Voice Screening Questions for Hiring: 25 Questions That Show How Candidates Think

The best screening question is the one every candidate hears, word for word. A 2022 re-analysis of a century of hiring research ranked structured interviews as the single strongest predictor of job performance, with a mean validity of .42. Voice screening questions let you run that structure before the first call, so you hear how a candidate reasons, not just what their CV claims.

Same questions, same rubric, scored on substance. Skip any of the three and the evidence stops applying. Below are 25 questions grouped by what they test, a rubric you can copy, and the bias risks we think most voice screening tools play down.

Why structured questions beat a CV skim

Most early screening is a CV skim and a 15-minute call where the recruiter asks whatever comes to mind. That second part is an unstructured interview, and the research on it is not kind.

In 1998, Frank Schmidt and John Hunter summarised 85 years of selection research in Psychological Bulletin. They put the average validity of structured interviews at .51, against .38 for unstructured onesSchmidt and Hunter 1998, Table 1. Predictive validity for overall job performance, corrected meta-analytic estimates across 19 selection procedures.. Structured meant the same job-derived questions for every applicant and a fixed scoring procedure. Unstructured meant neither, with only an overall impression at the end.

That paper shaped hiring advice for two decades. It also used statistical corrections that later researchers thought were too generous.

Paul Sackett, Charlene Zhang, Christopher Berry and Filip Lievens revisited those estimates in the Journal of Applied Psychology. Their argument was technical: many meta-analyses had over-corrected for range restriction, which inflated validity. Fixing that pulled most estimates down by .10 to .20.

Structured interviews came out on top at .42, ahead of job knowledge tests at .40 and cognitive ability tests at .31Sackett, Zhang, Berry and Lievens 2022, Table 3 and discussion. Structured interview mean operational validity .42, residual SD .19, lower 80% credibility value .18.. Unstructured interviews fell to .19. The authors suggest structured interviews, not cognitive ability, should now be the reference point other methods are judged against.

What .42 does and does not mean

A validity of .42 is a correlation, not a hit rate. It means structured interview scores and later performance ratings move together moderately, far better than chance and far from certain. The same paper reports a large spread around that mean, with a residual standard deviation of .19. Some structured interviews work very well and some barely work at all.

We read that spread as the most useful finding for anyone writing screening questions. The format is not magic. The quality of the questions and the discipline of the scoring decide which end of the range you land on.

What makes a voice screen a structured interview

A voice screen is a short form where candidates record spoken answers to set questions, on their phone, whenever suits them. It is close to a one way interview, minus the camera. Whether it counts as a structured interview depends on four things, and all four are under your control.

1. Questions come from the job. Each one maps to something the role actually requires, written down before you open applications.
2. Every candidate gets the same questions in the same order. No improvising, no extra questions for candidates you already like.
3. Answers are scored against anchors written in advance. You decide what a 1, 3 and 5 sound like before you hear anyone.
4. Each answer is scored on its own. Score question 1 for every candidate, then question 2, instead of forming one overall impression per person.

A recorded format makes rules 2 and 4 easier than a live call does. Nobody drifts off script, and you can listen to every candidate's answer to the same question back to back. That is hard to do with notes from 30 phone calls.

Voice, phone, video or written: what each screen is good at

The phone screen still wins one thing outright: you can ask "why?" a second time. If a role's shortlist is 5 people, a structured call is the better tool. Voice earns its place when you have 50 to 500 applicants and need a fair first cut. We compare voice and video tools in more depth in our voice-first vs video-first comparison.

25 voice screening questions, grouped by purpose

Pick 3 to 5 of these for a single role, not all 25. Choose one group as the backbone, usually past behaviour, and add one question from another group. Rewrite any question so it names a real situation from your job.

Motivation and fit (questions 1 to 4)

Use one of these at most. Motivation questions are easy to rehearse and predict less than behaviour questions, but they relax candidates before harder prompts.

1. What made you apply for this role, and what do you expect the hardest part of it to be?
2. Describe a piece of work you are proud of. What exactly was your part in it?
3. What would you want to have learned after your first 6 months here?
4. Tell us about a job or project that did not suit you. What did you learn about how you work?

Past behaviour (questions 5 to 10)

These ask what the candidate actually did, which is the strongest kind of interview question. Ask for one specific example, not a general habit.

5. Tell us about a time you had to learn a new tool or process quickly to meet a deadline. What did you do first?
6. Describe a mistake you made at work that someone else noticed. What happened next?
7. Tell us about a time you disagreed with a manager or teammate. How did it end?
8. Describe a task you finished that was badly defined at the start. How did you work out what was needed?
9. Tell us about a time you had too much to do. What did you drop, and how did you decide?
10. Describe the most difficult customer, client or colleague you have dealt with. What did you say to them?

Situational judgement (questions 11 to 15)

These present a realistic scenario and ask what the candidate would do. They work well for entry-level roles, where candidates may not yet have the experience behaviour questions assume.

11. A customer is angry about a late delivery that was not your fault. What do you say in your first 30 seconds?
12. You notice a colleague has made an error that will reach a client tomorrow. They have left for the day. What do you do?
13. Your manager gives you two urgent tasks and you can finish only one today. Talk us through how you decide.
14. You are halfway through a task and realise the instructions you were given are wrong. What do you do next?
15. A new teammate keeps asking you the same question. How do you handle it?

Explaining the work (questions 16 to 20)

Spoken explanation shows reasoning that a CV cannot. Adapt these to the role's real content, and keep the topic inside what a qualified candidate should know.

16. Explain one tool you use often to someone who has never used it.
17. Walk us through how you would check whether a piece of your own work is correct before handing it over.
18. Pick a term from your field that newcomers misunderstand and explain it simply.
19. Describe the steps you would take in your first week to understand how this team works.
20. Tell us how you would explain a delay to a client who is not technical.

Spoken communication, only if the job needs it (questions 21 to 23)

Use these only when speaking to customers or the public is a core duty, such as support, sales or field roles. Score clarity and structure, which you can define. Do not score accent, which you cannot fairly define.

21. Imagine we are a customer calling for the first time. Introduce yourself and ask what we need.
22. Explain our returns policy, pasted below, as you would to a customer on the phone.
23. A caller cannot hear you well on a bad line. Repeat your last answer as if speaking to them.

Closing the loop (questions 24 and 25)

These cost little and tell candidates the process is two-way. Treat them as optional and do not score them.

24. Is there anything about your experience that these questions did not give you a chance to mention?
25. What would you like to know about the role before the next step?

Leave notice period, location, work authorisation and salary range as tap or short-text fields. They have one right answer and take 2 seconds to tap. Recording them wastes the candidate's effort and your listening time.

How to score answers without a gut feel

A question without a rubric is an unstructured interview with better audio. Write anchors for each question before applications open, and score every answer on the same 1 to 5 scale.

Notice what the anchors leave out: fluency, confidence, accent, speaking speed and vocabulary. A nervous candidate with a strong example should beat a smooth one with no example. If you score spoken communication for a customer-facing role, use only questions 21 to 23 and anchor on structure and clarity.

Keep scorers independent. Have two people score the same answers separately, then compare. Large disagreements on a question usually mean the anchors are vague, not that one scorer is wrong. Score one question across all candidates before moving to the next, which limits the halo effect of a strong first answer.

Record the rubric next to the audio. Export answers to a spreadsheet and add a score column per question and scorer. That gives you an audit trail of what each decision rested on. It also gives you the data to check pass rates across groups, which the next section argues you should. We cover the sheet setup in sending form responses to Google Sheets.

How to run a voice screen this week

Here is the process end to end. It takes an afternoon to set up for one role.

In Would You Rather, each voice question has optional minimum and maximum lengths, and candidates can re-record before moving on. A per-question "type instead" box is off by default; we would switch it on for screening, for reasons covered below. Section branching lets you route candidates to role-specific questions based on an earlier answer, which we explain in conditional logic for forms. The tool stores and plays recordings. It does not transcribe or score them, and for screening we think that is the right default.

Where voice screening goes wrong

This is the section most vendors skip. A voice screen adds a new channel for bias, and the structure that makes interviews predictive does not remove it on its own.

Accent and voice are not job skills

A voice recording carries information you should not use: accent, pitch, age cues, a stammer, the sound of a regional language under the English. Most of it tracks national origin, gender or disability far more than performance.

US law is explicit here. The EEOC's national origin guidance says an employment decision may be based on accent only if effective spoken English is required for the job and the accent materially interferes with the ability to communicateEEOC Enforcement Guidance on National Origin Discrimination, issued 18 November 2016, section on accent discrimination. Opened 8 October 2026.. A noticeable accent is not enough. Our rule is stricter: do not score accent at all, and score clarity only for roles where speaking is the job.

Automated scoring inherits the recogniser's errors

If software transcribes or scores answers, its errors are not evenly spread. Koenecke and colleagues tested 5 commercial speech recognition systems on 19.8 hours of interviews in PNAS. The average word error rate was 0.35 for Black speakers and 0.19 for white speakersKoenecke et al., PNAS 117(14), 7 April 2020. 42 white and 73 Black speakers, 5 US cities, audio matched on age and gender. Systems from Amazon, Apple, Google, IBM and Microsoft., and the gap persisted on identical phrases. A model that scores a transcript scores those errors too.

That study is US English from 2020, and recognisers have improved since. We have seen no evidence that the gap has closed for Indian English, code-switched speech or speakers of regional languages, which is the population many Indian employers screen. We would not let software reject a candidate based on a voice answer.

Adverse impact is the employer's problem, even with a vendor tool

In the US, the Uniform Guidelines on Employee Selection Procedures set the four-fifths rule. A selection rate for any race, sex or ethnic group below four-fifths of the highest group's rate is generally treated as evidence of adverse impact29 CFR 1607.4(D), Uniform Guidelines on Employee Selection Procedures (1978), current eCFR text opened 8 October 2026.. In May 2023 the EEOC published technical assistance applying this to algorithmic tools. It said employers may be responsible for a discriminatory tool even when a vendor designed or administered it, and called the four-fifths rule only a rule of thumb.

That document no longer appears on eeoc.gov; the page returned "not found" when we checked on 8 October 2026, and law firms report it was removed in early 2025. The guidance was never binding. The Title VII obligations and the Uniform Guidelines it explained are still in force.

India: consent and data duties now apply to recordings

India has no single law equivalent to Title VII for private hiring. The Code on Wages, 2019, in force from 21 November 2025, bars discrimination on the ground of sex in recruitment for the same or similar work. A voice recording that identifies a candidate is personal data under the Digital Personal Data Protection Act, 2023. Its Rules were notified on 13 November 2025, with notice and consent duties phasing in over 18 months.

Lawyers disagree on whether candidate screening falls under the Act's employment exemption from consent. The low-cost answer is to give a clear notice anyway: what you record, why, who listens, how long you keep it, and how to ask for deletion. We are not lawyers, and an employment lawyer in your jurisdiction should check your setup.

Where this argument is weakest

The validity figures come from meta-analyses of in-person and phone interviews, mostly in the US and Europe. No study we found measures the predictive validity of short asynchronous voice screens specifically. We are reasoning that structure carries over to a recorded format, and that is an inference, not a finding.

Voice screens may also change who applies. Our analysis of voice survey drop-off shows that voice-only questions lose more people than text questions in survey research. If candidates who drop out at a microphone differ by age, disability, or comfort with English, the screen biases the pool before anyone scores it. A "type instead" option reduces that risk and costs you little.

Finally, the 2022 re-analysis has critics. Some researchers argue its range restriction choices under-correct, which would push estimates back up. The ranking of structured interviews near the top has held across both camps, so we rely on the ranking more than on the exact .42.

Where to start for your next role

Take the next role you are hiring for. Write down the 3 things a new hire must do well in month one. Turn each into one question from the bank above, write the anchors, and add one closing question. That is a 4-question voice screen with a 6 to 8 minute ceiling for the candidate.

Run it alongside your current process for one role before replacing anything. Compare who your phone screen shortlisted with who scored highest on the rubric. Where they differ, listen again and ask which evidence you trust more. If candidates answer in more than one language, read collecting answers in Hindi, Tamil or any language before you launch. If speaking ability itself is what you need to test, use an online speaking test instead of a screening question.

References

1. Schmidt, F. L. and Hunter, J. E., The Validity and Utility of Selection Methods in Personnel Psychology, Psychological Bulletin 124(2), 1998. Used for the 1998 validity estimates and the definition of structured and unstructured interviews.
2. Sackett, P. R., Zhang, C., Berry, C. M. and Lievens, F., Revisiting Meta-Analytic Estimates of Validity in Personnel Selection, Journal of Applied Psychology 107(11), 2022. Used for revised validity estimates, residual SD and the reframing around structured interviews.
3. Koenecke, A. and colleagues, Racial Disparities in Automated Speech Recognition, PNAS 117(14), 2020. Used for word error rates by speaker group.
4. US EEOC, Enforcement Guidance on National Origin Discrimination, 18 November 2016, opened 8 October 2026. Used for the accent standard.
5. US EEOC, Select Issues: Assessing Adverse Impact in Software, Algorithms, and Artificial Intelligence, 18 May 2023, archived copy. Used for vendor responsibility and the rule-of-thumb status of the four-fifths rule.
6. 29 CFR 1607.4, Uniform Guidelines on Employee Selection Procedures, eCFR, opened 8 October 2026. Used for the four-fifths rule text.
7. Government of India, Code on Wages, 2019, section 3, and Digital Personal Data Protection Act, 2023, with the DPDP Rules notified 13 November 2025. Used for Indian recruitment and data duties, read through law-firm summaries rather than the Gazette text.

The weakest point in this source base: no study here tests asynchronous voice screens directly, so the case for them rests on research about interviews in general, and the India section relies on secondary legal summaries.