What an AI interviewer does, and what it must not decide
An AI interviewer is software that runs the interview itself. It asks a question, listens to the answer, asks the follow-up that answer earns, and rates the conversation against criteria fixed before anyone was invited. It hands back a recording, a transcript and a scored report. A named person still decides.
3 different products are sold as an AI interviewer
| Recorded answers | Scripted voice bot | Live adaptive interview | |
|---|---|---|---|
| How it runs | The candidate records answers to set prompts, alone, on their own clock | A voice agent reads a branching script down a phone line | A two-way conversation in the browser, in real time |
| Can it follow up | No. The next prompt was written before the candidate spoke | Only where a branch was authored in advance | Yes. The next question is built from the answer just given |
| What it can observe | A rehearsed take, often the third one | Voice only, so nothing the candidate shows or writes | How someone reasons when pushed on their own claim |
| Where it fits | Very high volume, low-stakes qualification | Shift and hourly roles, phone-first candidate pools | Screening rounds where the depth of an answer decides the shortlist |
| What it cannot do | Tell a thin answer from a vague one | See a diagram, a screen or a written example | Replace the judgement call at the end of the round |
Inside a live AI interview
Two tiles, a running timer and a real exchange. Interview length is set per role. Example data.
The candidate joins in a browser after a device check. The rubric for the role is already fixed, so what varies is the questioning: the interviewer hears the answer and asks the thing a good human would ask next. Nothing is scored on screen during the interview, and there is no live transcript in the room.
What an AI interviewer is
AI interviewer Software that conducts the interview rather than assisting one. It asks, listens, probes the answer it just heard, and rates the conversation against criteria written before the first candidate was invited.
A recruiter writes the criteria for a role once. Candidates interview when it suits them. A scored report comes back per person, with the recording and transcript attached, and a person reads it before anyone moves.
Live interview AI, automated interviews and conversational AI recruiting
Teams searching for interview AI usually mean one of three things, and the three are not interchangeable. Live interview AI means a conversation happening now, where the software can change its next question. Automated interviews is the wider term: it covers a recorded screen with no conversation in it at all, and it says nothing about whether anyone listened.
Conversational AI recruiting is the one most often misread. In practice it is chat and scheduling: a bot on a careers page answering questions about shifts, qualifying an applicant with a few set questions, and booking a slot. That is upstream of an interview, not an interview. If a vendor answers a question about AI interviewing with a screenshot of a chat widget, those are different products.
The criteria are fixed before anyone interviews
Named parameters, each with a description and a weight. The weights have to total exactly 100. Example data.
This is the part that makes the round comparable. Each criterion is named, described and weighted for the role, and the template will not save until the weights reach 100. Questions adapt to the person in front of them. The rubric they are marked against does not move between candidates.
How an AI interview screening round actually runs
- Write the criteria Name what the role is actually judged on, describe what good looks like for each, and set the weights. Do this before a single invite goes out. Problem solving 30, system design 25, ownership 25, communication 20.
- Invite the shortlist Candidates sit in a stage on the role. Bulk invite moves a whole group from Ready to Invite to Invite Sent in one action.
- The candidate picks a slot Each candidate books inside a window you set, and can reschedule or cancel. Nobody waits on a recruiter's calendar to start.
- The interview happens A live two-way conversation in the browser, timed to the length set for the role, with follow-ups driven by what the candidate says. The interviewer hears a claim about a migration and asks what broke.
- The report lands A recording, a transcript with speaker labels and timestamps, and a scorecard. The candidate's status moves to Completed, then Pending Review.
- A person decides Approve, Reject or Hold, or move the candidate to the next stage. The decision and who made it are written into the stage history.
What lands on your desk afterwards
Each criterion out of 5, an overall out of 100, and the resume claims the interview actually tested. Example data.
Most pages in this category assert that the scoring is good. Few show the output. A resume is a set of claims, and the "Verified in this interview" section marks each one verified, refuted or unclear with the evidence beside it, so a disagreement is about a specific sentence rather than an impression.
What a follow-up is for, in 4 real moments
"We cut the failure rate on checkout by rewriting the retry logic."
Whether this candidate made the decision or stood near it. The follow-up asks which part was theirs and what they argued against.
More "we", a restatement of the result, and no decision they can name.
"The project slipped, but we learned a lot about scoping."
Whether there is a real failure behind the phrasing. The follow-up asks what they personally got wrong and what they do differently now.
A lesson with no mistake attached to it, or a failure that is secretly a success.
"Led the migration of our billing service."
Whether the claim on the resume survives one specific question. The follow-up asks for the thing that broke and how they found it.
Correct architecture vocabulary with nothing specific underneath it.
"I would start by gathering requirements from stakeholders."
Whether there is a method under the script. The follow-up asks which requirement they would drop first, and who would be unhappy.
The same generic sequence again, one level deeper, still with no trade-off in it.
What an AI interview agent must not decide
New York City's rule for automated employment decision tools gives the clearest definition anyone has written of the line you are trying to hold. A tool substantially assists or replaces discretionary decision making when you rely solely on its output, when you weight that output more heavily than any other criterion, or when you use it to overrule conclusions drawn from other factors.
Read the other way, that is a design brief. The score is one input among several, it is not the heaviest one, and it never overturns what a person concluded from the rest of the evidence. Colorado's law approaches the same problem from the other side, treating a system as a substantial factor when its output is capable of altering the outcome of a decision about employment.
So the practical test for AI interview agents is not how autonomous they are. It is whether a named human made the call on the record, with the evidence in front of them, and whether you could show that six months later. On The Cognitive, an advance, a rejection or a hold is a human action written into the candidate's stage history.
What to ask AI interviewers before you buy one
- Is the interview live and two-way, or a recording booth with prompts
- Show me a follow-up question the system wrote from an answer
- What exactly does it score, and is every criterion job-related
- Can I see the evidence behind one rating, in the candidate's words
- Who ran your most recent bias audit, and were they independent of you
- Do you write back to my ATS, or only import from it
- What happens to recordings, and can a candidate have theirs deleted
- What is the alternative route for someone who cannot use it
- Which languages does the interviewer actually speak today
AI interviewing and the rules that apply to you
| Where | What the rule requires | Where to check |
|---|---|---|
| Illinois | Notice, an explanation of how the AI works, and consent before the interview. Deletion within 30 days on request. Drafted around recorded, applicant-submitted videos. | Public Act 101-0260, the Artificial Intelligence Video Interview Act |
| New York City | A bias audit by an independent auditor within the past year, results published before use and kept up 6 months, plus 10 business days' notice. | Local Law 144, and the DCWP final rule at 6 RCNY 5-300 |
| Colorado | Employment is a consequential decision. A system that is a substantial factor in one is high-risk, with a duty of reasonable care against algorithmic discrimination. | SB24-205, with the effective date moved to 30 June 2026 by SB25B-004 |
| European Union | Systems used to recruit, select, filter applications or evaluate candidates are listed as high-risk, and profiling never escapes that classification. | EU AI Act Annex III point 4(a), and Article 6 |
Running AI interviews without making the hiring worse
Do
- Tell candidates before the interview that AI will evaluate it, and explain plainly how it works
- Keep every criterion tied to something the job actually requires
- Have a named person approve each advance, with the evidence in front of them
- Offer a route for anyone who asks for an alternative or an accommodation
- Keep the recording and the transcript, and honour deletion requests
Do not
- Scoring presence, accent, or how professional somebody looks on camera
- Letting a score auto-reject anyone with nobody reading the evidence
- Relying on a bias audit that is more than a year old
- Treating a vendor's compliance page as though it covered your obligations
How to integrate an AI interview copilot with an existing ATS
Ask the question at the level of mechanism, because that is where products differ. There are three moving parts: what pulls candidates in, what triggers the interview, and what goes back. Plenty of tools do the first two well and then hand you a PDF, which means somebody is retyping outcomes into the system of record every week.
The Cognitive imports candidates from your ATS and writes results back: advancing a stage, pushing the score, rejecting, attaching the report and adding notes. Greenhouse, Lever and Ashby are named in the product, and the company states more than 40 ATS integrations. When you shortlist AI interviewers that integrate with existing ATS solutions, ask which fields get written and which stage the write lands on.
One pipeline, from invite to decision
Candidates grouped by stage, with the real statuses, bulk invite and a push back to the ATS. Example data.
Interviewing at volume falls apart in the admin, not the interview. Candidates sit in stages such as Technical Interview 1 (AI) and Hiring Manager Interview, and move through Ready to Invite, Invite Sent, Scheduled, Completed, Pending Review and Approved. Stage history keeps the record of who decided what.
4 things worth remembering
- Live beats recorded for one reason: it can ask the question your prompt list did not anticipate
- The rubric is fixed per role. The questions are live and adaptive. Both halves matter
- Evidence you can read beats a number you cannot open
- A named person decides, on the record, or the tool has quietly become the decision
See it run
Start free at https://app.thecognitive.io/signup. Set the criteria for one open role, invite a shortlist you already sourced, and read the first scorecard. To sit on the candidate side first, take a live interview yourself at /try-interview.
Where to go next
Why live interviews go deeper · AI interviewer by role · AI interview questions generator · Take a live AI interview · AI hiring tools, compared · What is an AI interviewer
Frequently asked questions
What is an AI interviewer?
An AI interviewer is software that runs the interview itself rather than helping a human run one. It asks a question, listens, asks the follow-up that answer earns, and rates the conversation against criteria set before anyone was invited. It returns a recording, a transcript and a scored report. The hiring decision stays with a person.
Is live interview AI different from a recorded video interview?
Yes, and the difference is the follow-up. In a recorded screen the prompts were written before the candidate spoke, so a vague answer and a thin answer get the same next question, which is none. A live interview can hear a claim and ask what broke. If you want an AI interview simulation from the candidate's seat, /try-interview runs one.
Can an AI interviewer replace a phone screen?
It can replace the job the phone screen was doing, but the channel changes what is observable. AI phone screening is voice only, so it cannot see anything the candidate shows or writes, while video can. The Cognitive's interview runs as live video in the browser and there is no phone channel, so if phone-first candidate reach matters to you, ask vendors to name the channel they ship today.
How do AI interviewers integrate with an existing ATS?
Three mechanisms matter: importing candidates, triggering the interview, and writing results back. Many tools import and then hand you a report to retype. The Cognitive imports from your ATS and writes back, advancing a stage, pushing the score, rejecting, attaching the report and adding notes. Greenhouse, Lever and Ashby are named in the product, and the company states more than 40 ATS integrations.
Does an AI interviewer make the hiring decision?
It should not, and New York City's rule explains why in useful detail: a tool substantially assists or replaces discretionary decision making when you rely solely on its output, weight it above every other criterion, or use it to overrule other evidence. Keep the score as one input. On The Cognitive, Approve, Reject and Hold are human actions recorded in the candidate's stage history.
What does an AI interview agent actually do on its own?
In practice: send the invites, let candidates book a slot, run the conversation, write the scorecard and push results to your ATS. That is the part people mean by autonomous AI interview software for recruitment. What it does not do is decide. Ask any vendor selling agent language which of those five steps still needs a person, and which one ends in a decision.
Do I have to tell candidates an AI is interviewing them?
Where the rules bite, yes, and the duty falls on you as the employer. Illinois requires notice before the interview, an explanation of how the AI works, and the applicant's consent, and bars evaluating anyone who has not consented, for roles based there. New York City requires 10 business days' notice before an automated employment decision tool is used. Telling candidates plainly is the safer default everywhere. Check the rules where the role sits.
Can a candidate ask for their interview video to be deleted?
Under the Illinois Artificial Intelligence Video Interview Act, an applicant can request deletion and the employer has 30 days to delete the interview, including electronically generated backup copies, and to instruct anyone who received a copy to do the same. That Act is written around recorded, applicant-submitted videos, so how it reaches a live interview is a question for your counsel. Either way, ask any vendor how deletion works before you run a round.
What is conversational AI recruiting, and is it the same thing?
No. Conversational AI recruiting usually means chat: a bot that answers candidate questions, asks a few qualifying questions and books a time. Conversational AI recruiting software is generally sold for that top-of-funnel job. An AI interviewer runs the interview round itself and produces a scored record. The two sit at different points in the process and buying one does not give you the other.
Can an AI interviewer run in another language?
On The Cognitive, the interviewer speaks English, German, Russian, Dutch, French, Spanish, Italian, Portuguese and Polish. Language is set on the role. If you hire across markets, check the list per vendor rather than the headline count, and check that the scoring works in the same languages as the conversation.
Explore: Deep AI interviews · AI interviewer by role · Take a live AI interview · Interview question generator · AI screening software · What is an AI interview