Screening · Picked.ai

Product · Screening

Every applicant has a fifteen-minute conversation with the AI.

By voice or chat. In their language. On their device. We ask the questions a competent recruiter would ask. We do not ask trick questions. We do not score tone.

Post a role →See a sample transcript

Inside the screen

What we ask, and why.

Motivation. Why this role. Why now. What they walked away from to apply.
Must-haves. Specific skills and constraints you flagged at role creation. Hard yes or hard no.
Deal-breakers. Right to work. Notice period. Salary expectation. Working pattern. Location.
The role from their side. What they have done that maps. What they have not. What they want to learn.
Their questions for you. A good candidate has a good question. We capture them and pass them through.
Language and accommodation. The conversation runs in the candidate's preferred language. Extra time, dyslexia mode, or screen reader on request.

What this is not

Four things this is not.

01

Not a chatbot quiz.

We do not feed scripted questions in order. The follow-up is adaptive to the last answer.

02

Not a tone scorer.

We do not score on accent, fluency, or vocabulary. We score on what the candidate said, not how they said it.

03

Not a knockout.

We do not auto-reject on a single answer. The screen produces a numeric signal that feeds the next stage; reject decisions involve more than this one read.

04

Not a recording.

This is a live conversation. Nothing gets handed to a human to watch back later for grading.

Sample · sales role · 8 minutes in

Lightly edited. Shipping the live transcript view with V1.

Sample. Shipping with V1.

AI       So you closed three of the last four. The one that
         walked, what was that about?

CANDIDATE It was a fit thing. They came in on a price-led
         brief and we are not the cheapest option.
         I tried to reframe it as a value conversation twice.
         It did not land.

AI       What would you do differently next time?

CANDIDATE Earlier disqualification. The price signal was there
         in week one. I kept it warm because the pipeline was
         thin. I should have killed it and gone back to top of
         funnel.

AI       The role you are applying for now closes inbound mostly,
         not outbound. Does that change anything for you?

CANDIDATE It does. I am better at outbound than inbound, so I am
         coming in to learn that motion. I have done it briefly
         at [redacted] and the conversion curve was tighter than
         I expected. Worth doing more.

Eight years of recruiter conversations encoded into a prompt. We are not making this up; the question library has 4,800-plus items.

For every applicant

Built so the candidate is not the variable.

The screen runs in any language the candidate prefers. At V1, we support every language Claude is competent in, which is 25-plus in practice. Where we are not competent, we say so and offer English with extra time.

Accessibility is not a setting we add later. Every candidate can request: extra time, a screen-reader-friendly layout, a dyslexia mode that switches the font and contrast, a slower voice, or a written-only version. The request is one tap. Nothing is logged as a flag against them.

Voice: candidate's device, on browser. No app to install.
Chat: identical question set, identical scoring.
Pause and resume up to 72 hours.
Honest time estimate before the candidate commits.

Under the hood

~$0.10

Average cost per screen at V1, blended.

30%

Median pass-through rate to assessment.

14 min

Median screen length across roles.

Claude Sonnet

Model. Reasoning runs on Anthropic. Speech-to-text via Whisper. EU residency.

We publish per-stage cost and accuracy in the model cards. See /trust/model-cards once it's live.

For the candidate

We treat applying like a real conversation. Not a form.