Hiring GlossaryWhat is voice AI? How voice AI technology works in interviews
Voice AI is artificial intelligence that can hold a spoken conversation in real time: it listens to what someone says, understands the meaning, and responds the way a person would, instead of routing a caller through a menu of pre-recorded prompts. In hiring, voice AI usually shows up as a voice AI interview, an automated conversation that asks a candidate questions out loud, follows up based on what they actually say, and hands a recruiter a transcript and a scorecard instead of a stack of resumes to guess from.
How voice AI technology works
Voice AI is really a stack of technologies working in sequence, not one single model. The same stack is what powers conversational interviewing and AI screening tools more broadly, whether the use case is a structured hiring interview, a benefits intake call or a support triage line. Here’s what’s actually happening at each step of a voice AI interview.
Speech recognition
The system transcribes a candidate’s spoken answer into text in real time, including the pauses, the specifics and the hedging, not just a clean summary of what was said.
Natural language understanding
This layer interprets what the answer actually means: the reasoning behind it, where the candidate gets concrete versus vague and what they emphasize without being asked. It’s the difference between hearing words and understanding an answer.
Dialogue management and follow-up questions
This is what separates voice AI from a scripted phone tree or a fixed-menu chatbot. Instead of moving to a predetermined next question regardless of the answer, the system can probe a vague response, ask for a specific example or follow a thread the candidate opened. That same follow-up capability is also what helps a well-designed system catch scripted or AI-generated answers, since a real answer tends to hold up under a follow-up and a scripted one tends not to.
Scoring and structured output
The conversation gets scored against a rubric the hiring team set, not a generic skills bank, and the system produces a transcript, a scorecard and often suggested follow-up questions for whoever reviews the file next.
None of this replaces a recruiter’s judgment. It gives the recruiter something more complete to exercise that judgment on than a resume and a gut feeling.
Where voice AI shows up in hiring
Most hiring teams meet voice AI technology at the first-round screen: a conversation that used to mean a recruiter repeating the same 15- or 30-minute call dozens or hundreds of times. A voice AI interview runs that first conversation on the candidate’s schedule instead. Greenhouse’s guide to conversational AI screening breaks the technology down into four specific use cases.
- Finding the needle in the haystack: using voice AI to manage a crowded pipeline, so recruiters spend live conversation time on the candidates most worth it
- The conversational application: rethinking screening for a world after the resume, where a short structured conversation surfaces more than a static application can
- Precision shortlist screening: assessing an already-strong, small pool to find the best candidate rather than filtering noise out of a large one
- The optional pre-conversation: giving candidates useful context before they commit, without treating an opt-out as a strike against them
Why hiring teams are adopting voice AI technology
The case for voice AI usually comes down to a math problem small teams already know: application volume keeps climbing, recruiter headcount mostly doesn’t, and the gap gets filled by whatever shortcut is fastest, often a resume skim that misses people worth a conversation. Voice AI reduces the coordination load of first-round screening so recruiters spend their time where judgment matters, not by replacing the judgment itself.
- A wider applicant pool gets a real conversation instead of a resume-only pass, without adding recruiter headcount
- Review time per candidate drops once a recruiter is reading a transcript and scorecard instead of starting from a blank page
- Candidates answer the same questions in the same order, making it easier to compare people fairly
What to watch for
Speed is the easy part of this category. Verification is where most of the real risk lives, and it shows up in a few specific ways.
Fraud and scripted answers are a growing problem across remote hiring: one 2026 Greenhouse report found 91% of recruiters had encountered some form of candidate fraud. A voice AI tool that can’t tell a live, unscripted answer apart from a recording or an AI-generated response isn’t solving that problem, it’s automating it.
Bias is another risk: a rubric that was never checked for disparate impact just runs the same bias faster and at higher volume than a person would.
Transparency is where candidates are clearest about what they want, and the gap between what they want and what they get is wide. In one candidate survey, 70% said they weren’t clearly told AI would assess them, and only 21% believe employers use AI responsibly in hiring. Meanwhile 88% want at least one legal requirement that makes employers upfront about AI in interviews, with advance notice the most common ask.
The deeper issue is decision ownership. Voice AI should surface structured signal for a person to weigh. The moment a tool starts making the actual hire or reject call on its own, or a recruiter starts treating its output as a verdict instead of an input, the system has outrun the governance around it.
That gap between what AI can do quickly and what a hiring team can actually stand behind is exactly where trust breaks down, and it’s worth interrogating in any vendor demo, not just taking on faith.
What good voice AI technology looks like
Structured hiring is the standard a voice AI tool should be held to, not an exception made for it. A review of what makes AI interview screening good points to five design traits that build candidate trust, not just efficiency:
- Listening: the screen responds to what a candidate actually said, rather than moving through a fixed script
- Follow-ups: it probes vague answers by asking for specifics, which also helps catch scripted or AI-generated responses
- Answering questions: candidates can ask about pay, team structure or day-to-day work and get real answers back
- Patience: it doesn’t cut in during pauses, which especially helps non-native speakers and people who think before answering
- Transparency: it discloses that the candidate is speaking with AI, that a person reviews responses before any decision, and what happens to the recording afterward
Underneath all five is the same non-negotiable: the interview runs on criteria the hiring team sets, every candidate gets a full transcript and a documented rationale instead of a silent pass or reject, the system is checked regularly for bias, and a person owns the hiring decision, every time.
Greenhouse Voice AI is built against that bar, and the clearest test of it is what happens when a team that’s nervous about AI interviews actually tries one. At Railsware, a global software company, leaders expected candidate pushback. "In such a competitive tech talent market, we were worried that an AI interview would cause pushback," said Olia Ududiak, Chief of Staff at Railsware. Instead, 91 percent of candidates rated the interview excellent, and recruiter review time per candidate dropped from an hour or more to 7 to 10 minutes. One candidate put it plainly: the system “was able to ask meaningful follow-up questions, build on responses and explore topics in greater depth.”
That’s the pattern worth looking for in any voice AI tool, Greenhouse’s included: candidates feel heard rather than processed, and recruiters get their time back without losing the thread on who they’re actually hiring.
The bottom line
Voice AI is a real shift in how the first conversation between a company and a candidate happens, not a gimmick and not a shortcut around good hiring. Used well, it gives recruiters back the hours they used to spend on repetitive first-round calls and gives candidates a more consistent, more responsive first impression than a form ever could. Used carelessly, it’s just a faster way to make the same old hiring mistakes, or new ones nobody’s checked for yet.
The difference comes down to the questions asked earlier: Is the rubric the team’s own? Is every decision explainable? Does a person own the outcome? Those questions matter more than which vendor’s logo is on the tool.
If your team is evaluating voice AI interviewing for the first time, see how Greenhouse Voice AI approaches structured, human-reviewed screening.
