What a micro1 Zara Interview Records and Who Sees It

A micro1 AI interview is conducted by Zara, micro1's AI recruiter, and it captures more than the conversation. According to micro1's Candidate Privacy Notice, last updated 21 July 2026, the company collects "Audio and video recordings of your AI interview sessions," "Screen sharing recordings for technical assessments," "Transcripts of your responses," and "Assessment scores and evaluation results," alongside "Behavioral and performance analytics (e.g., information collected from our proctoring tools to assess the integrity of your interviews)" (micro1 Candidate Privacy Notice).

None of that appears on the pages micro1 shows candidates before the interview. This article reads the two sets of documents against each other: what the marketing pages describe, what the privacy notice and micro1's own published research document, and what neither one answers. Facts are current as of 11 September 2026. It is a reading of first-party documents, not legal advice.

The gap between the invitation and the notice

micro1's candidate-facing interview pages describe the experience in experiential terms. The AI interview guide promises "A conversational interview with our AI recruiter, Zara," "Open-ended technical questions based on your chosen role," and "A real-time result indicating whether you meet micro1's global certification criteria" (micro1 AI interview guide). The companion questions page adds that "No two candidates receive the same questions" (micro1 interview questions).

Neither page states that the session is recorded, that a screen share is captured during technical assessments, or that a proctoring tool runs alongside the interview. Those facts are in the Candidate Privacy Notice, which sits in a separate legal center and is not linked from the interview guide. micro1's general website privacy policy explicitly routes you there: "This Privacy Policy does not apply to candidates engaging with our AI interviewer (Zara)" (micro1 website privacy policy).

This pattern is not unique to micro1. The same split between an experiential candidate page and a disclosure-bearing legal document shows up across AI-led screening, and it is the reason what a Mercor AI interview records and who sees it needed the same treatment. The practical instruction is the same in both cases: the privacy notice is the document that tells you what happens.

What the notice says is collected

Two entries deserve attention. The screen-sharing line means that for technical roles the capture is a recording of your screen, not a snapshot of a code editor — a wider surface than most candidates assume when a browser prompts them to share. The identity line is conditional rather than universal; micro1 says it collects government ID "if necessary," which leaves the trigger undefined. If you reach that step, what interview identity verification actually checks covers what these checks compare and how long the images typically persist.

The notice sets no numeric retention period. It states only that micro1 keeps information "for no longer than necessary for the purposes for which it is processed," and for candidate profiles, "as long as necessary to fulfill the purposes of our Services." That is a purpose-bound standard, not a date.

The proctoring score is a separate number, and a human sees it

The most consequential detail is not in the privacy notice but in micro1's own research. In "Better Together: Quantifying the Benefits of AI-Assisted Recruitment," published 29 July 2025, micro1 describes a trial in which roughly 37,000 applicants were randomly assigned to a conventional résumé screen plus human interview, or to an AI-led structured video interview. In the AI arm, the system examined technical skills "plus soft-skills and proctoring scores," generating a skill report that human recruiters then reviewed to make advancement decisions (micro1 research).

That places the proctoring score inside the artifact a recruiter reads, next to your technical evaluation. It is not an internal backstop consulted only when something looks wrong. A low integrity signal and a strong technical answer arrive on the same page.

micro1 reports that candidates who cleared the AI interview passed the final human interview 54 percent of the time against 34 percent for the control group, and that nearly 40 percent of AI-selected finalists were employed within five months versus 23 percent of controls. These are vendor-reported results from micro1's own study of its own product, and the company has an obvious interest in the finding; treat the direction as informative and the magnitude as unverified.

What micro1 does not publish anywhere is the composition of the proctoring score. There is no first-party documentation of whether it weighs gaze direction, face presence, second-screen detection, tab focus, paste events, or response timing. By contrast, several assessment vendors publish per-signal documentation; micro1 does not.

Who receives your profile

The Candidate Privacy Notice states that micro1 "may use automated technologies, including artificial intelligence to: Engage in initial screening or matching," to "proctor interview sessions," and to "evaluate the overall strength of your application." It then adds a safeguard: "The overall assessment will be reviewed by a human evaluator before a final decision is made."

Disclosure is scoped to a particular kind of role. The notice says that "For certain roles where we are seeking candidates to complete tasks on behalf of our clients (e.g., if you are applying to be a data labeler/creator of content we provide to our clients), your candidate profile may be disclosed to such clients to assist us in selecting candidates with relevant skills."

That clause reads differently once you know what micro1 sells in 2026. Its homepage no longer presents the company as a developer-staffing marketplace; it describes "Data lab to train frontier models & evaluate agents" and lists three products — Realm for reinforcement-learning environments, Cortex for agent evaluation, and a robotics data line (micro1). The audience is AI labs, and the recruiting funnel feeds expert human data work. If you are interviewing with Zara, the pool you are being sorted into is largely expert-data and evaluation work for micro1's clients, and the clause above is the mechanism by which those clients see you.

The notice also names foundational model providers among its vendors, describing them as "large language model companies that process your information solely on our behalf...but never to train their models."

What Zara is, in micro1's own description

Zara is documented in a 2025 arXiv preprint co-authored by micro1 staff with affiliations at the University of Southern California and Stanford University, titled "Zara: An LLM-based Candidate Interview Feedback System" (arXiv:2507.02869). The paper describes four phases: "(1) personalized interview preparation, (2) the AI-led interview, (3) structured post-interview feedback, and (4) automated candidate query resolution." The system is "powered by OpenAI's GPT-4o, supplemented by additional LLMs, fine-tuned models, and a Retrieval-Augmented Generation (RAG) framework."

Two details from that paper are worth carrying into an interview. First, the feedback component deliberately narrows its scope: Zara "identifies two to three specific strengths and two to three improvement areas, explicitly excluding soft skills and communication to reduce subjectivity." Second, the scale reported is blunt about outcomes — over a three-day evaluation window the paper counts "4820 'unsuccessful' interviews conducted, of which 10.7% requested detailed feedback." Failing a Zara interview is a common event, and detailed feedback is available on request rather than by default.

The paper's focus is transcripts. It does not describe the proctoring layer, which leaves the privacy notice as the only first-party evidence that one exists.

What is not documented

Be skeptical of any guide that fills these gaps confidently. As of 11 September 2026, micro1's own materials do not state:

• how long the interview runs;
• whether a failed interview can be retaken, and after what interval;
• what behavior the proctoring tools flag, or what score threshold matters;
• how long recordings and transcripts are kept in calendar terms;
• whether the human evaluator reviews the recording or only the generated report.

Third-party pages state numbers for several of these. None of them cite a micro1 source, and the figures conflict with each other.

What you can ask for

The Candidate Privacy Notice grants candidates the ability to request "Access to, or a copy of, your information," "Deletion of your information," and to "Opt out of the processing of your information via automated decisionmaking." It also names a specific appeal route: "If you are not selected for a position, and you want to appeal the decision based on our use of automated technologies, please email us at [email protected]."

An appeal address that micro1 publishes itself is a stronger footing than most AI-led processes offer, though the notice does not commit to a response time or an outcome. Before you use it, it is worth knowing which requests carry legal weight in your jurisdiction and which depend on goodwill — what you can actually request when you are sent an AI interview sets out that order.

The practical read

Treat a micro1 Zara session as a recorded, screen-captured, proctored assessment whose output — including an integrity score — reaches a human recruiter as a single report, and whose resulting profile may be shown to micro1's clients if you are applying for expert-data work. That is a fair description of the process based on micro1's own documents, and it is materially more than the interview invitation tells you.

The preparation that follows is ordinary: know the role you selected, expect open-ended technical and scenario questions rather than a fixed question bank, and assume the screen you share is the screen that is recorded. If you want a structured approach to the format itself, start with how to prepare for an AI-led interview.