Greenhouse Voice AI Scores Your Transcript. Prep the Resume It Checks.

If you've applied to a company on Greenhouse or Ashby lately, the next email you get from them may not come from a recruiter. It may be an invite to a voice interview run by Greenhouse's Voice AI. The AI asks the employer's questions out loud, transcribes what you say, and holds your answers up against the resume you sent. Your accent and your nerves don't count. Your exact words do, and it quotes them, which changes how you should prepare.

What is Greenhouse Voice AI, and who is using it?

Greenhouse Voice AI is a two-way voice interview, built on Greenhouse's acquisition of Ezra AI Labs, that asks questions the hiring team wrote and scores your answers against criteria that team defined. It's live now for Greenhouse and Ashby customers, and Lever is listed as "coming soon."

The timeline is short:

• Before the deal: Zapier was one of Ezra's early customers and used it to screen candidates by voice.
• 5 May 2026: Greenhouse announces it will buy Ezra.
• 27 May 2026: the deal closes. Ezra founder Ophir Samson joins Greenhouse as Head of Voice AI.

The bigger reason you're likely to run into it comes from the market. An early-October roundup of screening software from HeroHunt concluded that for a mid-market team on a modern ATS, the screening built into Greenhouse, Ashby or Workable "will beat any bolt-on." An ATS (applicant tracking system) is the software that holds a company's job postings, applications and hiring pipeline. Once the default tool in that system can run an interview, plenty of teams will just switch it on.

How much of the market sits on these platforms

Most recruiters in Refolk's index who list one of these ATS platforms as a skill work on one where Voice AI is already live: 148 of 170, or 87%. I counted US recruiters and talent acquisition professionals by self-listed skill. It's a rough signal, since people tag skills sparsely and some employers show up in more than one list. Still, the gap is large.

That works out to 4.5 Greenhouse-tagged recruiters for every Ashby-tagged one. The Ashby employers lean toward scale-ups like Stripe, Waymo and Drata. The Greenhouse list includes large companies like Meta and Palo Alto Networks. Treat these counts as relative indicators, not market share.

There's one wrinkle for Ashby. Greenhouse's own July roundup says Ashby has its own AI Interviewer in private beta. So an "Ashby AI screening interview" could be either product, though the prep below works for both.

How do you know a voice AI interview is coming?

You usually find out when the invite arrives, because Voice AI sends it automatically once you reach a pipeline stage the employer picked. Nobody has to schedule a call, so the AI interview can be your first contact with the company.

Disclosure is often poor. Greenhouse's 2026 candidate report found that 70% of US candidates who went through AI evaluation said it wasn't clearly disclosed before their most recent AI interview. Look for these signals:

• An email that arrives right after you apply or move forward, with no recruiter name attached.
• Wording about a "voice interview," "AI interview," or a recorded conversation.
• A notice with opt-out instructions. Greenhouse gives employers candidate-facing templates that include an explicit opt-out line.
• Any mention that your answers will be transcribed or reviewed later.

Opting out is a real choice, but it isn't invisible. Scores, transcripts, rationale and opt-out signals all land on your candidate record in the ATS. Employers can make Voice AI optional, and Greenhouse's terms bar them from using it as the only reason for a hiring decision. What happens after an opt-out is up to the employer. If you skip it, email the recruiter directly and ask for a call so your file shows a request, not just a declined invite.

What does the Greenhouse voice AI interview actually score?

It scores the transcript of what you said, one question at a time, against criteria the employer wrote. It records audio and video but doesn't score them, and Greenhouse's terms say it may not read your emotion, sentiment or mood.

There's no universal rubric. You may have seen claims of a fixed 15-minute format graded on "skills, communication and problem-solving." Greenhouse's own materials say each team writes its own questions and scoring criteria, and length varies too. Greenhouse CEO Daniel Chait has talked about a 25-minute voice interview, and Zapier's Ezra interviews ran up to 30 minutes. Plan for whatever length the employer sets in the invite.

From what's public about the system, here's what is and isn't graded:

At Zapier, before the acquisition, each answer got a score from 1 to 4. Now every per-question score comes with a written explanation that quotes your own answers. That's the part that matters most. If you say "I helped out with a lot of the migration work," the recruiter will read that exact sentence as the evidence behind your score.

In practice, pauses, filler words and a shaky voice cost you almost nothing. Vague answers cost you a lot, because the AI puts them in front of a human, word for word.

Why is your resume the answer key?

Your resume is the document your spoken answers get checked against, so the biggest risk isn't a weak answer. It's an answer that contradicts what you wrote. Greenhouse says Voice AI includes fraud detection and a "resume authenticity layer that flags misrepresentation before it reaches your recruiters."

Greenhouse's sales pitch says this openly. AI-tailored resumes have started to look alike, so a structured spoken conversation becomes the real filter. Chait's version: "scammers hate a 25-minute voice interview at the start of the process." Every flag is tied to an exact moment in the transcript, and the hiring team reviews the evidence before making its own call.

The tool is built to catch one situation: a bullet that claims more than you can explain. "Led the platform migration" is fine if you led it. If you were one of six engineers and your part was the rollback scripts, the gap shows up the first time the AI asks you to walk through your role. Now the flag, the timestamp and your exact words all sit on your record.

AI-padded bullets are exactly what this layer looks for. Refolk builds your resume from your own work history and then tailors it to each posting, so every tailored bullet still traces back to something you did and a follow-up question doesn't catch you out.

How do you prepare for an AI phone screen, bullet by bullet?

Turn every bullet on the resume you sent that company into a 60-second spoken story you can tell without notes, then check that the transcript backs up the bullet. Delivery coaching won't move your score. Specifics will.

1. Pull the exact version you sent. If you tailored your resume for each application, the AI is checking you against that version, not your master copy.
2. Circle every verb. "Led," "owned," "built," "drove," "architected." For each one, ask whether a skeptical hiring manager on that team would accept it. If not, change it before you apply, not after the invite shows up.
3. Write a 60-second outline for each bullet:
   • Context (one sentence): the system, team or problem.
   • Your part (two sentences): what you personally did, in first person.
   • The decision (one or two sentences): an option you rejected and why.
   • The result: the number from the bullet and how you measured it.
   • The lesson (one sentence): what you'd do differently.
4. Prepare for the three follow-ups that test authenticity: What exactly was your role? How did you measure that? What went wrong?
5. Record yourself and read the transcript. Phone dictation is enough. The transcript is what gets scored, so read it the way the recruiter will. If a sentence wouldn't work as a quote in the rationale, rewrite it.

A before-and-after example

Bullet on the resume: Led migration of billing service to Kubernetes, cutting deploy time by [your number].

A weak spoken answer: "Yeah, so I led the migration, we moved everything over to Kubernetes, and deploys got a lot faster, it was a big team effort."

A strong spoken answer: "Billing deployed through a hand-rolled script that took most of an afternoon. I scoped the migration, wrote the Helm charts for the three billing services, and ran the cutover with one other engineer. We decided against migrating the reconciliation job in phase one because it touched the ledger. Deploy time dropped by [your number], measured from merge to production across the next month. I'd add load testing earlier next time."

The second answer gives the AI something to score in every sentence, and it matches the bullet exactly. If the honest version is "co-led," change the bullet to say so.

Picking which bullets to rehearse first is easier if you know what the posting cares about most. Refolk's fit score shows which of the posting's requirements your history actually supports and where it's thin. Rehearse the thin spots first.

Should you use a live AI assistant during the interview?

No. Outside help is the kind of thing Greenhouse's fraud layer flags, and while a flag doesn't lower your score, it gives the recruiter a timestamped moment to question.

Every flag is tied to an exact point in the transcript and reviewed by a human. Even if your scores are fine, a recruiter will see a marked moment where your answer looked assisted, sitting right next to a resume that may also have been AI-assisted. You're handing them the exact doubt the product was bought to catch.

The other problem is simpler. Live assistants produce polished, generic text, and the rationale quotes generic text back as weak evidence. Your own rehearsed specifics will score better than a fluent paragraph that could describe anybody.

Does voice AI hurt or help candidates without big-name employers?

It can help them, because it lets employers interview far more people than a recruiter's calendar ever could. Plural cites the "fewer than 7% of applicants" who currently get an interview at all, and voice screening widens the top of that funnel.

Recruiters are swamped. Applications per recruiter rose 411.8% between 2022 and 2025, from 146 to 746 a year. Zapier used Ezra to screen five times more candidates and advanced 30% of people it would otherwise have passed over.

If your resume doesn't carry a famous employer, a structured interview scored on what you say is one of the few places where substance can beat pedigree. That makes your prep more valuable, not less.

What should you do after the AI interview ends?

Follow up with a human, because about half of candidates who finish an AI interview hear nothing at all.

Here's what to do, in order:

• Ask the AI your questions during the interview. Samson says candidates "ask Ezra a ton of questions about the role or the compensation and the benefits." At Zapier, those questions were flagged so company reps could respond, so asking about pay range is fair.
• Write down what you were asked right after you hang up. These questions show what the team cares about, and they shape your next round.
• Email the recruiter within a few business days. Name the role, mention that you completed the voice interview, and add one specific point you'd expand on.
• Update your resume if a bullet felt shaky out loud. The next company's AI will ask about it too.

The rule doesn't change: anything you write, you need to be able to say.

FAQ

Can I opt out of a Greenhouse voice AI interview?

Sometimes. Employers can make Voice AI optional, and Greenhouse's candidate notice templates include an explicit opt-out instruction. The opt-out is recorded on your candidate record, so pair it with a direct email to the recruiter asking for a human screen.

Will my accent or nervousness lower my score?

No. Voice AI scores only the transcript, and Greenhouse's terms say it may not read your emotion, sentiment or mood. Audio and video are recorded but not scored. What lowers your score is vague content, because the written rationale quotes your exact words back to the recruiter.

How long is the Ezra AI interview, and what questions will it ask?

The employer sets both. Each team writes its own questions and scoring criteria, so there's no fixed rubric. Greenhouse's CEO has described a 25-minute voice interview, and Zapier's Ezra interviews ran up to 30 minutes. Expect questions about the experience on your resume, with follow-ups about your specific role and results.

Is an Ashby AI screening interview the same thing?

Often, yes. Greenhouse Voice AI is live for Ashby customers, so many Ashby employers use the same Ezra-built system. Ashby also has its own AI Interviewer in private beta, so the format may differ. The prep is the same either way: make every bullet something you can explain out loud in about 60 seconds.