, so ask for one basis in writing. You can upgrade plans at any time.

The plans differ in what they hold. Core includes Sourcing and CRM, and interview kits and scoring forms. It also includes AI Notetaker, scheduling, reporting and a single company login. Greenhouse’s pricing page also puts Talent Matching in Core, under its Real Talent name. Its support articles put Talent Matching in the Real Talent add-on. Plus adds a connector for business intelligence tools, Greenhouse Sourcing Automation, Texting and Real Talent fraud and spam detection. Pro adds unlimited CRM events, the audit log and enterprise data controls and security. It also adds a developer test environment and enterprise Real Talent features.

Add-ons sit on top of the plans. Voice AI is a separate annual subscription. HRIS Link is an add-on for Plus and Pro. Core customers can buy Sourcing Automation seats as an add-on. Two AI features also carry conflicting tier information. Greenhouse’s AI recruiting page puts AI report filters and CV anonymisation in every tier. Its pricing page puts report filters in Plus. A support article puts CV anonymisation in Pro.

Five terms in the agreement decide what you pay and when you can stop. Subscriptions renew automatically for one year unless the order form provides otherwise. Greenhouse notifies you at least 60 days before an automatic renewal for terms of a year or longer. Put that date in your diary. Fees are due in advance, and they are based on the services you bought and not on actual usage. Greenhouse reserves the right to increase fees on renewal. Fees are non-cancellable and non-refundable except in cases the agreement sets out.

The agreement also caps liability for each party. So each side’s liability stops at the fees received or payable in the 12 months before the event. Stated exceptions apply. Read that cap beside the AI addendum, where Greenhouse is not responsible for inaccuracies in AI output. Terms change, and this version is dated 1 February 2026.

The uptime promise is in the service level agreement for Greenhouse Recruiting, Onboarding and Sourcing Automation. Below 99.5% monthly uptime, Greenhouse awards 5 service credits. Below 99.0%, it awards 10. A service credit is Greenhouse’s own unit of compensation. A customer awarded more than 15 in a 90-day period can end the contract for cause. Voice AI is not covered, as the limits section explains.

Leaving has two clocks. Greenhouse provides an API key for export. If you want an export file, ask within 30 days of expiry or termination. Expect to pay its then-current data migration rates. At 90 days after the contract expires or ends, Greenhouse queues all customer data for deletion. The agreement defines data migration and data export as professional services. Put both dates in your diary.

Moving in is a separate job. Greenhouse puts the move of historical data from a former ATS at 4 to 6 weeks. Bulk import uploads active and historical candidate applications. Each import can hold up to 8,000 rows. Core, Plus and Pro all include bulk import.

Greenhouse reports a 589% three-year return on investment on its company page. Ask for the calculation behind that figure before you use it in your business case.

You could also run hiring steps with agents built on your own systems. We have mapped what that looks like. The HR Agent Map

Fourteen questions for the sales call

Use these questions on the sales call. Each one matches a point above. Each has the answer you want to hear, so you know when to push for more.

Which AI features are in the plan you quoted, and which are paid add-ons? Good answer: A written list by feature, with the price of each add-on on the order form.
Your pricing page puts Talent Matching in Core. Your support articles put it in the Real Talent add-on. Which is right? Good answer: One written answer, with the plan or add-on named on the order form.
Your AI recruiting page puts AI report filters and CV anonymisation in every tier. Your pricing page puts report filters in Plus. A support article puts CV anonymisation in Pro. Which tier has each one? Good answer: The tier named for each in writing, and a date by which it is live in our account.
Does Talent Matching give each candidate a single score or only a category? Good answer: A clear answer, shown on screen in a demo with what a recruiter really sees.
Will you set our fees on one basis, headcount or hiring volume, and write it on the order form? Good answer: One named basis, with the headcount or volume figure it uses.
Will you narrow the AI addendum clause on using our data to develop your AI features? Good answer: Yes, in writing, so our candidate data is used only for our own account.
What uptime and security terms cover Voice AI, since the standard ones do not? Good answer: A written uptime figure and security terms for Voice AI in the order form.
Does Voice AI score communication, and does it count toward the score by default? Good answer: A named list of what is scored, with any weight set in writing.
When a rule rejects an applicant automatically, how does our team find out? Good answer: A report or alert that covers every automatic rejection, on our plan.
What notice will we get before renewal, and what cap will you put on a fee increase? Good answer: The notice date and a written cap on any increase at renewal.
What is one service credit worth in money, and what counts as a CRM event on Pro? Good answer: A written money value per credit, and a written definition of a CRM event.
How do we get our candidate data out when the contract ends, and what does the export cost? Good answer: A named export format, a price and a written window longer than 30 days.
Where is our candidates’ data processed, for each provider you use? Good answer: A named country for each provider, and a line we can give a candidate who asks.
Did the Klue incident reach any system that holds candidate data? Good answer: A written yes or no, with the systems involved named.

Alternatives to Greenhouse

If Greenhouse comes off your shortlist, look at Ashby, Lever, Workable and SmartRecruiters. Each one differs from Greenhouse on a different point.

Ashby is the closer match if you want to see a price before a sales call. Ashby puts a company-size selector and a monthly total on its Foundations plan, which covers up to 100 employees. Its AI Interviewer is an early access waitlist today. Greenhouse sells Voice AI now, in the US. If you want an AI interview this year, compare the two on that point.

Lever differs on where its automation sits. At Lever, auto-screening comes with the Enterprise edition or the Advanced Automation add-on. At Greenhouse, auto-reject comes with the Plus and Pro plans. Ask each vendor what its screening rule does to an applicant.

Language is one point to check between Workable and Greenhouse. Workable’s Language Kit translates what candidates see, from job post to offer signing. Enterprise and Premier plans can have up to 9 extra languages, and Standard plans one. Its Agent does not support languages other than English yet. Greenhouse’s Voice AI works in 11 languages, and its Notetaker transcribes English only. If you hire in more than one language, ask each vendor which AI parts work in yours.

On uptime, SmartRecruiters makes a different promise. SmartRecruiters suits organisations with at least 50 employees and a significant number of open requisitions. It promises 99.9 percent annual system availability in its service level agreement. Greenhouse pays compensation below 99.5% monthly uptime. Ask each vendor for its uptime promise in the same terms before you compare them.

Our market guide also covers HR systems that include recruiting, such as BambooHR, Rippling and UKG. If you hire only a few people a year, ask whether you need a separate hiring system at all.