Four Kinds of AI Interview, and What Each One Records

What does a one-way video interview actually capture?

A one-way round shows you a question, gives you a short prep window, and records one answer with nobody on the call. The recording exists either way, and what reads it depends on the platform. HireVue's published statement says its AI relies only on the words in a candidate's answer and does not analyze video, facial expression, background, or tone of voice ¹, so on that platform the score behind the recording comes from a transcript rather than the picture.

The same statement names a choice the consent screen offers: you can opt out of AI scoring, and the vendor says opting out does not remove you from consideration, the response goes to a recruiter who rates it by hand instead ². That is one vendor describing one product, not a rule for every tool. What is a rule, in one jurisdiction, is the EU AI Act's prohibition on AI systems used to infer a person's emotions in the workplace ³, which the European Commission says took effect with the Act's other prohibitions in February 2025 ⁴. It covers exactly the face and voice reading the older advice tells you to perform for. It does not say every video tool complies with it, it does not reach hiring outside the EU, and it does not say a recruiter never watches the recording. Check what applies where you are applying.

Lighting and eye contact stop being the thing worth an hour. Content does not. A model reading a transcript reads the same words whether you deliver them smoothly or not, so a flat answer built around a real, specific example carries further than a polished one built around a generic claim. On HireVue, how many retries you get and how long you get to prepare are set by the employer deploying the assessment rather than fixed by the platform ⁵, so read the instructions screen rather than assume you get one shot.

The employer side asks whether this round is even worth keeping once every candidate's answers are read off a second screen, which is worth reading if you want to know what your recruiter is actually being told about the format you just sat.

What does a chat or text screener ask?

A chat or SMS screener asks a short set of fixed questions before anyone reads your name: work authorization, location, availability, salary range, whether you hold a required license. These are knockout and scheduling questions, and the answer that counts is the literal one. What the invite will not tell you is whether a rule or a model reads what you send back.

There is no performance to give here, and no camera to worry about. Answer the actual question asked, in the specific terms it was asked in, and move on. A vague answer to "are you authorized to work in this location" invites a follow-up message or ends the thread; it does not earn a second look for effort. If a question genuinely does not fit your situation, say so in one line rather than guessing at the answer the form wants.

What you will not get here is a chance to make a case. Save the context and the specifics for the round that can actually weigh them, and treat this stage as paperwork rather than a test of how you communicate.

Treat a live AI interview like a real conversation.

A newer format replaces the recorded monologue with something closer to a call: an AI conducts a live phone or video conversation, asks a follow-up based on what you just said, and lets you ask questions back. One vendor's own candidate page describes the AI summarizing an answer, then asking a follow-up drawn from it, before sending the exchange to a human recruiter who decides whether to reach out ⁶.

Read that description as a vendor's own claim about its own product, not an independent finding. The shape of the format is still worth preparing for on its own terms, rather than treating it as a longer version of the recorded round. One study has asked people who had actually sat an AI-evaluated interview what they made of it: in a convenience sample of 225 respondents in China and 213 in Pakistan, AI-based interviews were negatively associated with perceived transparency and with interactive communication, and higher trust in AI weakened that response ⁷. Two countries, self-reported and correlational, so it does not transfer to applicants elsewhere. The gap it names is visibility, and visibility is something you can ask about directly. Ask, before or during the round, who reads the transcript and whether a person makes the actual call, and answer as if a follow-up is coming, because the format described above is built around them.

Who reads it when a notetaker sits in on a human interview?

Whoever already reads the hiring file: your interviewer, the recruiter, and anyone else on the loop, because what the notetaker produces lands in that file next to your name. The format is an ordinary interview with a person, except that a notetaker is transcribing and summarizing in the background. The conversation itself is unchanged, questions and follow-ups from a live person exactly as they would run without software in the room, so there is nothing about it to prepare differently.

What is new is the document produced after it: a transcript, plus an AI-written summary of that transcript. The summary is the part worth caring about, because it is an interpretation rather than a record. The employer side of this exact question asks whether a line like "seemed nervous and rehearsed" belongs in a hiring file at all, and the answer there is no, on the grounds that the transcript underneath is what a decision should be able to rest on.

You are allowed to ask, before the interview starts, whether a notetaker is running and who besides your interviewer will read what it produces. It is a plain question, and asking it before the call costs nothing. If something you said gets paraphrased back to you inaccurately later in the process, that is a fair moment to correct the record rather than let a summary stand in for the conversation.

See how it works >