A Karat interview is a technical screen that the hiring company has outsourced. The engineer on the call does not work for the employer you applied to — they work through Karat, follow a scripted rubric, and hand back a recommendation. The session is recorded end to end, and the recording goes to the employer along with your score.
That structure changes what preparation means. There is no rapport to build with a future teammate, no chance to read the room and adjust, and no interviewer discretion to lean on. There is a rubric, a recording, and a second reviewer who can watch the whole thing again. This guide covers what Karat documents about that process as of September 1, 2026, drawn from its candidate and customer materials.
Who is actually interviewing you
Karat calls its interviewers Interview Engineers, abbreviated IVE. Karat's candidate page describes the role plainly: "Your Karat interview will be conducted by an IVE — a professional software developer (just like you!)" (Karat, candidate experience).
They are contractors in a vetted pool, not employees of the company hiring you. Karat states that "to join the Karat IVE community, an engineer must have a minimum of 3 years of experience and a degree in computer science or engineering, as well as fluent English language skills" (Karat customer FAQ).
Two practical consequences follow. First, your interviewer cannot answer questions about the team, the codebase, or the role — they do not have that information. Save those questions for the recruiter. Second, scheduling is unusually flexible because the pool is large and distributed; Karat says 60% of its interviews are scheduled outside core business hours, on nights, or weekends. If your best thinking happens at 8pm, that is an option here in a way it usually is not.
The hour, section by section
Karat's candidate-facing description of the standard interview is specific: "Your interview will be an hour long, and will have three major sections: A brief introduction, 10 min. of discussion questions, 40 min. of programming questions" (Karat).
The ten minutes of discussion questions are the part candidates most often underprepare. They are not warm-up small talk — they are scored. In Karat's own guidance to hiring teams, the mix shifts by level: interns get roughly 10 to 15 minutes of discussion plus 45 to 50 minutes of coding, mid-career engineers get about 10 minutes of discussion, 30 minutes of coding, and 20 minutes of knowledge questions, and senior interviews add past-work discussion and architecture review (Karat, guide to technical interviews).
You do not need to prepare a local environment. Karat runs the session in the browser, telling candidates it can "run everything we need through your web browser" (Karat candidate FAQ). Test your browser, camera, and microphone in advance, because the hour does not stretch to accommodate setup problems.
What gets recorded, and who ends up holding it
This is the part worth reading twice. Karat's candidate FAQ states that "the entire video call, including video and audio, your code, and a full playback, is recorded and shared" with the hiring company. Karat's candidate page adds that "we'll send the recruiter the video recording of the interview, including a summary that highlights your performance on the questions."
A full playback means the code artifact is not just a final snapshot. Reviewers can watch how the solution was built, not only what it ended up being.
Retention is documented on the customer side. Karat says it "deletes candidate data upon the request of the candidate or after a maximum of 5 years from the date of the interview," and that it "retains candidate personal data for as long as a candidate has an active account (but in no event longer than five years from the candidate's last date of activity)" (Karat customer FAQ). Separately, Karat "grants customers a perpetual license to interview results," including the overall recommendation and quantitative scores, which survives the end of the customer's contract.
Read those two facts together. The recording has a five-year ceiling and a deletion path you can invoke. The scored result does not — once an employer holds your recommendation, they hold it. If you reapply to the same company in three years, assume the earlier result is still on file.
Integrity here is a person watching, not a telemetry feed
Most of the proctoring a candidate meets in remote hiring is automated: browser events, webcam analysis, tab-focus logs. Karat's documented approach is different in kind. The reviewer is human, and the signal is behavioral.
Karat tells customers that "each IVE is trained to monitor for and report abnormal candidate behaviors which may indicate cheating, such as typing outside of the browser window without explanation, frequently looking between monitors, or writing code in an unusual top-down or line-by-line manner." When something is flagged, "the interview will be rewatched and reviewed in its entirety by a second IVE specifically trained in Karat's quality control standards" (Karat customer FAQ).
Notice what those three examples have in common: every one of them is ambiguous on its face. Plenty of engineers glance at a second monitor because their notes are there. Plenty write code linearly because that is how they think. The signal is not the behavior itself but the behavior without explanation — Karat's own phrasing.
That points to a concrete habit for this format: narrate. Say "I'm going to check the documentation for this method signature" before you look away. Say "I'm going to sketch this top to bottom and then fill in the helper" before you do it. In a session where a human is judging whether your behavior is explained, explaining it is not a trick — it is the thing being measured. For contrast with the automated model, see what HackerRank proctoring detects and records, which logs events for later review rather than relying on an interviewer's judgment in the moment. If a flag has already happened to you elsewhere, what happens when a coding assessment flags you covers the review process that follows.
The AI rules depend on which module you were assigned
There is no single Karat answer to "can I use AI," and candidates get this wrong in both directions.
Karat's default candidate guidance is restrictive but not absolute. Documentation lookup is fine — candidates may use Google and Stack Overflow, and are told to let the interviewer know when they do. Generation is not: "Using ChatGPT to generate an approach or solution is not allowed" (Karat candidate FAQ).
But the platform is modular, and the employer chooses. Karat tells customers that "some of these modules involve the use of a built-in AI Assistant, others allow for the use of external resources which may include genAI tools, while the remaining modules do not allow the use of external resources" (Karat customer FAQ).
On December 10, 2025, Karat launched NextGen Interviews, which it describes as "the first human-led, AI-enabled talent evaluation solution." In that format candidates "tackle complex, multi-file projects with an integrated AI assistant while collaborating live with Karat's expert Interview Engineers," and the Interview Engineer probes reasoning and trade-offs in real time. Karat's announcement names Atlassian, Duolingo, PayPal, and Citi among its customers (Karat, NextGen Interviews launch).
So the same vendor can run a closed-book screen for one employer and an AI-native project interview for another, in the same week. The only reliable move is to ask your recruiter which module you are scheduled for, in writing, before the call. Verifying AI rules before a technical interview covers how to ask that question so the answer is usable — a tool being available in the environment is not the same as permission to use it.
The redo interview is real, and you have 24 hours
Karat offers something most technical screens do not: a second attempt on request, not on appeal.
Karat describes the mechanism as giving "candidates the chance to redo their first-round technical interview if they believe they didn't do their best." The process runs in four steps — the candidate requests a redo within 24 hours of the original interview, Karat sends an invitation to schedule, the candidate completes an interview with new questions, and both recommendations go to the hiring company (Karat, redo interviews).
The last step matters and is easy to miss. A redo does not replace the first result; it adds a second one, and the employer sees both and decides. That is still worth taking if the first hour genuinely went badly — a weak result plus a strong result reads differently than a weak result alone — but it is not an eraser.
Karat reports that across an analysis of more than 200,000 interviews, redos produced a 20% reduction in false negatives, a 14% improvement in hiring yield, and a 9% boost in candidate progression. Those are vendor-reported figures from Karat's own data, published to sell the feature to hiring teams; treat them as a directional claim rather than an independently verified result.
Rescheduling is separate and more limited: candidates can reschedule up to two times, and should do so at least 24 hours before the scheduled slot.
How a recommendation is produced
Karat's pitch to hiring teams is consistency, and the mechanics reflect that: "detailed scoring rubrics ensuring all candidates are evaluated equivalently" and a "quality control process with a second interviewer validating scoring" (Karat customer FAQ).
Its guidance on building those rubrics tells interview designers to "identify observable candidate actions and results that demonstrate a candidate has those competencies," then translate the completed rubric into a recommendation placing the candidate below, at, or above the hiring bar (Karat, guide to technical interviews).
The candidate-side implication is that unobserved competence does not count. A rubric scores what an interviewer can see and write down. Reasoning you did silently, a trade-off you considered and rejected without saying so, a constraint you noticed but never named — none of it reaches the scorecard. In a format built on observable actions, thinking out loud is not a stylistic preference.
Results are not immediate. Karat tells candidates the hiring company will "reach out to you within a few business days once they've reviewed your Karat results."
What this guide cannot tell you
Most of the detail above comes from Karat's customer-facing FAQ, which describes what the product can do for buyers. It is not a contract with you, and it does not reveal which modules, question sets, or AI configuration your specific employer purchased. Karat also does not publish the rubric itself, so "meets the bar" remains a vendor-and-employer determination you cannot reverse-engineer from the outside.
Two things are worth confirming directly with your recruiter rather than inferring: whether your interview permits AI tools, and whether a redo is available for your employer's process.
Preparing for the format, not just the questions
The distinctive demands of a Karat interview are narration, structure under a timer, and spoken discussion answers — not exotic algorithms. The rehearsal that transfers is a full hour, out loud, with a browser IDE and no local tooling, including the ten minutes of discussion questions most people skip in practice.
Running that as a mock interview with an explicit scoring rubric is the closest available analogue, because it forces the same behavior a rubric rewards: stating the approach, naming the trade-off, and saying what you expect before you run anything.
Control is a desktop AI interview assistant for Windows and macOS, with a free allowance of five messages and two minutes of voice transcription for testing the workflow. For a Karat screen specifically, use it for preparation. Karat's default candidate guidance prohibits using a model to generate an approach or a solution during the interview, and the session you would be using it in is recorded in full and rewatched when anything looks unexplained.
Specific Karat questions
Share
Explore with AI
Continue exploring
Related guides
Assessment Guides · 4 min read
Karat Redo Interview: The 24-Hour Rule and What Employers See
Every Karat candidate can request one redo within 24 hours. New questions, new interviewer, both scores go to the employer. The full policy and Karat's own numbers.
Assessment Guides · 4 min read
What Is a Karat Interview? What the Name Means
Karat is a third-party technical screen: a live hour with a contracted Interview Engineer, recorded and scored for the employer. What the invitation means.
Assessment Guides · 4 min read
Does Karat Record Your Screen?
Karat records the video call, your microphone, and everything typed in Karat Studio, not your desktop. What is captured, who gets it, and how long it is kept.