What is a one-way video interview, and what other formats are there?
A one-way video interview is a set of preset questions a candidate answers on camera, alone, at a time they choose, and a reviewer watches the recordings later. AI interviews use the same one-way pattern by text or voice too, and some tools hold a live conversation instead. Which format each question allows is your call.
A one-way interview is answered alone and watched later; a live video call has a person on the other side, and a conversational interview has an AI interviewer that asks follow-ups as the candidate talks. What a one-way interview is, and how it compares with a live one sets out the three side by side.
The "AI" in an AI interview is usually the scoring, and sometimes the interviewer. The candidate still answers questions a recruiter chose or approved. What changes from tool to tool is how the answer is given, and whether anything talks back.
Text, voice and video
In a text interview the candidate types each answer. The reviewer gets the words exactly as written, which suits roles where writing is the job. Some tools also let the candidate upload a document instead, such as a written task.
In a voice interview the candidate records audio in the browser or an app. The reviewer gets the recording and, on most tools, a transcript. There is no camera, so a candidate on a phone in a shared room can still answer.
In a video interview the candidate records on camera. The reviewer gets the recording and a transcript, and can see how the candidate presents. It asks the most of the candidate: a working camera, a quiet room and enough light.
Where software scores a voice or video answer from its transcript, not the sound or the picture, the format changes what you can watch back, and what a cheating check can look at, more than it changes what gets scored.
No tool we checked publishes a time saving or a completion rate with a method behind it, so this guide quotes neither.
How is an AI interview scored?
An AI interview is scored against a scorecard: each criterion gets a score from the answers linked to it, and the scores are weighted into a total. A pass-or-fail rule, such as a notice period limit, can override the total, and the total maps to a recommendation. You approve or override that recommendation.
Any tool should be able to show you the whole chain, in this order:
- A scorecard lists what the role needs, each item with a weight.
- Each question is linked to the criteria it tests.
- The candidate's answers are the evidence.
- Each criterion gets a score, with a written reason.
- The scores are weighted into a total.
- The total, and any pass-or-fail rule, gives a recommendation.
If a tool can't show you one of these steps, you can't check its score.
The arithmetic is a weighted average: multiply each criterion's score by its weight, add the results, and divide by the total weight. Recagn publishes a worked example on a scorecard, which Recagn calls a rubric, with four criteria weighted 30, 25, 15 and 30. Scores of 4, 2, 3 and 4 give (4×30 + 2×25 + 3×15 + 4×30) ÷ 100 = 3.35 out of 5. That rubric's cut-offs are 3.5 for Qualified and 2.75 for Maybe, so 3.35 is Maybe. The 3.35 is a recommendation; the recruiter approves or overrides it.
A pass-or-fail rule sits outside the average. A candidate who can't join inside the client's notice period fails it, whatever the total says, and the recommendation follows the rule.
Ask these of any tool's scoring before you trust a number from it:
- Does each score come with a written reason you can check against the answer?
- Is the total worked out by a fixed rule, or does the model produce it?
- Are the weights and cut-offs yours to set before the first interview?
- Is a check for cheating kept apart from the score, or mixed into it?
- Is a voice or video answer scored from its transcript, and can you read that transcript?
Each step, with a worked example, is set out in how AI interview scoring works.
A score reads the answers, not the person. It can miss what an answer meant, which is why the written reason belongs beside it, and why a person reads both before deciding.
What should an AI interview decide, and what should a person decide?
An AI interview should recommend, and a person should decide who is invited, which answers count and who is rejected. The software can run the interview and score it against the scorecard; whether a score or a flag holds needs someone who knows the client. Keep rejection as your own step, never an automatic one.
| The software can | A person decides |
|---|---|
| Run the same questions for every candidate, at any hour | Which questions, and which formats each allows |
| Score each answer against the scorecard, with a reason | Whether the reason holds |
| Flag signs of cheating | Whether a flag matters, after asking the candidate |
| Recommend who moves on | Who is rejected, and who the client sees |
Source: this guide's reading of the steps above · not a legal requirement
A rejection decides something about a person's next job. If a tool rejects on its own, a scoring error becomes a lost candidate, and nobody has read the answer that caused it. When a person makes the call, the same error is a recommendation someone can overrule.
The same goes for what the client sees. A score sent straight to the client becomes the agency's view before anyone at the agency has checked it.
Two questions show where a tool draws the line:
- Can the tool reject a candidate, or send a rejection email, without a person clicking?
- Can the client see scores before you've reviewed them?
In India, the Digital Personal Data Protection Act's duties for recruiters, such as notice and consent, start on or about 13 May 2027, under the commencement notification of 13 Nov 2025. The Act gives candidates no right to a human review of an automated decision, so a person deciding is a design choice, not a legal requirement. This is not legal advice.
The DPDP Act and recruitment sets out the notice, consent and dates for an agency.
AI interview vs human interview: what changes for the candidate and the hire?
In one field experiment, candidates interviewed by an AI voice interviewer were 12% more likely to get a job offer than those a recruiter interviewed. Recruiters still made the hiring decisions, in a working paper by Luca Henkel and Brian Jabarian covering 70,000 job candidates at one recruitment company, as Erasmus University reported on 19 Aug 2025.
The study assigned candidates at random to a human recruiter, to an AI that held a live voice conversation, or to a choice of the two. It tested conversational voice interviews at one firm. It did not test one-way recorded interviews, so its results don't carry over to every format above.
Two more findings come from the same source. When given the choice, 78% of candidates preferred the AI interviewer. Technical problems hit about 7% of the AI-led interviews.
Recruiters rated candidates' performance higher in the AI-led interviews. When they decided, though, they gave more weight to the standardised test scores, according to the university's summary of the paper. The interview supplied evidence, and the people used it alongside other evidence.
A human interview is still the better tool for some work:
- Selling the role to a candidate who isn't looking.
- Negotiating CTC and notice period.
- Senior hires, where the conversation is part of the assessment.
- Judging how a candidate will get on with a client's team.
An AI first round helps in a different way. Every candidate gets the same questions, answers at a time that suits them, and leaves answers that a client can read later. That is a description of how it works, not a measured result.
Either way, the interview gathers evidence; you decide what it means.
What do candidates see in an AI interview?
Candidates should see, before the first question, that AI evaluates their answers and that a person makes every decision. A good AI interview then lets them answer at a time they choose, re-record before saving, and know how long each answer can run. What candidates are told depends on your tool and your wording, so read the consent screen yourself.
The candidate usually gets a link by email or message, opens it on a phone or a laptop, and agrees to how their answers are used. Then the questions come one at a time. Each one says which formats it allows, and voice and video answers have a time limit.
A candidate who doesn't know AI is involved can feel misled when they find out. A candidate who is told plainly, and told that a person decides, knows what they are agreeing to before they start.
Take these questions to any vendor, and open its candidate screen as a candidate would:
- Is AI evaluation disclosed before the interview starts, in plain words?
- Is it clear that a person makes the decision?
- Can the candidate re-record, and is there a time limit per answer?
- Does it work on a phone, and on a weak connection?
- What happens to the recordings, and for how long are they kept?
Recagn's version is on what candidates see and agree to.
Can candidates cheat an AI interview?
Yes, some can: by reading a prepared or AI-written answer, by looking up answers off-screen, or by having someone else take the interview. Software can flag the first two from the answers and, on video, the picture; a stand-in who answers well is much harder to catch without an identity check. A flag is a reason to ask the candidate, and you decide whether it matters.
Cheating
A check on the answers can flag text that reads as AI-written, an answer read aloud from a script, or a claim that contradicts the CV. On video it can also flag eyes that keep moving to the same spot off-screen. Some help leaves no trace in the words or the picture, so a flag is a reason to ask the candidate, not proof. The ways candidates cheat, what any tool can see of each, and what to do with a flag are in the guide to interview cheating.
Proxies
A proxy is someone else answering in the candidate's place. An AI interview without an identity check can't confirm who is speaking or typing, however good its other checks are.
Three habits help in any process, with any tool:
- Compare the voice or face in the first round with the live client interview.
- In the live round, ask about one specific answer from the first round.
- Check ID before an offer, not before the first round.
The question to ask a vendor is short: do you verify identity, and when? A tool that checks ID at the start adds a step for every candidate. A tool that doesn't leaves the check to you, later in the process.
What does one AI interview cost in India: per interview or a flat plan?
Published AI video interview prices in India run about ₹185–500 on a vendor's smallest paid monthly plan or single purchase; a plan is cheaper once its credits are used. From vendors' own pages, 10 Nov 2026, GST not stated; Hyring Free and yearly billing go lower. Recagn costs at most ₹149.95 a video interview on Starter (₹176.94 with 18% GST).
| RecagnOur product | Hyring | Interviewer.AI | UnoJobs | |
|---|---|---|---|---|
| How you pay | Flat plan, monthly Starter, ₹2,999 a month | Flat plan or pay-as-you-go Pro, $99 a month billed monthly | Flat plan Essential, $199 a month billed monthly | Per interview |
| Per video interview | at most ₹149.95 a video interview on Starter (₹176.94 with 18% GST) | About ₹185 $99 ÷ 50 credits | About ₹372 $199 ÷ 50 credits | ₹500 flat, 20-minute interview |
| One more after the plan | Top-up Pack, paid plans only | $5 (about ₹468) 1–99 credits | $3 (about ₹281) pack of 50 at $150 | ₹500 |
| GST | Before 18% GST, GST figure shown | GST not stated | GST not stated | GST not stated |
Source: each vendor's own pricing page, checked 10 Nov 2026 · USD at ₹93.56 (Wise mid-market rate, 17 Nov 2026) · Recagn from its plan catalogue
The prices are from Hyring's pricing page, Interviewer.AI's pricing page and UnoJobs' AI interview page. Hyring and Interviewer.AI bill in US dollars, so their rupee figures move with the exchange rate.
A per-interview price is simple, and it suits a desk that runs a handful of interviews a month. A flat plan's per-interview figure assumes every credit is used. It is also a ceiling on what the interview itself costs, because the plan pays for other things too, such as seats, job posts or CV reads.
The breakeven is plain arithmetic on the table: divide the plan price by ₹500. Hyring Pro, about ₹9,275 a month, costs less than UnoJobs from about 19 interviews a month. Interviewer.AI Essential, about ₹18,644, costs less from about 38. Both figures assume the same GST treatment on each side, which none of the three pages states.
Two exceptions sit below the range. Hyring's Free plan includes 50 credits a month at $0, with one seat and one active job post. Hyring Pro billed yearly is $79 a month, about ₹148 a video interview, against $99 billed monthly (checked 10 Nov 2026).
Format changes the price too. On Recagn a text or voice interview uses one credit, a third of a video one, so it costs at most ₹49.98 a text interview on Starter (₹58.98 with 18% GST). Interviewer.AI's conversational interview uses two credits against one for its async video interview.
Every vendor we checked, yearly billing and the cost per candidate are in AI interview pricing in India.
For who each tool suits, the AI interview tools in India list sets eight of them side by side, grouped by how each is sold.
When should you not use an AI interview?
Skip an AI interview for senior or executive search, for a role with only a handful of candidates, and where the first call has to sell the role. It also suits candidates poorly when many of them lack a quiet place to record or a steady connection. Decide which calls you keep before you switch.
An AI first round helps most when many candidates answer the same questions. With five candidates, you will speak to each of them anyway, and a scored round in front of that call adds a step without saving one.
Better for a senior search with five candidates: calling them yourself.
Better for a desk that wants candidates phoned: an outbound phone screener such as Hyring's AI Phone Screener, which calls each candidate itself (checked 10 Nov 2026).
Better for roles where the client wants live probing from the start: a person, or a conversational tool, for the first round.
How Recagn does this
Recagn is our product. Here is one real run through it, broken down by format. The role is a Digital marketer in Bhubaneswar, adapted from a real public job posting with the company details removed. Recagn drafted the rubric and its questions from the brief. Recagn won't attach a candidate to a role until its rubric is locked, and editing and locking it is the recruiter's step.
Recagn scored the run on 8 Nov 2026. The four candidates are invented, and their answers were performed from scripts. The scored voice answers were generated by text-to-speech from those scripts.
The formats in one run
The role had seven questions, and each question set which formats it allowed. Running the interviews came off the recruiter's plate. The questions, and which formats each allowed, stayed theirs.
| Format | Questions that allowed it | How the four candidates answered | What Recagn scored |
|---|---|---|---|
| Text | 6 of 7 | Question 6 only, the ad copy: all four typed it (4 answers) | The typed text, as written |
| Voice | 6 of 7 | The other six questions, all four by voice (24 answers); question 2 allowed voice only | The transcript of each recording |
| Video | 5 of 7 | None in the scored run | Nothing: the clip below is a performed take, not scored |
| Document | 6 of 7 | None | Nothing |
Source: Recagn worked example, scored 8 Nov 2026 · adapted from a real public job posting; company details removed · invented candidates · voice answers generated by text-to-speech from a script
Take Ujjawal P., the candidate Recagn recommended as Qualified. His voice answer to question 4 feeds Analytics, Tracking & Reporting (weight 22%), which Recagn scored 4 out of 5 from his answers to questions 2, 3 and 4. His typed ad copy for question 6 feeds Written Communication & Ad Copy (weight 10%), also scored 4 out of 5. The Qualified is Recagn's recommendation; the decision is the recruiter's, and none was made in this run.
Every interview in the run used one credit, because none had a video answer. One video answer would have made that whole interview three credits. Recagn charges per interview, never per answer.
The same answer on video
The clip sequence
- 0–4s. Question 4 of 7:
How do you set up conversion tracking across GA4, Tag Manager and Meta Pixel, and how do you turn weekly/monthly data into decisions? Give an example of an insight that changed what you did.
- 4–36s. Ujjawal chooses Video recording and answers on camera:
I set GA4 events for form submits and call clicks, fire them through Tag Manager, and mirror them with the Meta Pixel. I found form conversions were double-counted because the thank-you page fired the tag twice. After I fix it, one Google campaign that looked cheap was actually average, so I moved budget to the better one. The bug show up when GA4 had more conversion than the CRM had leads. Thank you.
- 36–41s. He stops, then Save & next. His answer is saved and question 5 of 7 opens.
Screen recording · answer performed from a script · invented candidate
The clip is a person performing Ujjawal P.'s scripted answer to question 4 on video, in Recagn's real interview screen. The take adds one more sentence at the end. It was not scored, so it shows what a candidate sees and does, not a result.
How the scoring is checked
Recagn scores each criterion from 1 to 5 from the transcript of what was said, with a written reason. Its code, not the AI model, then works out the weighted total and the recommendation, and a failed hard gate makes the recommendation Disqualified whatever the total.
Scoring answers against the rubric comes off your plate. Whether a score holds stays yours: you approve the recommendation or override it before you set a decision. Recagn recommends. Only you can reject a candidate.
What Recagn checks for cheating, and what it doesn't
A separate authenticity check reads the whole interview for answers read aloud, signs of reading off-screen and AI-written text. It gives a reason for each finding and a 0 to 10 likelihood that sits beside the score. It is never added to the score, and it never rejects anyone.
On video answers, Recagn also checks sampled frames for looking away at off-screen material and for reading patterns. A CV check marks each claim on the CV as backed up, inflated or contradicted by the answers.
When you switch them on, Recagn also proctors the interview. On every plan it records when a candidate leaves the page or switches tabs, and what they copy or paste. From Starter it adds webcam photos every 20–30 seconds and a face match between the candidate's ID and a selfie; from Growth, webcam video and a recording of the whole screen. Recagn's rules turn these into a short list of findings, such as a second face in a photo or long time away, which sits beside the score and is never added to it. The hiring client never sees it.
Recagn doesn't lock the browser or force fullscreen, and its face match doesn't check that the ID is genuine, so a stand-in with a matching ID can still pass. It also doesn't phone candidates or hold a live conversation: it asks at most one follow-up question, on typed answers only, on questions where you turn follow-ups on.
Checking CV claims and flagging likely cheating come off your plate. Whether a flag matters stays yours.
What candidates agree to
Before the first question, candidates see one screen that lists everything the interview records, from their answers to any webcam, screen or ID check. It says the recruiter sees these records and the hiring company doesn't, and that a person makes every decision. Candidates tick that they agree and that they are 18 or older. What candidates see and agree to sets out the rest, including how long each record is kept.
A candidate can re-record a voice or video answer before saving it, unless you cap the number of takes. Each voice or video answer runs up to 3 minutes unless you set another limit. Candidates never pay Recagn.
How Recagn runs the whole first round, with prices by format, is on Recagn's AI interview software page.
Questions people ask
Is an AI interview the same as an AI phone screen?
No. In an AI interview the candidate opens a link and answers in the browser or an app, at a time they choose; an AI phone screen calls the candidate, as Hyring's AI Phone Screener does (checked 10 Nov 2026). The phone screen is a call the candidate takes when it rings, while the interview is answered from the link, usually to fixed questions.
Can Recagn tell if someone else took the interview?
Partly. With an ID check on, from Starter, Recagn compares the face on the candidate's ID with a selfie and, with webcam photos on, with the face during the interview, and lists a mismatch or a second face. It doesn't check that the ID itself is genuine, so it can't rule out a stand-in with a matching ID. Recagn's rules list what happened beside the score; you decide whether it matters.
Does Interviewer.AI charge more for a conversational AI interview?
Yes. Interviewer.AI uses 2 credits for a conversational AI interview against 1 for an async video interview, and offers conversational interviews from Professional up, not on Essential. Professional is $399 a month billed monthly, or $3,990 a year billed annually, for 120 credits a month, on its pricing page on 10 Nov 2026, prices in USD, GST not stated. At the monthly price that is about ₹622 a conversational interview if every credit is used, at ₹93.56 to the dollar.
Does a video interview cost more than a text or voice one on Recagn?
Yes. A video answer makes the whole interview three credits instead of one, so on Starter it costs at most ₹149.95 (₹176.94 with 18% GST) against ₹49.98 by text or voice (₹58.98 with 18% GST). Other vendors price by format too: Hyring charges 1 credit for a video interview and half a credit for a phone screen (checked 10 Nov 2026). AI interview pricing in India compares vendors format by format.
How long is an AI interview?
As long as the vendor or the recruiter sets it: UnoJobs runs a 20-minute AI video interview, and Interviewer.AI sets 20 minutes per interview on Essential and 30 on Professional (each vendor's own page, checked 10 Nov 2026). Neither publishes how long candidates actually take; the length follows the number of questions and the time allowed for each answer.
Does Recagn score a candidate's accent or appearance on video?
No. Recagn scores the transcript of what was said, not the voice or the picture, and its scorer is told not to use or infer protected attributes such as name, age, gender or photo. Frames from video answers are used only for the cheating check, which is told not to infer identity, appearance or accent. With an ID check or webcam photos on, faces are compared only to confirm the same person answered, and that never touches the score. A transcript can mishear a word, so read it beside the score; you decide who moves on.