AI video interviews or chat interviews: which should candidates get?

Three formats that test different things

Most AI interviewers use one of three formats. In a text interview the candidate types answers in a chat and the AI asks follow-up questions. In a one-way video interview the candidate records spoken answers to set questions, sometimes put to them by an animated avatar, and the recording is scored or summarised. In a voice interview the candidate has a spoken conversation with the AI, in the browser or over the phone.

Vendors tend to argue about accuracy. The more useful comparison is simpler: each format measures something different, and each asks a different amount of the candidate before anyone has told them much about the job.

What each format actually measures

Text measures what the candidate says: their experience, the facts you need such as expected salary, notice period and availability, and how clearly they explain something in writing. It measures nothing about how they speak.

Video measures spoken communication, and it captures a great deal you did not ask for: a face, an accent, a bedroom, a noisy household. Some of that is information you are not supposed to decide on, and once a reviewer has seen it, it is hard to unsee.

Voice sits between the two. You hear fluency, pace and how someone handles a follow-up in real time, without the camera.

What each format asks of the candidate

A text interview needs a phone and a few spare minutes, and it can usually be paused and picked up later. That suits the people who apply for most volume roles: someone between shifts, on a bus, or answering between two other things.

A video interview needs a working camera, somewhere quiet and presentable, decent light, and enough mobile data to upload several recorded answers. It also asks the candidate to be comfortable being recorded by software before they have spoken to a person, and not everyone is.

A voice interview needs a quiet place and a block of uninterrupted time, because a spoken conversation is usually harder to pause halfway through.

Each of those requirements removes some candidates before a single answer is judged. A format fewer people finish is, in practice, a filter nobody designed.

When text is the wrong choice

When speaking is the job. For a call centre agent, a salesperson, a receptionist or a teacher, how someone sounds is part of the work. A text first round tells you they can describe the job, not that they can do it out loud.

When written fluency would filter unfairly. A typed interview rewards people who write well. For a warehouse or kitchen role, that is a bias the format introduces on its own. Keep the questions short and concrete, and read the answers for substance rather than polish.

When you are worried about borrowed answers. A typed answer can be drafted with help from another AI tool, and the transcript shows what was sent, not who wrote it. Questions about the candidate’s own specific experience, with follow-ups on the details, are much harder to outsource than “why do you want this job”.

When video is the wrong choice

When applicants are on their phones and short of time. Retail, food and beverage, logistics and other volume roles draw applicants who will answer a few questions in a chat and will not set up a recording.

When appearance must not matter. If nothing about the job depends on how someone looks, a format that shows you adds risk without adding information.

When the system scores tone or expression. Inferring confidence, enthusiasm or personality from a face or a voice is the hardest part of AI interviewing to defend. If a vendor scores these, ask what evidence each score rests on. There is more on this in is an AI screening interview fair to the candidate?

A rule that works for most teams

Match the first round to the question it has to answer. Most first rounds exist to find out whether a candidate has the basic experience, wants a salary in range, can start in time and can work the hours. None of that needs a camera, and a format more people finish answers it for more of your applicants.

Where spoken skill genuinely matters, test it in the second round: with a person, or with a voice or video step for the shortlist only. You ask more of fewer candidates, at the point where they know they are being seriously considered.

The questions a first round should ask, whatever the format, are in the questions a first round actually has to answer.

Questions to ask a vendor, whatever the format

• Are candidates told they are talking to an AI before the first question?
• Does every score come with the transcript or recording it was based on?
• Does the system reject anyone on its own, or does a person decide?
• Does it score anything the candidate did not say, such as tone, expression or background?
• Can a candidate pause and come back, and what happens if their connection drops?
• Is there a route to a person for a candidate who will not use it?

Where we stand

Blue’s interviews are typed by default, for the reasons above: most first rounds need facts rather than a performance, and a phone-friendly format that can be paused reaches the most applicants. An employer who needs spoken answers can set an opening to Voice or Video instead, and the candidate can still switch to typing. Blue tells candidates it is an AI and asks them to agree before the first question, and every score carries its reasoning and the full transcript. It is the wrong tool for judging how someone speaks, and we would rather say so here than have you find out in the second round.

For text, video and voice tools available in Malaysia side by side, see the 8 best AI interviewer tools in Malaysia.