The Zara AI interview: what it is and what it actually scores

Why this matters

Nearly every role on micro1's board routes through the same gate: an AI interview with Zara. It is unpaid, it happens before you know your rate or whether a project has a seat for you, and its length depends on what the role assesses: micro1's free practice interviews are labeled 18 minutes, and its knowledge base builds real interviews from skill modules of about 5 minutes each plus any exercise the role adds.

Most of what is written about it online is guesswork. But micro1 published a peer-reviewed-style research paper describing Zara in detail (Zara: An LLM-based Candidate Interview Feedback System, on arXiv), and that paper, plus micro1's own expert-facing pages and knowledge base, answers a lot of what candidates are guessing at. This guide is built on those, and it flags clearly where we are relying on anything less solid. For the questions themselves, see our companion guide, micro1 interview questions.

It is not only micro1

Alignerr's own onboarding page states that some jobs may require additional steps "such as a Zara™ interview or domain assessments".

micro1 built Zara and publishes research on it. Alignerr lists it as a step in its own process. We have not seen either company describe the commercial arrangement between them, so we are not going to characterise it, but in practice, preparing for Zara helps on more than one platform.

What Zara actually is

From micro1's paper: Zara is "an AI-driven recruiter primarily powered by OpenAI's GPT-4o, supplemented by additional LLMs, fine-tuned models, and a Retrieval-Augmented Generation (RAG) framework."

It operates in four phases, and candidates typically only notice the second one:

1. Personalised interview preparation: practice interviews generated for your role and skill set
2. The AI-led interview: the part everyone means when they say "the Zara interview"
3. Structured post-interview feedback: issued on request
4. Automated query resolution: Zara answering your questions about the process

Phases 1 and 3 are free, useful, and largely ignored.

Use the practice interview

micro1's paper describes phase one plainly: candidates "first engage with Zara through customized practice interviews tailored to their targeted roles and skill sets", and these "help candidates familiarize themselves with the anticipated interview questions."

A practice run against the same system that will assess you, generated for the specific role you are applying to, is an unusually direct rehearsal. If you do one thing from this page, do that. And do it for the role you are actually applying to, not a generic version, because the generation is role-specific.

The interview adapts, and you are allowed to interrupt it

Two things from the paper change how you should behave in the room.

It adapts. Zara "dynamically adjusts questions based on candidates' responses". Your answers steer what comes next, so a thin early answer narrows the ground you get to cover.

You can ask it questions. The paper states Zara answers "any clarification questions the candidate has mid-interview", and micro1 credits exactly this for an improvement in its internal question-quality scores, from 8.38 to 8.60, because it enables "more precise and effective questioning".

That is the opposite of the usual advice about automated interviews. Asking Zara to clarify an ambiguous question is a designed-for behaviour, not a tell. Candidates who treat it as a one-way recording and answer a question they have misread are giving up something the system was built to offer.

Answer with substance, and steer towards the job ad

Because the interview adapts to you, you are partly steering it. Steering it towards the role's stated scope costs nothing.

Read the listing's requirements immediately before you sit down and keep them in front of you. Where a listing names something specific, that specificity is deliberate: micro1's AMO physics role names two ytterbium isotopes, and its STEM Expert listing names the exact failure modes you will be asked to catch. Those are the terms the assessment is likely to move towards.

Then put most of your effort into the reasoning. Name the constraint you are working within. Say what you would check. State the trade-off you rejected and why. On a role where the job itself is written explanation, which is most of the high-rate work on this board, the interview is a sample of the product.

What the feedback covers, and what it deliberately leaves out

If you request a report, micro1's paper says Zara "identifies two to three specific strengths and two to three improvement areas, explicitly excluding soft skills and communication to reduce subjectivity."

That sentence describes how the feedback is written, not how the assessment is scored. micro1's knowledge base, written for employers who use Zara, states that soft skills are "automatically assessed in every interview", so do not assume delivery is ignored just because the report leaves it out. micro1 does not publish its scoring rubric or any pass threshold.

The paper also notes the feedback passes a "reflective review" step with guardrails that "systematically prevent overly negative language", and its own system prompt asks for feedback that is "specific, constructive, supportive, sugar-coated". So expect a gentle document. Read it for the improvement areas, which are the actionable half.

The statistic worth acting on

Over a three-day window, micro1 recorded 4,820 unsuccessful interviews. About 10.7 per cent of those candidates requested detailed feedback.

Nearly nine in ten people who did not get through walked away without asking for the report that would tell them why, on a platform where you can apply to other roles, and where the same interviewer assesses all of them. A rejection with a written account of what was weak is a direct input into the next attempt. If you are rejected, ask.

Separately, micro1 reports its automated query system resolves around 75 per cent of candidate inquiries without a human, and that Zara-led interviews scored an average 4.37 out of 5 on Net Promoter Score across 400 candidate ratings.

A human does look at it

micro1's paper: "Employers use Zara to efficiently identify promising candidates, whose AI-generated assessments are subsequently reviewed by human recruiters."

Its experts page adds that "human recruiters review profiles and match experts with relevant opportunities based on verified skills and project requirements", and that "certification decisions are based on demonstrated knowledge rather than background, location, or personal characteristics."

So the AI is a filter with a human behind it, not a sole decision-maker. It also means a strong interview is not a placement: micro1 states elsewhere that passing certification does not guarantee you a project.

Proctoring: the Ava model is real

Several third-party guides describe a second AI, named Ava, that proctors the session. When we first wrote this guide we could not confirm that from any micro1 source. We can now.

micro1's knowledge base page "What is Proctoring?" says proctoring can be switched on per interview. When it is, candidates must share their entire screen with Zara, use a webcam throughout and limit themselves to one screen; Zara "monitors and logs the candidate's environment and actions", and afterwards the data "is analyzed using the Ava Model", which micro1 calls its most accurate proctoring model, "with a 93% success rate in detecting cheating." Each proctored interview gets a proctoring score and a breakdown of flagged behaviours. Breaking the rules (adding a second monitor, stopping screen sharing, turning off the webcam) suspends the interview until you put things back.

Two caveats. That page is written for employers configuring Zara, and micro1 does not say which of its own certification interviews have proctoring switched on. And the specific numbers some third-party guides attach to Ava (gaze-drift limits, automatic rejection below a set score) still do not appear in anything micro1 publishes, so we are not repeating them. The practical advice is unchanged: sit the interview honestly, in one window, on one screen.

Not every AI assessment is Zara

Outlier runs a different shape. It puts onboarding at 30 to 90 minutes by its own estimate, including skill screenings, and then adds project-specific onboarding where you review the guidelines and complete assessment tasks before any paid work begins.

That changes the preparation. Where Zara is an interview about your expertise, an Outlier project assessment is a sample of the work itself, graded against the guidelines you were just given. Read the guidelines properly before starting the tasks; they are the rubric.

Alignerr sits between the two: sign up, apply, identity verification, billing signup and onboarding, with some jobs adding a Zara interview or domain-specific assessments. Mercor runs its own AI interview with a different retake rule; see the Mercor AI interview.

What nobody publishes

These gaps are the questions people actually have:

• There is no published pass mark. micro1 does not state a score threshold, and neither does the paper
• There is no published rubric. We know feedback excludes soft skills; we do not know what the assessment weights
• There is no published retry policy. Whether, and how soon, you can re-interview for another role is not stated anywhere we can find
• There is no published acceptance rate. The 4,820 figure is a count of unsuccessful interviews in one window, not a pass rate
• There is no single published length. micro1 has removed the 20 to 40 minute figure its experts FAQ used to give; its practice interviews are labeled 18 minutes and its knowledge base describes durations block by block

If you have been through it, the gaps above are where first-hand accounts are worth most.

Before you sit it

• Do the practice interview for the specific role, not a generic one
• Have the job ad open. Zara adapts to you, and you should be framing answers against the role's stated scope
• Ask for clarification when a question is ambiguous. The system is built for it
• Answer with substance over polish, but do not assume delivery is ignored: micro1's knowledge base says soft skills are assessed in every interview
• Set up as if proctoring is on: one screen, webcam working, nothing else open
• Budget the time as unpaid. It comes before you know your rate, and micro1 says certification is not placement
• If it goes badly, request the feedback report. Almost nobody does

Sources

Everything above is drawn from the platforms' own published material, and attributed in the sentence that uses it:

• Zara: An LLM-based Candidate Interview Feedback System, arXiv:2507.02869. micro1's own paper, the source for Zara's architecture, its four phases, the feedback method, and every figure quoted here except the proctoring rate
• micro1's experts page, for the hiring process, the fairness statement, and country eligibility
• micro1's interview prep page, for the 18-minute practice interviews
• micro1's knowledge base (kb.micro1.ai), "What is Proctoring?" and "Soft Skills", for the Ava model, the 93 per cent figure, the proctoring rules and soft-skill assessment
• Alignerr's process page, for the Zara interview and domain assessments in its onboarding steps
• Outlier's own onboarding estimate and project assessment description
• The micro1 listings live on this site

First published 7 September 2026; proctoring and interview length updated 25 September 2026; merged with our former "Five things that decide an AI interview" guide on 3 October 2026. The Ava model is confirmed by micro1's knowledge base; the specific thresholds third-party guides attach to it are not. We are not affiliated with micro1, Alignerr or Outlier.