Only 7% Refuse AI Screens. The Engineers You Sourced Weren't Asked

If you run outbound, Greenhouse's finding that only 7% of candidates would refuse an AI-led interview looks like permission to put a voice AI screen between a cold-sourced engineer's reply and your first call. The data doesn't support that, and Greenhouse's own Head of Voice AI explains why.

The figure comes from Greenhouse's 2026 Future of Hiring Candidate Report. On Nov 1 it ran in Recruiting Brainfood 525, which reaches over 39,000 subscribers, in the slot the issue credits as "supported by our friends at Greenhouse." So it was the sponsor's message, not Hung Lee's own commentary.

What did Greenhouse's 2026 candidate report actually find?

Greenhouse surveyed 2,832 active job seekers and found only 7% would refuse an AI-led interview outright. The headline figures come from the US sample of 1,200, and every respondent was already looking for work.

The numbers that matter:

• Sample: 2,832 job seekers aged 18 to 64 across the US, UK, Ireland, Germany and Australia, released Oct 20 to 21.
• Preference: candidates were 2.7x more likely to take an AI interview than to withdraw if no human interviewer was available.
• Experience: 48% of US respondents took an AI interview in the past year, and 70% of those said it let them show their skills well. In the UK and Ireland the figures were 42% and 57%.
• Silence: 81% of US respondents were rejected in the past year without speaking to anyone at the company, human or AI. The figure was 86% in Germany and 85% in Australia.

81%

That last number frames everything else. If four in five candidates get rejected without a single conversation, an AI interview beats silence. That tells you about frustrated applicants. It tells you very little about engineers who weren't looking.

Why don't the 7% and 38% figures contradict each other?

They measure different things. Greenhouse's May report counted people who had walked away from a process over an AI interview at least once, while the October report counted people who say they would refuse one outright.

The May 1 report surveyed 2,950 active job seekers. 38% had walked away from a hiring process because it included an AI interview, and another 12% said they would. The same report found 59% of US candidates prefer a live interview with a human, against 12% who prefer an AI-conducted one.

"Walked away once" is a low bar, because one bad pre-recorded video screen puts you in it. "Refuse outright" is a high bar, because you have to rule out every AI interview on principle. The same person can be in May's 38% and in October's 93% who wouldn't refuse. Both statements are true at once.

Who did both Greenhouse surveys leave out?

Neither survey asked anyone who wasn't looking for a job. That leaves out the passive engineer you cold-messaged, who is the person this decision actually affects.

Look at how the 2.7x figure is framed. It compares taking an AI interview with withdrawing "if a human interviewer wasn't available." That framing assumes the candidate is already in a process. They applied, they put in effort, and if they're in the 81%, they've been rejected without a conversation somewhere else. With that sunk cost, almost any conversation looks worth taking.

A passive candidate has no sunk cost. They didn't apply. They answered your message because something in it was interesting. If the next step is a bot, saying no doesn't look like withdrawing. It looks like not replying.

That's why AI interview drop-off in outbound is invisible. A sourced engineer who goes quiet shows up as "no response," not "declined AI screen," and no survey counts a non-reply as a withdrawal. The number that matters most to sourcers is the one nobody measures.

Why does an AI screen fit inbound applicants but not outbound?

An AI screen fixes a volume problem: too many applicants and not enough recruiter hours. Outbound has the opposite problem, because the scarce thing is the candidate's attention, not the recruiter's time.

You don't have to take my word for this. Ophir Samson, who founded Ezra and is now Greenhouse's Head of Voice AI, described the product's main use on a DataCamp podcast: "the majority of the use cases are that the company will receive a bunch of inbound applicants, so they'll get 2,000 applications." Zapier, an early Ezra customer, screens candidates by voice. That's the right fit. A big inbound pile is where a first-pass voice screen earns its cost.

Outbound looks very different. In Refolk's index of professional profiles:

These counts match on job titles. Plenty of people source under a plain "Recruiter" title, so 1,286 understates sourcing capacity. Read it as a sign of how rare the specialist title is, not as a headcount.

~167

With inbound, recruiter time is scarce and candidate attention is plentiful, because the candidate asked to be considered. With outbound, the right engineers are a finite list and each one may give you a single reply. Putting an AI screen there saves the cheap resource (a recruiter call) and spends the expensive one (that reply). In Germany, with a pool about 13x smaller by this count, each wasted reply costs even more.

So the leverage in outbound comes earlier, in deciding who gets the human call at all. That's the part Refolk is built for. You describe the engineer in plain English and get back a ranked shortlist from GitHub, LinkedIn, and the open web, so your recruiter hours go to people who actually fit.

What does an AI first contact cost you with a sourced candidate?

For a sourced candidate, the first conversation after they reply is their first impression of your company. In Greenhouse's May data, 34% came away with a more negative view of the employer after an AI interview.

The top reasons people walked away from a process in May:

1. Pre-recorded video scored by AI with no human present (33%)
2. The company didn't disclose how AI would be used (27%)
3. AI monitoring during the interview (26%)
4. An AI-led interview was required, with no alternative (26%)

On top of that, 51% never heard back after an AI interview.

An inbound applicant who has a bad AI screen still wanted the job, and many will sit through it anyway. A passive engineer at Datadog or Vanta didn't need you to begin with. When their first contact is a scored recording, they read it as a signal about how your company treats people, and they draw that conclusion before they've met anyone on the team.

A passive engineer who replied gave you one conversation. Don't hand it to a bot.

Where should the AI screen go in your hiring process?

Use the AI screen on inbound applications, with disclosure and human review, and keep a human on the first call with anyone you sourced. The October respondents' own requests point to the same split.

Inbound: AI screening done properly

1. Disclose up front. 88% of October respondents want at least one legal requirement for AI disclosure.
2. Offer a live alternative. A required AI interview with no alternative was a walk-away reason for 26% in May.
3. Skip pre-recorded scored video. It was the top walk-away trigger at 33%.
4. Say a human reviews the answers. 49% want to know that.
5. Close the loop. 51% never heard back after an AI interview, which undoes whatever goodwill the screen earned.

Outbound: a human first

1. Build the shortlist before you write a word. Tight targeting is what makes human calls affordable. Refolk handles this step: ask for the people you want and get them back ranked.
2. Send a personal first message about the work, not the process.
3. Make the first call human, with the recruiter or the hiring manager. A human call meets the disclosure and human-review expectations by default, so you don't need workarounds.
4. Add structured or AI-assisted assessments only after the candidate has opted in, and say so plainly when you do.

If you're on Greenhouse or Ashby, the tooling is already within reach. Ashby has its own AI Interviewer in private beta, and Greenhouse Voice AI is also live for Ashby customers. So the question isn't whether you'll have an AI screen. It's which stage of which pipeline gets it.

Keeping humans first does cost time. Every first call takes a recruiter hour, and with dedicated sourcer titles this rare, most teams don't have a spare hour for a role that's been open for months, let alone one they'd otherwise pay an agency fee to fill.

Some companies have clearly decided the human hours are worth it. Anthropic and Rippling both show up in Refolk's index as employers of dedicated technical sourcers, among the 1,286 US profiles with that title. Companies that still pay people to source are betting that a passive candidate's first contact deserves a person.

How should recruiters read vendor-funded candidate surveys?

Read a vendor survey the way you'd read a candidate's profile: check who produced it and what they gain from it. Greenhouse's surveys are useful, but the timeline matters.

• May 1: Greenhouse publishes the 38% walk-away report.
• May 5: Greenhouse announces it is buying Ezra AI Labs, a voice AI interviewing company.
• May 27: The deal closes.
• Aug 1: Greenhouse Voice AI launches.
• Oct 20 to 21: The 7% report is released.
• Nov 1: The 7% stat runs in Greenhouse's sponsor slot in Brainfood 525.

None of this makes the data wrong. Greenhouse, led by CEO Daniel Chait, surveyed active job seekers, and active job seekers are exactly who its Voice AI screens. The sample fits the product. Within that population, a candidate survey and a product roadmap can both be honest. The mistake would be applying either one to engineers who were never in the sample.

FAQ

Should recruiters use AI screening for passive candidates?

Not for the first contact. Both Greenhouse surveys covered only active job seekers, and its Head of Voice AI says most customers use the product on inbound applicants. A passive candidate you sourced has no sunk cost, so an AI screen is more likely to end in silence than in a completed interview. Use a human first call. Bring in AI-assisted assessments only after the candidate has chosen to enter your process.

Is Greenhouse's 7% figure wrong?

No. It answers a narrow question: 7% of active job seekers say they would refuse an AI-led interview outright, in a hypothetical. That fits alongside May's finding that 38% had walked away at least once, because one person can do both. The figure is accurate for applicants. It just tells you nothing about people who weren't applying.

Do AI interviews hurt employer brand?

According to Greenhouse's May report, sometimes. 34% of candidates came away with a more negative view of the employer after an AI interview, and 51% never heard back after one. The most common walk-away trigger, at 33%, was pre-recorded video scored by AI with no human present. On inbound, disclosure, a live alternative and closing the loop address the top triggers directly.

When can AI interview screening fit into outbound recruiting?

After the candidate has opted in. Once a sourced engineer has had a human conversation and agreed to a formal process, a disclosed, AI-assisted technical screen is a reasonable step, as long as a human reviews the result. What you shouldn't do is make it the first thing a cold-sourced candidate sees after replying.