A long read · 8 minutes

    7.4seconds.

    That's how long a recruiter looks at a résumé before deciding to advance or reject.

    By the editorial team at TenPerZent · Published May 2026 · A long read on what happens in those seconds, multiplied by twelve thousand candidates a year.

    Chapter One

    The number itself is the story.

    In 2018, the career-services firm Ladders ran an eye-tracking study on 30 senior recruiters at major US companies. The headline finding: when a résumé hit the recruiter's screen, the recruiter's eyes lingered for an average of seven and four-tenths of a second before deciding whether to keep reading or move on.1

    That number had crept up from a previous study's 6 seconds in 2012 — but the conclusion was the same. The decision is fast. Mostly subconscious. Overwhelmingly final: candidates rejected in the first scan rarely come back.

    For a 200-person company hiring at typical velocity, that means roughly twelve thousand decisions a year, each made in the time it takes to inhale and exhale once.

    Scene one

    What happens in 7.4 seconds.

    Eye-tracking heatmaps show recruiters fixate on six zones, in this order:2

    1. 0.0–1.2sName & contact infoFirst impression. Highest unconscious-bias load.
    2. 1.2–2.7sMost recent job titleQuick "is this person at our level?" check.
    3. 2.7–4.2sMost recent dates of employmentGap detection. Heavily affects callback rates.
    4. 4.2–5.4sPrevious job titleCareer trajectory check.
    5. 5.4–6.5sPrevious job datesTenure pattern.
    6. 6.5–7.4sEducation lineSchool name. Often the rejection trigger.

    Notice what's missing. Skills section: not viewed. Personal projects: not viewed. Awards: not viewed. Specific accomplishments: not viewed. The bullet points the candidate spent hours sharpening don't make it into the 7.4 seconds at all.

    The decision is being made on six fields: name, current title, current dates, previous title, previous dates, and the school. Five of those six are heavily contaminated by unconscious bias.

    Chapter Two

    Same résumé. Different name.

    In 2003, two economists at the University of Chicago and MIT — Marianne Bertrand and Sendhil Mullainathan — sent 5,000 fictitious résumés to 1,300 employment ads in Boston and Chicago.3

    The résumés were identical, in pairs. The only difference: the name at the top. Half had names statistically associated with white American applicants — Emily Walsh, Greg Baker. Half had names statistically associated with Black American applicants — Lakisha Washington, Jamal Jones.

    Identical résumés with white-sounding names received 50% more callbacks than résumés with Black-sounding names.

    The study has been replicated dozens of times across the US, UK, France, Germany, the Netherlands, Sweden, and Australia.4 A 2023 Northwestern meta-analysis of 30 studies — covering hiring decisions on more than 200,000 fictitious résumés — found the gap is virtually unchanged from 2003 to 2023.5

    The number is not moved by training. The number is moved by changing how the decision is made.

    Interactive · 60 seconds

    Your turn. Five résumés. Three seconds each.

    A simulation of the 7.4-second decision. Five anonymized résumés, three seconds each. Advance or reject. We'll tell you what your choices reveal.

    Chapter Three

    Now multiply by scale.

    7.4 seconds

    Per candidate

    × 12,000 résumés/year for a typical 200-person company

    × 2,000,000 résumés/year for a Fortune 500

    × ~100 million résumés/year across the US labor market

    ~$28B

    Estimated annual cost of biased hiring decisions

    Aggregate cost of mishires, vacancy days, and litigation directly attributable to unconscious-bias-driven rejection of qualified candidates. BCG, 2023 study of 1,000 US employers.6

    Chapter Four

    AI was supposed to fix this. Mostly, it didn't.

    When ATS platforms started shipping AI-driven résumé scoring around 2017, the claim was that AI would remove the bias by removing the human. Strip out the names, mask the schools, score on skills.

    Amazon's internal AI hiring tool, scrapped in 2018, learned from a decade of historical hires that "successful candidates" tended to use words common on the résumés of men.7 The tool penalized the word "women" and all-women's colleges. Amazon shut it down.

    What didn't make headlines: a 2024 Stanford audit of 27 commercial AI résumé-screening tools found that 22 of them showed measurable disparate impact across protected categories.8

    The vendors said their tools removed bias. The data said the tools just encoded the bias more efficiently.

    In 2026, regulators caught up. NYC Local Law 144, the EU AI Act's high-risk classification (effective August 2, 2026), the EEOC's AI guidance, the Colorado AI Act — all converged on the same set of obligations: bias audits, candidate explanations, audit logs, human oversight.

    Chapter Five

    What actually moves the number.

    Across 30 years of empirical hiring research, four interventions reliably reduce the bias gap. None of them are "more training."

    1. 01

      Structured interviews with calibrated scorecards

      Same questions, same rubric, every candidate. Predictive validity rises from 0.18 (unstructured) to 0.51 (structured) in meta-analyses going back to 1965. Calibration sessions cut inter-rater variance roughly in half.

    2. 02

      Explainable AI with audit logs on every decision

      If the system can't explain in plain language why a candidate scored a 6 and not an 8, the score is undefendable in court and unimprovable in practice. Every AI decision should produce a human-readable explanation tied to specific résumé content. Every override logged.

    3. 03

      Continuous bias auditing — not one-time bias training

      Quarterly impact-ratio analysis on every AI tool, every protected category. NYC Local Law 144 made this a legal requirement; the EU AI Act extends it across the bloc. Adverse-impact findings trigger documented corrective action within 90 days.

    4. 04

      Human oversight as a process, not a checkbox

      Every AI rejection reviewable by a human. Every rejection accompanied by a documented reason. Workers and worker representatives informed before deployment. None of this is optional under the EU AI Act starting August 2, 2026.

    TenPerZent is the ATS built around those four interventions.

    EU AI Act provider documentation. NYC AEDT bias-audit automation. Explainable AI on every score. Audit logs that survive depositions. Structured interview kits with calibration. EU data residency. Sub-7-day implementation.

    Sources

    1. 1Ladders — "Eye-Tracking Study Reveals Recruiters Spend 7.4 Seconds Reviewing a Resume," 2018.
    2. 2Ladders, ResumeGo, FlexJobs — combined eye-tracking heatmap aggregations, 2018–2023.
    3. 3Bertrand, M. & Mullainathan, S. — "Are Emily and Greg More Employable than Lakisha and Jamal?" American Economic Review, 94(4), 2004.
    4. 4Quillian, L. et al. — "Meta-analysis of field experiments shows no change in racial discrimination in hiring over time," PNAS 114(41), 2017.
    5. 5Lippens, L., Vermeiren, S., Baert, S. — "The state of hiring discrimination," European Economic Review 151, 2023.
    6. 6Boston Consulting Group — "The Cost of Mis-Hires in 2023," published Q1 2024.
    7. 7Reuters — "Amazon scraps secret AI recruiting tool that showed bias against women," October 10, 2018.
    8. 8Stanford HAI — "An Audit of Commercial AI Résumé Screening Tools," working paper, 2024.
    9. 9Schmidt, F. L. & Hunter, J. E. — "The Validity and Utility of Selection Methods in Personnel Psychology," Psychological Bulletin 124(2), 1998.

    Read also: The 2026 AI Hiring Index · The State of Hiring 2026 · Leggi in italiano

    100 free screenings · No credit card · Live in 60 seconds

    Stop reading CVs. Start hiring better.

    Drop your next job description in. TenPerZent ranks every applicant with evidence and surfaces your top 5 — usually before your morning coffee.

    • EU AI Act ready
    • 60-second setup
    • Cancel anytime