RecruitingAI screeningVoice AI

How to conduct an interview, from first minute to score

HireQwik November 9, 2026 11 min read

Across our customers’ hiring pipelines in India between March and August 2026, our AI interviewer held 2,589 completed interviews, about 475 hours of conversation. Building an interviewer that runs those conversations means writing down, explicitly, almost every rule a good human interviewer follows by instinct. Ask one question at a time. Draw questions from the job and the candidate’s own résumé. Don’t count an answer as a deal-breaker unless the candidate actually said it. Quote their words when you score.

None of that is new. Structured interviewing is one of the best-supported methods in the selection research. What writing the rules down shows is how many of them are easy to skip when nobody’s checking. This guide on how to conduct an interview is the list, written for the person on the interviewer’s side of the table: a hiring manager doing their first few interviews, a team lead who’s been asked to join a panel, or a recruiter training either of them.

What does it mean to conduct an interview?

Conducting an interview means running a planned conversation that gathers evidence about whether a candidate can do a specific job: preparing questions and a scoring scale in advance, asking the same core questions of every candidate, probing vague answers, taking notes on what was said, and scoring each answer soon after.

The word doing the work there is “evidence”. An interview that ends with “I liked him” has produced a feeling. One that ends with three scored answers and a quote under each has produced something a hiring manager, a debrief or a future you can actually use.

Before the interview: prepare in 20 minutes

You don’t need hours. You need these five things, in this order:

  1. Read the résumé and any earlier scorecard (5 minutes). If the candidate went through a screen, start where it ended. Don’t re-ask what’s already answered; note the one or two things it flagged as unclear.
  2. Pick your questions (5 minutes). Three to five main questions, each tied to something the job needs. Our guides to behavioral interview questions and situational interview questions cover the two most useful types.
  3. Read the anchors (3 minutes). What does a strong answer to each question contain? What does a weak one look like? If there are no written anchors, write two lines per question now.
  4. Plan the time (2 minutes). Allocate minutes per question and keep five for the candidate’s questions. Our interview agenda template does the arithmetic.
  5. Set up (5 minutes). Quiet room or a working video link, the notes template open, phone on silent. Be on time.

Google’s interviewer guidance puts the first and last of these into one rule: start on time, spend a few minutes building rapport, and leave at least five minutes for the candidate’s questions.

The first three minutes: how to open an interview

The opening sets how much the candidate will tell you. A nervous candidate gives short, safe answers; a candidate who knows what’s happening gives real ones. Here’s a script you can adapt:

“Hi, I’m Kavya, I lead the support team you’d be joining. Thanks for making time. We have 40 minutes. I’ll ask you about five questions, mostly about work you’ve done, and I’ll take notes as we go so I don’t miss anything. We’ll keep the last five minutes for your questions. Sound good? Let’s start with your current role. What does a normal day look like?”

That covers everything a candidate needs: who you are, how long, what kind of questions, why you’re typing, and when they get to ask. The first question is deliberately easy. It’s a warm-up, so don’t score it.

Keep the small talk short and the same for everyone. If you chat about cricket with one candidate and launch straight in with the next, you’ve already run two different interviews.

Asking questions: one at a time, then probe

One question at a time. “Tell me about a difficult customer, how you handled it, and what you’d do differently” is three questions, and most candidates will answer one of them. Ask the first, then follow up.

Use a probe ladder for vague answers. Most first answers are general. The skill of interviewing is getting from general to specific without leading:

Candidate saysProbe
“We improved the process.”“What did you personally change?”
“It went well.”“How did you know? What was different afterwards?”
“I usually handle it by staying calm.”“Tell me about the last time it happened. What did you actually say?”
“We had a lot of tickets.”“Roughly how many a day, and how many were yours?”
“I’d escalate it.”“To whom, and what would you tell them?”

Don’t lead. “You’d check with the customer first, right?” tells the candidate the answer. Ask “What would you do first?” and wait.

Let silence work. After a question, count to five before you rephrase. Many candidates need a few seconds to pick an example, and the answer after a pause is often the better one.

Steer long answers politely. If an answer runs past three or four minutes, interrupt with a specific follow-up rather than a “let’s move on”. We wrote about the right length for answers, and how to steer them, in interview answer length.

Talk less than 20 percent of the time. If you find yourself explaining the company for five minutes, that’s five minutes of evidence you didn’t collect. Save the pitch for the end, or for candidates you’re keen on.

Phone and video interviews: what changes

Most of the above applies whatever the medium, with three adjustments. On a phone interview you lose facial cues, so confirm understanding out loud (“Just to check I’ve got that right, you managed the queue yourself?”) and leave longer pauses, because candidates can’t see that you’re waiting for them. On a video call, agree what happens if the connection drops before you start, and look at the camera, not at your notes, when you ask a question. In both, tell the candidate when you’re typing, so a silence doesn’t read as disapproval. If the first round is a phone screen, our guide to the phone screen interview covers that format in more depth.

Taking interview notes you can actually score

Notes are where most interviews lose their value. The usual notes are adjectives: “confident”, “good communicator”, “a bit vague”. Those are conclusions, and they can’t be checked later. Write what the candidate said instead.

A simple interview notes template, one block per question:

QuestionWhat they said (keywords and short quotes)Probes askedScore (0-3)
Q1: A time you handled an angry customer“refund delayed 9 days”, “called back same day”, “got finance to approve partial refund”“What did you say first?”
Q2: …

Three habits make notes useful:

  • Capture numbers and names of things (tools, team sizes, volumes, dates). They’re the easiest details to verify and the hardest to make up.
  • Mark your probes. An answer that needed three probes to get specific is different from one that was specific from the start.
  • Leave the score column empty until the answer ends. Scoring mid-answer is how first impressions become final scores.

This is one place where recording helps a great deal. In HireQwik every AI interview is recorded and transcribed, so HR can check exactly what the candidate said, and the per-question scores quote the candidate’s own words. Human interviewers rarely have that luxury, which is why the quotes in your notes matter so much.

Closing the interview

The last five minutes decide how the candidate remembers the whole conversation. Give them real time for their questions, answer honestly, and tell them exactly what happens next and when. “You’ll hear from our recruiter by Friday” beats “we’ll be in touch” every time. If you want a full script, we’ve written a separate guide on how to end an interview.

Don’t signal the outcome. “You’d be great here” to a candidate the panel later rejects is the kind of thing candidates repeat to their friends.

After the interview: score within the hour

Score each answer against its anchor as soon as the candidate leaves, before your next meeting, and before you talk to anyone else who interviewed them. Then write one line of evidence per score. If you’re part of a panel or a multi-round process, bring those scores to the interview debrief meeting unchanged; that guide covers what happens next.

Why so fast? Memory of an interview fades quickly and gets rewritten by your overall impression. The specific answer to question three is gone by evening; the feeling that “she was strong” remains, and it will fill in the scores for you.

What not to ask in an interview

Every question should connect to the job. That rule alone rules out most of the risky ones. In the United States, federal law prohibits employment discrimination based on race, colour, religion, sex, national origin, age (40 or older), disability or genetic information, and the EEOC counts transgender status, sexual orientation and pregnancy as part of sex. In India, questions about caste, religion, marital status, pregnancy or family plans invite complaints and tell you nothing about whether someone can do the work.

Common traps, and what to ask instead:

Don’t askAsk instead, if it matters for the job
“Are you married? Planning a family?”“This role has rotational night shifts. Can you work them?”
“Where are you originally from?”“The role is based in Pune. Can you work from there by the joining date?”
“How old are you?”Nothing. If there’s a legal minimum age, check it in documents, not in the interview.
“Do you have any health issues?”“The job involves standing for most of an eight-hour shift. Is there anything you’d need to do that?”

How to train new interviewers

Google requires employees to complete formal interview training before they’re allowed to interview, and the content of that training is mostly what’s in this post: the process, the scoring, legal limits and how to write feedback. You don’t need Google’s scale to do the same thing. Here’s a four-step program that works for small teams:

  1. Learn the process (1 hour). The interview stages, the role’s question areas, the 0-3 scale with anchors, and what not to ask. Our post on interview process stages is a good pre-read.
  2. Shadow (two or three interviews). The new interviewer sits in, takes notes and scores silently, then compares their scores with the lead interviewer’s afterwards.
  3. Reverse-shadow (two or three interviews). The new interviewer leads; an experienced interviewer watches, scores, and gives feedback straight after.
  4. Calibrate. Have the new interviewer and two experienced ones score the same recorded interviews. Sign them off when their scores land within a point of the others on most questions.

Then keep calibrating. Interviewers drift: some get harsher over a long drive, some get kinder. Once a quarter, have the team score the same recorded interview and compare. The gaps tell you who needs a refresher, and sometimes they tell you an anchor needs rewriting.

Why structure and training at all? Because they’re what make interviews predict anything. The most cited recent review of selection research, a 2022 meta-analysis led by Paul Sackett and summarised by HumRRO, found that structured interviews had the highest average validity of any selection method it examined. Most of the steps above are just structure, applied consistently.

What building an AI interviewer taught us about conducting interviews

I’ll end where I started. When you have to write every interviewing rule down for software, you find out which ones matter. Four stood out:

  • Questions come from two places: the job and the candidate’s résumé. In HireQwik the interview questions come from the job and from the candidate’s own résumé. Questions from the job keep it fair; questions from the résumé keep it specific and make rehearsed answers harder.
  • A deal-breaker has to be heard, not assumed. A deal-breaker answer counts only if it matches what the candidate actually said; otherwise the candidate goes to On Hold for a person to check. Human interviewers assume all the time (“she’s in Delhi, she won’t relocate”). Ask.
  • Every score needs a quote. Each AI interview ends with a recommendation (Strong Go, Go, On Hold or No Go) and per-question scores that quote the candidate’s own words. If a score can’t point to something the candidate said, it’s an opinion.
  • The interviewer recommends; someone else decides. The AI recommends and HR makes the final call. The same separation helps human interviewers: your job in the room is evidence, and the decision comes after, against a rule.

To see those four rules applied from the candidate’s chair, try our live AI interviewer and compare how it opens, probes and closes with the checklist above. To see what it hands your human interviewers afterwards, book a demo.

Frequently asked questions

How do you start an interview as the interviewer?

Start on time, introduce yourself and your role, explain the format in one or two sentences (how long, how many questions, that you will take notes, and that there will be time for their questions), then ask an easy first question about recent work. Keep small talk short and the same for every candidate, and never score it.

What should an interviewer not do?

Do not ask questions that differ from candidate to candidate without a reason, decide in the first minute, talk more than the candidate, ask two questions at once, score from memory hours later, or ask about anything unrelated to the job such as religion, caste, marital status, pregnancy or family plans. These lead to weak and sometimes unlawful decisions.

How do you train new interviewers?

Use four steps: teach the process and the scoring scale, have the new interviewer shadow two or three interviews and score them silently, then reverse-shadow (they lead while an experienced interviewer watches), and finally calibrate by comparing their scores with experienced interviewers on the same candidates. Sign them off when their scores land within a point of the experienced ones.

See your own candidates screened

Book a 30-minute demo. Bring a live JD and we'll screen against it, then start with a pilot on your own candidates before committing to anything.