What Eightfold's AI Interviewer Records and Scores

Eightfold AI Interviewer is an autonomous screening interview: a conversational AI asks adaptive questions, evaluates your answers against criteria derived from the job description, and hands a structured summary to a human recruiter who makes the decision. Unlike the asynchronous video interviews many candidates have met since 2020, Eightfold states on its AI Interviewer product page that the system "does not analyze facial expressions, tone of voice, or accent; evaluation is based on what candidates demonstrate through their responses." That single sentence is the most useful thing to know before you sit one, and it changes how you should prepare. This guide covers what the product documents as of September 13, 2026, what an independent bias audit measured, what the system watches for on the integrity side, and which of those guarantees do not apply.

What the interview itself looks like

Eightfold describes AI Interviewer as holding "adaptive interviews" that understand "context and nuance" rather than reading from a fixed script. In practice that means follow-up questions are generated from your previous answer, so two candidates for the same role do not necessarily get the same question list.

The company's "360 Interview" format bundles several evaluation types into one session — functional, coding, soft-skills, and language. Eightfold's product page states that "an engineering interview always includes a coding round" and that "a skilled trades evaluation completes in 20-25 minutes." Treat those as vendor-published descriptions of configuration defaults, not guarantees about the specific interview your employer set up: the length and the mix of rounds are chosen by the company that invited you.

Eightfold is explicit that the AI does not decide: "Human teams retain full control over every outcome — AI surfaces the signal, people make the call." What the recruiter receives is the structured assessment and an intelligence summary, not merely a pass or fail.

What it scores, and what it says it ignores

The claim that Eightfold's AI Interviewer ignores facial expressions, tone, and accent is a meaningful departure from the video-scored model, where delivery and visual signal were historically part of the evaluation surface. If you are used to preparing for the kind of asynchronous video screen where HireVue records and scores your responses, the preparation shifts: rehearsing eye contact and energy buys you less here, while the completeness and specificity of your spoken answers carry more of the weight.

Two cautions belong next to that claim. First, it is a vendor statement about the scoring model, not a statement that no camera is on — Eightfold's own integrity features reference camera manipulation, which implies video capture is available to the system even if the evaluation model does not score appearance. Second, "does not analyze" describes the model's inputs, not the human reviewer's; a recruiter watching a flagged recording is not bound by the model's exclusions.

What the bias audit measured

Eightfold publishes an independent bias audit for AI Interviewer, conducted by BABL AI Inc. and dated "Audited June 29, 2026 | Signed: July 28, 2026." The audit defines a group's scoring rate as the proportion of candidates in that group who scored at or above the overall median score of the full population.

All reported ratios cleared the 0.80 four-fifths threshold, and Eightfold's September 2026 announcement describes its Responsible AI program as including bias audits compliant with New York City's Local Law 144. The caveat worth carrying: the race and ethnicity comparison was run on synthetic data, which tests the model against constructed profiles rather than against how real applicants of different backgrounds actually performed. A passing audit on synthetic inputs is weaker evidence than a passing audit on real ones, and the published summary is the source for both.

The integrity layer: what it flags

Eightfold's product page states that AI Interviewer runs "12+ cheating signals during interviews — including tab switching, camera manipulation, and duplicate voices," and that suspicious activity is flagged "for human review." The company also lists identity verification through ID.me and CLEAR.

Read those three named signals carefully, because they describe the shape of the system:

• Tab switching is a browser-level signal about focus leaving the interview window. It says nothing about what you switched to.
• Camera manipulation targets virtual cameras and injected video feeds — the identity-fraud problem, not the notes-on-a-second-screen problem.
• Duplicate voices targets a second person answering for you, or the same voice appearing across multiple candidate accounts.

All three are aimed primarily at impersonation and proxy interviewing rather than at reference material. Eightfold does not publish the remaining signals, so the "12+" figure is a vendor count you cannot independently verify, and you should not infer from the three named examples what the undisclosed ones do.

Identity verification through ID.me or CLEAR is a separate step from the interview and involves handing government ID or biometric data to a third party under that provider's terms, not Eightfold's. If your interview invitation includes one, the candidate guide to interview identity verification covers what those checks collect and what you can ask before completing one.

Practice mode and employer mode are not the same product

On September 7, 2026, Eightfold ran "AI Career Day," offering anyone 18 or older a free five-to-ten-minute practice screening interview with AI Interviewer, with personalized feedback afterward and no application, resume, or job posting required. The company stated that these practice interviews are "not shared with employers, not used in hiring decisions, not used to train our AI" and are deleted within 30 days.

Those protections attach to the practice event, not to the product. When an employer invites you to an Eightfold interview, the retention period, who can view the transcript, how long the assessment stays attached to your candidate record, and whether it is reused for future requisitions are all governed by that employer's configuration and privacy notice — not by the Career Day terms. If you tried the free version and formed an impression of how your data was handled, do not carry that impression into a real interview. Ask the recruiter.

The practice run is still worth doing on its own terms, because the single most common surprise candidates report about autonomous AI interviews is the format itself: no interviewer to read, no permission to think out loud for thirty seconds, and a follow-up question that arrives the instant you stop talking.

What you can ask for

Eightfold's AI Interviewer sits inside a broader legal argument about AI hiring tools that is unresolved as of September 2026. In Kistler v. Eightfold AI, Inc., No. 3:26-cv-00559, filed in the U.S. District Court for the Northern District of California in January 2026, two California job seekers allege that Eightfold functions as an unregistered consumer reporting agency under the Fair Credit Reporting Act by generating predictive Match Scores about applicants without the disclosure, written authorization, and dispute rights the FCRA requires. Eightfold disputes the characterization, stating that it "does not 'lurk' or scrape personal web history, social media or the like to build 'secret dossiers'" and that candidates can access and correct their data.

Two things to hold separately. The complaint targets Eightfold's candidate matching and scoring, not the AI Interviewer product specifically. And an allegation is not a finding — the case is in early stages and no court has ruled on whether the FCRA applies. What it does establish is that the question of whether you are entitled to see and dispute an algorithmic assessment of yourself is live, not settled.

Independent of that case, several jurisdictions already give candidates concrete entitlements around automated interviews — notice, consent, and in some cases the results themselves. The rules differ by state and by where the employer operates; what AI hiring rules entitle you to know in 2026 sets out which ones attach to you. If the format itself is the problem rather than the data handling, asking the recruiter for a human interviewer is a more realistic request than candidates generally assume, and Greenhouse's 2026 survey suggests plenty of people vote with their feet instead: Fortune reported in May 2026 that around 38% of candidates had withdrawn from a hiring process because it included an AI interview, while about 51% of those who completed one were ghosted or still waiting.

How to prepare for one

Prepare for content, not delivery. Because the evaluation is built from the job description and scored on what you say, the highest-value preparation is having concrete, specific examples ready for each requirement listed in the posting — named systems, numbers, your actual role in the outcome. Vague competence reads as vague.

Expect to be cut off less gracefully than a human would cut you off, and expect no visible reaction to calibrate against. Answer completely the first time rather than waiting for a prompt to elaborate. If the role is engineering and a coding round is included, treat it as a timed assessment and narrate your reasoning in text or speech as the format allows, since that reasoning is the evaluated artifact.

Do not plan around the integrity layer. Tab switching, camera manipulation, and duplicate voices are the three signals Eightfold names publicly out of a larger set it does not disclose, and an interview where a machine logs the session and a human reviews the flags is a poor place to test what a system can see.

The short version

Eightfold's AI Interviewer is a conversational screening tool that scores the substance of your answers, publishes a dated third-party bias audit, runs an undisclosed set of integrity checks aimed mainly at impersonation, and hands a structured summary to a human who decides. Its documented refusal to score appearance, tone, and accent is a genuine improvement over the video-scored generation of tools, and the free practice interview Eightfold opened on September 7, 2026 is a reasonable way to learn the format without an employer watching. What remains unresolved is the part candidates care about most: what happens to the assessment afterward, and whether you get to see it.

Control is a desktop AI interview assistant built for live, human-led interviews and screen-share workflows on macOS and Windows — a different situation from an autonomous screen that logs integrity signals and routes them to a reviewer. If what you need right now is format practice for an AI interviewer, start with the guide to preparing for an AI interview instead.