Which AI Interviews Work on a Phone, and Which Don't

Which Formats Already Expect a Phone?

Chat interviews and phone-call interviews already expect a phone on at least some platforms. Sapia's candidate explainer describes its Chat Interview as a text-based interview completed by typing on a phone, tablet or computer: five to seven mostly behavioural questions, answered in 50 to 150 words each, with no timer and an emailed link that saves your progress ¹.

All of that comes from Sapia's own explainer, with no audit behind it.

Sapia also answers the camera question directly: no video recording and no camera, with the candidate assessed on the written answers ¹.

The live AI interviewer is the other format that already accepts a phone. Alex's About page describes an always-on first-round interviewer that talks to each candidate by phone call, video or text, at a time the candidate picks, and claims a cumulative million candidates interviewed and 30 languages supported ². The counter is the vendor's own and unaudited.

The page also says nothing about how the answers are scored or who reads the transcript afterwards.

What actually happens in an AI interview sorts the formats out, and the format is what decides the device.

Check the Platform's Own Device Page First

The platform's name is in your invitation, and its own help pages are where the device answer lives. A general article cannot know whether you were sent a browser recording, a typed chat or a proctored coding session, and those three do not have the same answer. Open the platform's help link before the deadline rather than after it.

Three questions about the round matter more than the device name:

• Does it record or share your screen at any point
• Does it ask you to type at length, or to write code
• Does it check your identity against a photo ID on camera

Each of those is a section a phone handles badly. CodeSignal's own page says its proctoring records video, audio and screen activity for the entire assessment session ³. The same page says identity verification prompts the candidate to present a valid government-issued photo ID, which is matched to the person on camera, and that the verification status goes to the hiring team with the results ³. Those are capabilities an employer switches on, and the page does not say which are defaults.

A ten-minute typed answer on a phone keyboard is a different exercise from the same answer on a laptop.

Whether you have to agree to screen or camera recording is a separate question from which device you use, and worth reading before you agree to either.

If the instructions do not say, ask the recruiter. Whether the round is recorded in the browser, and whether any part of it needs a keyboard or a screen share, are questions with factual answers. They are logistics questions, and a recruiter can answer them.

What a Phone Costs You Mid-Answer

The expensive failure is not the device. It is the connection dropping while you are answering. HireVue's help article says a candidate disconnected mid-answer should assume that answer has most likely been submitted: re-entering the interview takes you to the next question, and there is no way to reset the interrupted one ⁴.

The vendor's advice is to finish the interview and then contact the company you applied to, which can send a retake if it decides one is needed. HireVue says it does not have permission to provide another interview itself ⁴. Nothing in that obliges an employer to grant a retake.

A stable connection beats a fast one.

Closing the browser is not a reset button either. Willo's knowledge base says its timer counts up until the answer is submitted, and that the time recorded includes any retakes ⁵. It adds that candidates invited by email, SMS or an ATS integration cannot reset the clock by restarting the assessment, closing the browser or refreshing the page ⁵. That describes Willo's own mechanism. On other platforms the recording can cut off automatically when the allotted time ends.

There is a softer edge to this. HireVue's 2025 explainability statement says an answer is only scored by AI when the candidate has consented and the system finds enough content in the response ⁶. It names a too-short answer, one not given in the time allowed, speech it cannot understand, and a microphone problem as the ways that fails, and says those responses go to human review instead ⁶. The vendor calls it a small proportion of candidates and publishes no figure.

A response routed to a person has changed queues, and the statement says nothing about how that review turns out. Whether a human ever watches your AI interview covers what that review means, and if a technical failure cost you an answer, the recruiter is the person to tell.

Set the Phone Up Before You Start

Prop the phone at eye level on something solid, put it in do-not-disturb, and plug it in. A handheld phone shakes, an incoming call can cut into a recording the way the last section described, and a recording round drains a battery faster than you expect. Wired headphones with a microphone take one variable out of it.

Use the practice round for the parts you control. HireVue's help center says its OnDemand practice questions are generic, can be redone without limit, and are never uploaded or submitted to anyone, and that once you move past the playback they are gone ⁷. That is where to hear your own audio, look at what is behind you, and find out whether the microphone picks up the street.

By default the real questions cannot be previewed, and the real answers cannot be watched back ⁷.

Give the preparation an hour, most of it on specific examples rather than lighting. How to prepare for a one-way video interview lays that hour out.

If you have no suitable device at all, say so early. A recruiter with a week can move a deadline or offer another format, and one told an hour beforehand has fewer options. Asking for time costs less than sending a take you already know is bad.

See how it works >