One-way recorded interview
- Who leads
- Fixed prompts
- Candidate experiences
- Records answers without a live conversational response
- Hiring team should inspect
- Question quality, accessibility, and how recordings are reviewed
Guide Book · State of AI in Hiring
The State of AI Voice and Avatar Interviews in 2026
A guide for talent leaders
AI interviews are moving from experiments to everyday hiring workflows. Some systems speak with candidates through a voice interface. Others add an on-screen avatar that asks questions, responds to answers, and represents the employer’s hiring brand. Their value depends on the quality of the conversation, the evidence collected, and what hiring teams do with it.
0%
of HR pros using AI in recruiting say it saves time or increases efficiency
SHRM 2025
0%
more likely to get a job offer when interviewed by an AI voice agent (one field study)
Field experiment, 70,000 applicants
0%
final-round pass rate from an AI-assisted pipeline, vs 34% traditional
RCT, 37,000 applicants
0%
of HR teams don’t formally measure the success of AI investments
SHRM 2026
01
The formats
How AI voice interviews work
An AI voice interview is a spoken conversation between a candidate and an AI interviewer. The system asks questions, processes spoken responses, and may ask relevant follow-ups. It can create a transcript, summarize responses, or organize evidence against predefined role criteria. The precise capabilities vary by product and configuration.
This is different from a one-way video interview, in which candidates typically record answers to a fixed set of prompts. A conversational voice interview can respond during the session. It is also different from a live interviewer copilot, which assists a human-led conversation rather than conducting the interview itself.
An AI avatar interview adds a visible digital interviewer to the voice conversation. The avatar has a chosen appearance and voice and can make the experience feel more like an employer-led interaction.
Keep in mind
The visual layer does not automatically make the assessment more accurate or fair. Evaluate the conversation and decision process beneath it.
Four interview formats, compared
From fixed prompts to a human-led conversation with AI assistance
AI voice interview
- Who leads
- AI interviewer
- Candidate experiences
- Speaks with a responsive voice system
- Hiring team should inspect
- Follow-ups, transcript accuracy, role criteria, and human review
AI avatar interview
- Who leads
- AI interviewer with a visual presence
- Candidate experiences
- Speaks with an on-screen interviewer
- Hiring team should inspect
- Everything in a voice interview, plus visual clarity and candidate experience
Human-led interview with an AI copilot
- Who leads
- Human interviewer
- Candidate experiences
- Speaks with a person while AI assists the panel
- Hiring team should inspect
- Whether guidance helps the interviewer collect and evaluate evidence
The formats can coexist. A team might use an AI avatar for an initial conversation and a human panel with a copilot for a later, deeper assessment.
Perspective 01
From experiment to everyday
Placeholder — add a pull quote from a talent leader on how AI interviews are moving into everyday hiring.
FN
Full Name
Designation, Company
02
The volume problem
Why voice and avatar interviews are gaining attention
High-volume hiring creates a practical constraint: every qualified candidate needs a meaningful chance to explain their experience, but recruiter calendars do not expand with application volume.
A well-designed AI interviewer can offer a consistent first conversation outside ordinary office hours and prepare structured material for human review.
This is a narrower and more useful claim than “AI makes better hiring decisions.” The immediate opportunity is to collect comparable, job-relevant information from more candidates. The next question is whether the collected information helps people make better decisions.
0%
say AI in recruiting saves time or increases efficiency
HR professionals whose organizations use AI in recruiting
Source: SHRM, 2025 Talent Trends. Applies to AI in recruiting broadly; it does not establish that every voice or avatar interview system improves hiring outcomes.
03
The evidence
What the research says about AI-led interviews
Two studies illustrate both the promise and the need for careful interpretation.
Study A · 2026 field experiment
70,000 applicants randomly assigned to human recruiters or AI voice agents
Likelihood of receiving a job offer
Indexed to the human-recruiter interview group (= 100)
Human recruiters evaluated the interviews and made the hiring decisions in both groups.
The researchers associated the result with more structured and consistent information collection. It does not mean a 12% offer gain should be expected at another employer, nor does it test every avatar design.
Study B · 2025 randomized study
37,000 junior-developer applicants; traditional vs AI-assisted pipeline
Pass rate at the same final human interview
Among candidates who reached the final round
+20 percentage points for the AI-assisted pipeline.
But the pool changed, too
The AI-assisted process tended to select applicants who were:
- Younger
- Less experienced
- Fewer advanced credentials
Why Study B matters twice
Structured interviews may alter the quality of the candidate pool reaching humans, and teams must examine who the system advances, not just how many people it processes.
Read with care
These are studies of particular processes, populations, and systems. They support evaluating AI-led interviews seriously; they do not prove that all AI voice or avatar interviews outperform human interviews. Evidence for an avatar’s visual presentation specifically is more limited than evidence about structured conversational information collection.
Perspective 02
Reading the research
Placeholder — add a pull quote from a talent leader on what the evidence on AI-led interviews does and does not prove.
FN
Full Name
Designation, Company
04
Interface vs. substance
The avatar is the interface; the interview design is the substance
An avatar can make the interaction recognizable and branded. Candidates may find a visible interviewer easier to follow than an audio-only interface, while others may prefer audio or a human alternative.
Those preferences should be tested with actual candidates rather than assumed.
For example, if “stakeholder communication” is a criterion, a useful interview asks about a real situation, the candidate’s action, and the outcome. It should not infer communication skill from accent, appearance, or how closely the candidate resembles a preferred presentation style.
A structured interview playbook helps teams specify what they intend to evaluate before the interview begins.
InterfaceAvatar · voice · branding
Substance — five design choices
01
Role criteria
What skills and experiences must the interview explore? Separate essential criteria from preferences before generating questions.
02
Questions
Do questions invite evidence about the work, or merely polished generalities? Use the same core assessment areas for candidates in the same role.
03
Follow-ups
Can the system ask for a specific example or clarify an incomplete answer without drifting into unrelated or inappropriate topics?
04
Evidence
Can recruiters see the relevant answer, transcript, and assessment rationale rather than only a summary score?
05
Review
Can a qualified person challenge a transcription error, reconsider a score, and make the final decision?
05
Candidate experience
Candidate experience is part of interview quality
Before the first question, candidates should know…
- That they are speaking with an AI system
- What the interview will cover
- Approximately how long it will take
- What information is recorded
- How people will review the result
- How to request an accessible route or accommodation
During the conversation, test whether the system…
- Allows a reasonable pause
- Handles a request to repeat a question
- Recovers from background noise
- Responds appropriately when an answer is unclear
- Preserves what the candidate actually said in the transcript
- Performs in every language you intend to use
A polished avatar cannot compensate for interruptions, inaccurate transcription, or irrelevant follow-ups.
Language support requires the same care. Test question quality, transcription, follow-ups, and evaluation in each language. “Supports a language” is not the same as “performs equally well for every speaker and role.”
Example · JayT by JobTwine
JobTwine’s AI Avatar Recruiter, JayT, is designed to conduct first-round conversations using configured questions and criteria and to prepare evidence for recruiter review. The hiring team remains responsible for deciding who moves forward.
This is an example of where an avatar can fit in a broader workflow, not a substitute for testing its performance in the employer’s own roles.
Perspective 03
Interface vs. substance
Placeholder — add a pull quote from a talent leader on why interview design matters more than the avatar.
FN
Full Name
Designation, Company
06
Beyond speed
What hiring teams should measure beyond speed
A voice interview can be available around the clock and reduce scheduling effort. Those are useful operational gains, but they are only the beginning of an evaluation.
SHRM’s 2026 State of AI in HR report found most organizations are not yet measuring AI investments rigorously. The figures refer to AI in HR overall, not specifically to AI interviews.
How HR teams measure AI investments
Share of HR professionals
16 of every 100 HR teams use their own ROI metric for AI
Source: SHRM, 2026 State of AI in HR.
The AI interview scorecard
Follow the candidate through the whole process — from first click to performance in the role
1
Access & completion
Can candidates start and complete the interview without avoidable friction?
EvidenceStarts, completions, technical failures, accommodation requests
2
Conversation quality
Does the interviewer ask relevant questions and useful follow-ups?
EvidenceSampled recordings and transcripts reviewed against the playbook
3
Evidence quality
Can reviewers locate the answers behind each assessment?
EvidenceAnswer-to-criterion mapping, transcript accuracy, reviewer notes
4
Recruiter efficiency
Does the system save time without shifting hidden work downstream?
EvidenceScheduling time, review time, re-interviews, exception handling
5
Candidate experience
Do candidates understand the process and feel able to show their skills?
EvidenceCandidate feedback and complaints, segmented where appropriate
6
Selection outcomes
Who advances, and do later human assessments support the initial evaluation?
EvidenceAdvancement rates, later-stage results, adverse-impact monitoring
7
Hiring outcomes
Are hires successful in the role?
EvidenceJob-related performance and retention measures over time
Compare these measures with a suitable baseline. A higher interview completion rate is valuable, but it cannot by itself show that the questions identify qualified candidates. An improvement in later-stage pass rates needs to be examined alongside the composition of the candidate pool and the criteria used to advance people.
Practical starting point
See JobTwine’s interview intelligence evaluation guide and adapt its pilot structure to voice and avatar interviews.
Perspective 04
Measuring what matters
Placeholder — add a pull quote from a talent leader on measuring AI interviews beyond speed.
FN
Full Name
Designation, Company
07
Human in the loop
Human review must be able to change the outcome
“Human in the loop” has little meaning if the reviewer only sees a score and clicks approve.
Rubber stamp
ScoreApprove
The reviewer sees a number and clicks through. AI effectively decides.
Meaningful review
EvidenceJudgmentDecision
The reviewer inspects evidence and has the authority to disagree.
A meaningful reviewer should be able to
- Inspect the candidate’s relevant answers and the criteria applied
- Correct transcription or interpretation errors
- Consider the context the system may have missed
- Override an assessment with a documented reason
- Escalate an accessibility, fairness, or technical concern
AI can shape a decision even when a human formally makes it: it determines what appears in a summary, which statements are highlighted, and sometimes which candidates receive attention first. Training reviewers to question that framing is as important as keeping a human approval step.
A flag is not proof
The same principle applies when interview fraud signals are present. A flag is a reason to investigate, not proof of misconduct. Examine the underlying evidence and give candidates an appropriate way to address a concern. Read more about JobTwine’s approach to candidate fraud detection.
08
Playbook
How to run a responsible AI interview pilot
Start with one or two roles where the team can define job-related criteria and review enough interviews to detect failures.
Agree in advance on the baseline, the measures above, the human decision owner, and the circumstances that require an alternative interview route.
Phase 1
Before launch
- Create the playbook
- Review questions for job relevance
- Test voice, avatar, language, accessibility, and recording experience
- Document candidate notice and data handling
- Train reviewers
Phase 2
During the pilot
- Sample complete conversations — strong, borderline, and unsuccessful
- Compare AI summaries and scores with the underlying answers
- Record candidate feedback and technical failures
- Log reviewer overrides and cases needing a human interview
Phase 3
After the pilot
- Compare results with the existing process
- Look at time, candidate experience, later-stage performance, and selection patterns
- Revise the playbook — or stop where evidence shows a problem
- Don’t declare success just because more interviews happened
Compliance note for U.S. employers
Selection procedures remain subject to applicable employment law when AI is used. The EEOC’s guidance on selection procedures is a starting point for understanding job-related validation and adverse impact. Specific legal duties depend on the employer, location, and use case; involve the appropriate legal and compliance teams in deployment decisions.
09
The operating model
Where AI voice and avatar interviews fit in the hiring workflow
The most useful pattern is a connected, reviewable process.
People lead AI conducts, people review
- 1
Define the role
Hiring teams agree on skills, criteria, and an interview playbook.
- 2
Invite the candidate
The candidate receives clear notice and a usable interview option.
- 3
Conduct the conversation
The voice or avatar interviewer asks core questions and relevant follow-ups.
- 4
Review the evidence
Recruiters inspect the transcript, answers, and structured feedback.
- 5
Continue human assessment
A hiring manager or panel explores areas requiring deeper judgment, potentially with an interviewer copilot.
- 6
Record the decision
The team documents its reasoning and keeps the hiring decision with people.
ATS & existing review process
Connected through JobTwine integrations — no separate evidence silo
Perspective 05
People own the decision
Placeholder — add a pull quote from a talent leader on keeping humans accountable for hiring decisions.
FN
Full Name
Designation, Company
10
Conclusion
What talent leaders should take from 2026
AI voice interviews create a way to conduct more structured conversations at a scale that would otherwise be difficult. Avatar interviews add a visible, configurable interviewer to that experience.
The strongest case for either format is consistent collection of job-relevant evidence that people can inspect and challenge.
Research suggests that AI-led or AI-assisted structured interviews can improve outcomes in particular settings. It also shows that technology can change who advances. That is why the operating model matters.
01
Define what the system may assess
02
Test the candidate experience
03
Measure outcomes beyond speed
04
Empower reviewers with the evidence and authority to disagree
The JobTwine principle
AI handles repetitive recruiting work. People retain the hiring decision.
See how an avatar interview and subsequent human review could work in your process.
Source and methodology note
This guide draws on SHRM’s 2025 recruiting AI findings (Talent Trends) and 2026 State of AI in HR report, plus two research papers: a 2026 field experiment on voice AI in firms (70,000 applicants) and a 2025 randomized study of AI-assisted structured video interviews (37,000 junior-developer applicants). The research papers evaluate particular interview systems and applicant populations; their results are not product-performance claims for JobTwine.