An AI phone screen interview often replaces the first recruiter call. A voice bot asks about your background and the role’s requirements. Then it sends the hiring manager a transcript, a summary and a rubric score. No person is there to notice what you left out. So answer each requirement question with a clear yes or no first. Then give one concrete piece of evidence, like a product, a number or a year. Many of these tools ask weak follow-up questions, a recruiting consultant told CNBC. A vague answer may never get a second chance. Long pauses can also make the bot move on before you finish thinking. Before the call, find out which requirements the screen checks. The job posting usually lists them. In New York City, an employer using an automated hiring tool must also name them in a notice. This post follows one Android engineer through a full AI screen, with answers you can adapt.
The AI phone screen interview hands the hiring manager a scored report
These screens are now common. Greenhouse surveyed 2,950 active job seekers in five countries for its 2026 Candidate AI Interview Report. It found that 63% had been interviewed by AI. Greenhouse sells voice AI interviews itself, so it has a stake in that number. Hirevue, another vendor, launched a voice-based AI interviewer in June 2026. Its release says the tool runs two-way voice interviews with every applicant, at any hour.
The output is what decides your next round. A September 2026 CNBC Make It article describes what the hiring manager receives. It’s a transcript of the call and an evaluation of the candidate. That evaluation can include a short summary and a rubric score against the job requirements. It can also include a final overall score or ranking. The manager decides who advances from that report.
A human recruiter screen works differently. The recruiter remembers that you sounded sharp and fills in gaps. Our post on how the recruiter screen sets your level covers that version. The AI screen has no memory of how you came across. It has the words in the transcript and a score built from them.
The rubric is the job posting, so read it first
The screen checks the role’s requirements. Hirevue’s Dina Taylor told CNBC that what the tool screens for depends on the company and the job. CodeSignal’s AI Interviewer page is more specific about its own product. Employers map their competencies to a 1 to 5 scale. Every candidate is scored on the same rubric. CodeSignal is selling the tool, so read that as a product description.
You can’t see the rubric. The job posting is its main source, though. Copy the requirements from the job posting into a list. Mark each one as a yes, a partial or a no for you.
For a New York City role, there’s a second source. Local Law 144 covers automated tools used in hiring. The city’s enforcement FAQ, dated June 2023, says what the notice must contain. Employers must tell New York City candidates they use such a tool. The notice must name the qualifications or characteristics the tool will assess. It must also explain how to request a reasonable accommodation. It must arrive 10 business days before the tool is used. It may sit on the employer’s careers page rather than the job posting. So check there, because that list is close to the rubric. Sara’s post on AI-graded behavioral answers covers the bias audit side of the same law.
The running example: one Android posting and its five checks
Take an Android engineer we’ll call Omar. He applies for a mid-level Android role. The posting asks for three or more years of Kotlin. It asks for Jetpack Compose shipped in production. Kotlin Multiplatform is listed as a nice-to-have. The role is hybrid in Austin, three days a week.
Omar’s list looks like this. Kotlin is a yes, with five years. Compose is a yes, with two years in production. Kotlin Multiplatform is a partial, from a side project only. Austin is a yes. He also expects one light technical check. Our own question bank tags short concept questions, like the Activity lifecycle, as phone-screen material. He prepares the ViewModel on rotation as his likely one.
Before the call, Omar writes one sentence of evidence for each item. The screen is short. Those sentences are what the rubric can score.
Vague answers fail because the follow-up may never come
Katrina Kibben is a recruiting consultant. She told CNBC that many AI interview tools “really stink” at follow-up questions. If your first answer is thin, the bot may not ask you to say more. A human would. Kibben called vague answers the most common mistake. Her advice is to open with a very clear yes or no.
Products differ here. CodeSignal says its tool probes thin answers with follow-ups, based on the employer’s criteria. You won’t know which kind you have until the call. So answer as if no follow-up will come. If one does, you get extra room to add detail.
Here’s the weak version of Omar’s Compose answer.
Yeah, I’ve done quite a bit of Compose. We migrated a bunch of screens a while back and it went pretty well.
A transcript of that answer has no yes, no duration and no product. “We” hides what Omar built himself. A rubric line for “Compose in production” has little to score.
Pauses and spoken jargon are the two traps a person would forgive
Nicole Kaiser is a technical talent scout who has tested these tools. She told CNBC a bot may interrupt a long pause. It may also skip to the next question before you finish thinking. A human interviewer waits. So don’t think in silence. Say a holding line out loud. “Let me think about that for a few seconds” works. Then think.
Kaiser also said the tools can struggle with accents and that clear enunciation matters. Android answers are full of terms a transcriber may mangle. Think of Compose, coroutines, Room and Hilt. Say them slowly and in full. Say “Kotlin coroutines” rather than “coroutines” alone. Context words make a garbled term easier to read in the transcript.
The full sample: Omar’s answers to all five checks
Here’s how Omar answers each question. The wording is ours. The bot’s questions will differ. The requirements behind them won’t.
Question: Do you have experience building UI with Jetpack Compose in production?
Yes. I’ve shipped Jetpack Compose in production for two years. At a grocery delivery app, I rebuilt the checkout flow in Compose. About 400,000 people use it each month. I also built the Compose UI tests for that flow.
Question: How many years have you worked with Kotlin?
Five years. All of my Android work since 2021 has been in Kotlin. That includes Kotlin coroutines and Flow for networking and data.
Question: Have you used Kotlin Multiplatform?
Not in production. I built one side project with Kotlin Multiplatform. It shares a Ktor networking layer between Android and iOS. My closest production work is a shared Kotlin module that two of our Android apps use.
Question: What happens to a ViewModel when the user rotates the device?
The ViewModel survives the rotation. The activity is destroyed and recreated. The ViewModel stays in its ViewModelStore. It’s cleared only when the activity finishes for good. Then onCleared runs.
Question: This role is hybrid in Austin, three days a week. Does that work for you?
Yes. I live in Austin. Three office days a week works for me.
The numbers in these answers are illustrative, so use your own.
Why each part of Omar’s answers scores
Every answer starts with the requirement’s verdict. “Yes,” “Five years” and “Not in production” can each match a rubric line on their own.
The second sentence carries the evidence. A product, a user count and a year are hard to misread in a transcript. They also give the manager something to ask about in the next round.
The partial answer is honest. It still scores. “Not in production” is a clear no. The side project and the shared module are the closest real evidence. That’s better than a vague yes that falls apart in the human round.
The technical answer is short on purpose. A screen checks that you know the basics. The deeper version of the same ViewModel question is in our post on Android domain interview follow-ups. Save that depth for the human round.
Salary may come up too. The screen may ask for your expected range. Treat it the way you would with a person. Our post on what to say when a recruiter asks for your current salary has the wording.
Ask two questions before you start the screen
In the Greenhouse survey, 70% of candidates said nobody clearly told them upfront about AI evaluation. So the invite may not explain much. Send the recruiter a short note first.
Hi Maya, thanks for the invite to the AI screen for the Android Engineer role. Two quick questions before I take it. Does a person read the full transcript, or only the score? And if my connection drops, can I restart the session?
The first answer tells you how much the score alone decides. The second protects you from a technical failure counting as a weak answer. If you need an accommodation, ask in the same note. In New York City, the notice should explain how.
Then plan for silence afterward. Of the Greenhouse respondents who completed an AI interview, 51% never heard back. Put a follow-up date in your calendar a week after the screen. If nothing comes by then, write to the recruiter once.