They call it an unfair advantage.
So was the eye.

Less than everyone hopes. The landmark meta-analysis of 85 years of research (Schmidt & Hunter, 1998) found that even the best combinations of hiring methods reach only moderate validity — around .63 to .65. Later corrections by Sackett et al. (2022) cut many published estimates further. Interviews are a filter with noise in them. Treating them as a sacred precision instrument — and calling help with them “cheating” — overstates what the instrument ever was.

Yes — that is the finding. In controlled experiments, Dana, Dawes & Peterson (2013) showed that unstructured interviews made predictions worse than test scores alone, while leaving interviewers just as confident. The conversation feels like insight and measures like noise. A copilot that grounds the conversation in your actual work makes the exchange more informative, not less.

Almost everything except the answer. Stable judgments form from a face in a tenth of a second (Willis & Todorov, 2006). The first 20 seconds of greeting predict final ratings (Prickett et al., 2000). Attractiveness and accent biases are measured at d=.37 and d=.47 (Hosoda et al. 2003; Spence et al. 2022). Interview anxiety degrades qualified candidates (McCarthy et al., 2021, N=8,343). When the room scores names, faces, accents and nerves, a tool that steadies your delivery of what you actually know is a counterweight — not an unfair edge.

Yes. Google abandoned its famous brainteaser interviews after its own analysis found they predicted nothing (Laszlo Bock, 2015). Google’s published hiring guidance now prescribes structured, rubric-scored interviews precisely because unstructured ones are weak predictors. Most interviews you will sit in this year are still unstructured. The people who ran the research kept the noisy format anyway — and you’re the one told to trust it unaided.

It does, measurably. In a live FAANG-style technical experiment, candidates using ChatGPT passed 73% of interviews versus 53% for the control group — and not one of 32 was suspected (interviewing.io). In a preregistered study of 245 video interviews, AI users scored significantly higher on performance and content, with delivery unchanged; the only penalty was perceived honesty (Canagasuriam & Lukacik, 2024). The thing interviews claim to measure goes up. What goes down is how it looks to a skeptical observer — which is why Clozo keeps every word in your voice.

The objection: the candidate agreed to be tested without help. Using a copilot breaks that agreement.

The answer: a rule only binds when it exists, and most employers never wrote one down. Only 19% of job seekers say live-interview AI crosses a line; 27% have never even seen an employer policy on it (Greenhouse, 2025). Meanwhile their side ran AI on you — screening, scoring, rejecting — and 70% of the time never told you (Greenhouse, 2026). Meta now runs pilot interviews with AI built in and grades how well you use it. Amazon bans it. The norm is contested, not settled — and the rulebook was edited quietly, by one side, for a decade.

The honest residual: if a specific employer bans AI and you use it anyway, you are breaking their rule — their house, their call. Know the rules of your room. That’s not a reason to be afraid; it’s a reason to be deliberate.

The objection: the interviewer believes they are hearing your unaided thinking. Anything else is deception.

The answer: you speak every word. The copilot arranges what you already know — built from your résumé and the live conversation — and the judgment about which parts to say, and how, stays yours. In the only preregistered experiment on live-interview AI, performance and content went up while delivery was unchanged; the sole penalty was how honest it looked to evaluators (N=245, 2024). Deception is claiming competence you don’t have. Articulating competence you do have is called explaining.

The honest residual: the interviewer’s assumption of unaided thinking is real, and your answer changes it — that is a choice you own. Own it deliberately.

The objection: the skill that won the offer wasn’t yours. The job will demand the real one.

The answer: training wheels, not a wheelchair. The answers are grounded in your actual work — you still deliver them, live, in your own voice, and you still have to back them up when the interviewer digs in. The job itself will hand you AI anyway: the market pays a 56% wage premium for AI skills (PwC), and Meta already tests candidates on using AI inside the interview. Practicing with the copilot is practicing for the job as it actually is.

The honest residual: if you claim competence you don’t have, that debt comes due in the role. Don’t claim what you can’t cash. That’s the line — and we draw it below.

The objection: you’re competing against honest people playing by the old rules.

The answer: the old rules were never fair. Identical résumés, different names: 50% fewer callbacks. Accent bias alone is measured at d=.47. 88% of employers admit their filter discards qualified people. AI assistance lifted novice workers 34% — by spreading the knowledge of the top performers (NBER). The people the current filter already disadvantages are the ones the copilot helps most. Fairness is leveling the field, not defending a tilted one.

The honest residual: some rooms will still judge AI use, and norms vary. Unfair feelings don’t disappear because the argument is good. That’s why the choice stays deliberate — and yours.

The objection: the interview measures the job. Pass it artificially and you’re a fraud in the seat.

The answer: the interview barely measures the job even at its scientific best — unstructured interviews can make predictions worse than tests alone, while interviewers feel more certain. Google killed its brainteasers for predicting nothing. Meanwhile the job itself is AI-assisted: passing an interview with AI tests exactly the skill the role now requires — performing with the tools everyone uses.

The honest residual: in roles where unaided recall genuinely is the job — some medical, legal or safety-critical work — the line moves. Know your room, and place yourself honestly on it.

The objection: detection software is real, funded and improving.

The answer: in live settings it almost never fires. In interviewing.io’s controlled FAANG-style experiment, zero of 32 AI-assisted candidates were suspected by any interviewer — and only 11% of those companies even use detection software. 81% of those interviewers suspect AI in the abstract; almost none can point at it in the room. Clozo’s architecture is built for exactly this world: a native macOS app whose window is excluded from screen sharing by macOS itself, nothing joining the call, nothing in the browser — and the app verifies its protection is live during every session.

The honest residual: the arms race is real and tools evolve; a tiny minority of employers run dedicated detection. That’s why the architecture is native and OS-level rather than a browser overlay — and why we publish the matrix, axis by axis, here.

The objection: the signal dies when the whole market optimizes. Everyone loses.

The answer: the loop was already running — and employers started it. Applications per recruiter are up 412%; 21% of companies reject with no human check; 1 in 7 postings is a ghost. Greenhouse, the hiring platform, calls it an “arms race, not a hiring process” and an “AI doom loop.” Their own CEO’s fix: “We don’t need more hoops or friction; we need better signals.” A candidate speaking grounded answers in their own words is a better signal — better than a keyword-optimized résumé ever was.

The honest residual: the systemic question is real, and no single candidate can answer it. Your interview, though, is still yours to win — this year, in this market.

Only the rule-maker can say — and most never wrote the rule down. Only 19% of US job seekers say using AI in a live interview crosses the line, and 27% have never even seen an employer policy on it (Greenhouse, 2025). Meanwhile 63% of job seekers have faced an AI-run interview, and 70% were never told (Greenhouse, 2026). Meta runs pilot interviews with AI built in and grades how well candidates use it. The rule being “broken” was never shown to you. Our position: you speak every word, the answers are grounded in your own résumé, and honesty is yours — see where your line is.

Deception means making someone believe something false. A candidate who uses a copilot grounded in their own résumé, and speaks every answer in their own words, asserts nothing false — the preregistered Canagasuriam & Lukacik study (2024) found AI-assisted candidates scored higher on performance and content, with only perceived honesty lower. The honest residual: where an employer explicitly forbids AI, concealment can become its own wrong — which is why we say know the rules of your room, and never claim competence you can’t cash. Faking competence you don’t have is a different act with a different name: fraud.

Yes — before, during and after. A ResumeBuilder survey of 948 business leaders (October 2024) found 51% of companies use AI in hiring, 82% of them review résumés with it, and 21% reject candidates at some stage with no human review. Harvard Business School found 88% of employers agree qualified candidates get vetted out for not matching exact keywords. Employers have run AI-scored video interviews since 2017, when Unilever adopted them.

Rarely, in live settings. In interviewing.io’s controlled experiment with FAANG-style technical interviews, not one of 32 AI-assisted candidates was suspected by any interviewer, and only 11% of those companies even use detection software. 81% of FAANG interviewers suspect AI use in general — suspicion is common, verification is not. Interview Clozo is a native macOS app whose window is excluded from screen sharing by macOS itself, with nothing joining the call and nothing in the browser; the app verifies its protection is live during the session.

The baseline was never fair. White-sounding names received 50% more callbacks in the famous NBER résumé experiment; attractiveness and accent biases are measured at d=.37 and d=.47; 88% of employers admit their filters discard qualified people. AI assistance lifted novice workers 34% in the NBER field study — it spreads the advantage the top already had. Fairness is leveling the field, not preserving a biased starting line.

If the answers are grounded in your real work, yes — the copilot arranges what you know; you still deliver it, and the job itself will have AI in it. PwC measures a 56% wage premium for AI skills, and Meta now runs interviews with AI built in, grading how well candidates use it. The honest line: don’t claim competence you can’t cash. A real person answering in their own words is the opposite of the deepfake fraud the DOJ prosecutes.

It depends on the room — and the norm is contested, not settled. Meta allows AI in its pilot coding interviews; Amazon has told candidates not to use it. Greenhouse, the hiring platform, calls AI-enabled interviewing “an arms race, not a hiring process,” and its research found only 19% of job seekers think live-interview AI crosses the line. 22% of job seekers already report using AI during live interviews. Our guidance: know the rules of your room — and never fake a competence you don’t have.

Three main lenses, and none of them settles it alone — we work all three in the lenses section above. Consequentialists ask about outcomes: interviews already mispredict job performance (structured interviews validate around .63–.65), so help that lets a qualified, anxious candidate show what they know can improve outcomes — as long as the competence is real. Deontologists ask about duties: the duty of honesty bars asserting what is false, not using help to assert what is true; the duty to respect rules binds you where a rule was actually declared. Virtue ethicists ask about character: help that amplifies real skill is prudence; help that substitutes for skill you lack is where character leaks away. Our position sits where all three agree: grounded, spoken by you, inside the rules of your room.