Pith. sign in

REVIEW 3 major objections 4 minor 63 references

Better Together: Quantifying the Benefits of AI-Assisted Recruitment

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read In a 37,000-applicant trial, an AI-led first interview lifted final-stage hiring pass rates from 34% to 54%.

desk verdict Valuable experiment, overclaimed headline: the 20pp pass-rate gap is an arm-specific pipeline contrast, not a significant ATE. read the letter →

arxiv 2507.08029 v1 pith:YSGVIFUS submitted 2025-07-08 cs.CL cs.CY

classification cs.CLcs.CY
keywords AIrecruitmentrandomizedfieldexperimenthuman-AIcollaborationstructuredvideointerviewresumescreeningskillinflationtreatmenteffecthiringoutcomes
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that inserting an AI-conducted structured video interview before a human hiring stage improves who gets hired. In a randomized field experiment with about 37,000 applicants for a junior developer job, 54% of candidates who came through the AI-assisted pipeline passed the final blind human interview, versus 34% from a resume-screening pipeline, an average treatment effect of 20 percentage points (SE 12 pp.). The paper also reports that five months later, AI-pipeline applicants were more likely to report new employment, and that one in five candidates claimed skills the AI interview rated as absent. If correct, the result means AI screening can surface qualities that resumes miss and can do so at lower recruiter cost.

What carries the argument

The load-bearing mechanism is the AI-conducted structured interview and its automated skill report: an LLM asks and adapts questions, assigns each required skill a four-level label (Not Familiar, Junior, Mid-level, Senior), combines this with a proctoring score and a soft-skills rating, and hands the report to human recruiters, who then select finalists. The comparison arm uses a resume-scoring algorithm plus human recruiter review. All finalists face the same human interview with interviewers blind to pipeline, and the paper's treatment effect is defined conditionally on reaching that final stage ($$\tau = E[Y|W=1,Z=1] - E[Y|W=0,Z=1]$$). The machinery matters because it converts an unstructured screening task into a standardized behavioral assessment, which is what the paper argues captures candidate quality that resumes miss.

What would settle it

If a replication held recruiter selection rules fixed, for example by using the same resume-score threshold in both arms, and the final-interview pass-rate gap fell to near zero, the paper's causal interpretation would be refuted. A simpler check would compare the pass rates of AI-passing candidates selected by recruiters who saw the AI report against a matched group of control candidates with identical recruiter-visible credentials; if the gap disappears, selection differences, not interview quality, drove the result.

Watch

Extended reading notes

Core claim

The central discovery is that an AI-run conversational interview, with an automated skill report that recruiters use for shortlisting, selects candidates who perform substantially better in a subsequent blind human interview than candidates selected by resume screening. The paper estimates the difference in pass rates at 20 percentage points (SE 12 pp.) by simple difference in means, and 26 pp. when adjusting for observed characteristics; it interprets this as evidence that the AI interview elicits information beyond resume observables. In the follow-up, among the 70 finalists, the AI-selected group was 17.2 pp. (SE 8.2 pp.) more likely to report a new job, and the wider comparison of all AI-passing candidates against top-resume candidates showed a 5.9 pp. (SE 2.9 pp.) difference. The paper also finds the AI system ranks younger, less experienced candidates higher than resume scores do, and flags inflated skill claims in about 21% of interviewed candidates.

Load-bearing premise

The claim stands or falls on whether the 35 AI-selected finalists and the 35 resume-selected finalists are comparable enough that the 20-point pass-rate gap is caused by the AI pipeline rather than by different recruiter selection intensity or different information in the two arms.

Editorial extensions

If this is right

  • If the 20-point pass-rate gain is real, employers can reach a hirable shortlist with roughly 44% fewer final interviews, directly cutting recruiter time.
  • AI screening shifts the cost of initial evaluation onto applicants: about 5,600 candidates spent roughly 40 minutes each, trading each saved recruiter hour for about 45 candidate hours.
  • Because the AI system selects younger, less experienced candidates and flags resume inflation, its adoption could widen opportunities for early-career applicants while reducing the value of padded skill lists.
  • The blind final interview means the pass-rate gap is not explained by interviewer knowledge of pipeline, so the improvement is attributed to which candidates the AI stage advances.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension would test whether the same gain appears in non-technical roles; the paper's mechanism of behavioral skill verification suggests the effect should be largest where resume credentials predict performance poorly.
  • The 20-point estimate likely bundles the AI interview's information value with changes in recruiter behavior, because recruiters in the two arms saw different decision aids; a sharper experiment would hold selection intensity fixed across arms.
  • If the AI interview's advantage comes largely from weeding out inflated skill claims, its benefits should shrink in settings with verified credentials or reference checks.
  • The candidate-time cost of about 45 candidate-hours per recruiter hour saved implies that the socially optimal use of AI screening depends on whose time is valued more, a trade-off the paper notes but does not resolve.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper reports a randomized field experiment on AI-assisted recruitment. Approximately 37,000 applicants for a junior developer position on LinkedIn were randomly assigned (70/30) to an AI-assisted pipeline (an AI-conducted structured video interview, followed by human recruiter selection) or to a conventional resume-screening pipeline. The primary outcome is the pass rate in a final human interview that was blind to pipeline assignment: the abstract claims 54% vs 34%, an average treatment effect of 20 percentage points (SE 12 pp.). Secondary analyses examine LinkedIn-documented new employment, dropout and self-selection into the AI interview, demographic heterogeneity, skill misreporting, and comparisons of AI-led versus human-led interview quality. The paper concludes that AI-assisted recruitment identifies higher-quality candidates and exposes resume inflation.

Significance. If the headline estimate were credible, this would be valuable evidence on AI hiring tools from a large, real-world experiment with a blind final interview. The scale (~37,000 applicants), the randomized assignment, and the collection of downstream LinkedIn outcomes are strengths, and the authors are transparent about several limitations (single organization, short observation window). However, as detailed below, the primary causal claim is not identified by the design, and the unadjusted headline estimate is not significant at the 5% level despite the table's star convention. The secondary claim about skill misreporting treats the AI's ratings as ground truth without a human benchmark. The paper therefore contains useful descriptive findings, but the central causal conclusion as currently framed is not supported.

major comments (3)
  1. [Section 4.2, Eq. (1); Table 2] The estimand in Eq. (1) conditions on Z=1, the event of being selected for the final human interview, but this event is not the same across arms. In the treatment arm, Z=1 means passing the AI interview and formal thresholds, then being hand-selected by recruiters from 286 AI-passers. In the control arm, Z=1 means being hand-selected by recruiters from roughly 11,000 resume-screened applicants. Because Z is a post-treatment outcome of the pipeline, random assignment of W does not make the two Z=1 groups exchangeable; conditioning on Z opens a selection/collider path. The 20-percentage-point gap (Table 2) is therefore a descriptive difference between two differently selected subpopulations, not an average treatment effect 'attributable to AI assistance' as the abstract and Section 4.2 claim. The OLS and GRF adjustments do not repair this because selection may act on unobservables such as the AI transcript and recruiter judgment. The authors should reframe the estimate as a pipeline contrast or adopt principal-stratification/bounds methods that acknowledge the post-treatment selection.
  2. [Table 2] The difference-in-means estimate is reported as 0.200 with standard error 0.118, giving a t-statistic of approximately 1.69 and a two-sided p-value of approximately 0.09. Under the paper's own star convention (p<0.05), this result should not receive a ** marker. The OLS and GRF estimates are significant at the 5% level, but the unadjusted headline estimate is not. The abstract's wording, 'yielding an average treatment effect of 20 percentage points (SE 12 pp.)' without noting the lack of conventional significance, overstates the evidence. The table and abstract should be corrected to reflect the actual significance.
  3. [Section 5.3, Table 5] The claim that 'roughly one in five candidates inflate skill lists' defines misreporting as disagreement between the candidate's resume and the AI's proficiency label ('Not familiar'). No human benchmark is provided to establish that the AI's rating is accurate ground truth. The authors themselves note in Section 5.3 that 'some divergence may reflect genuine differences in self-perception rather than intentional deceit' and that 'future work should benchmark the model's judgments against human interviewers,' yet the abstract and conclusion state that the AI 'exposes widespread skill inflation' and 'identifies one in five inflated skill lists.' Without an independent validation, the term 'misreporting' is not justified; the finding should be described as 'AI-rated non-familiarity with claimed skills' or should be supplemented by human evaluation of a subsample.
minor comments (4)
  1. [Appendix B.1, Table 7] The column labeled 'S.E.' does not appear to report standard errors of the difference in means. For example, Years of Experience has a reported S.E. of 5.365, which is close to sqrt(3.87^2 + 3.71^2), i.e., the square root of the sum of variances rather than the standard error of the difference of two means. Please recompute the standard errors and the associated balance tests.
  2. [Section 4.2] The estimator name 'Augmented Inverse Propensity Matching' should be 'Augmented Inverse Probability Weighting' (or 'Augmented Inverse Propensity Weighting').
  3. [Section 6.1] The workload-savings arithmetic is not derivable from the reported pass rates. The text states that the AI screen raised the first-round pass rate from 34% to 54% and that recruiters 'needed to interview 44% fewer candidates before identifying a hirable individual,' but 1 - (0.34/0.54) is approximately 37%, not 44%. The '160 to 82 hours' figures are also unexplained. Please clarify the assumptions behind these numbers.
  4. [General] There are numerous typographical and formatting issues, including missing spaces ('andResume Score'), inconsistent spacing around p-values, and a mismatch between the abstract's total applicant count (37,000) and the sum of the treatment and control sample sizes in Table 1 (32,169). Please proofread carefully.

Circularity Check

1 steps flagged · score 4.0 of 10

Secondary skill-misreporting claim is definitional; main randomized ATE rests on independent human judgments and is not circular.

  1. self definitional [Section 5.3 'Resume vs. AI-Vetted Skill Verification'; Table 5 note]
    "Columns 1–2 use the subset of the treatment group of the randomized experiment that participated in the AI interview and flag candidates as misreporting if the AI interview rates any of the three required skills: React, JavaScript, or CSS as Not familiar. ... A misreport is flagged when the AI recruiter rates a required technology as Not familiar."

    The paper defines the phenomenon it claims to detect—resume skill inflation—as disagreement with the AI interviewer's own proficiency labels ('Not familiar'), with no human or external skill benchmark. The conclusion that 'AI Recruiter identifies candidates who were unable to showcase skills that they reported on their resumes' (Section 6) is therefore true by definition: 'unable to showcase' is operationalized as the AI's rating. The 21.3% and 21.6% rates are rates of the AI's own labels, not independent evidence that resumes are inflated. The paper's own limitation—'Future work should benchmark the model’s judgments against human interviewers'—confirms that no external validity check was used. This is a secondary mechanism claim, not the main ATE.

full rationale

The main quantitative claim is not circular. The primary outcome is a live human interviewer's pass/fail decision made blind to pipeline assignment (Section 4.1: 'interviewers conducted evaluations and made hiring decisions independently, blind to each candidate’s original pipeline assignment'), so the 20-pp difference is an observed contrast between two groups of candidates judged by independent humans; no parameter is fit to the outcome and then renamed a prediction. The covariate-adjusted and AIPW estimates are alternative estimators of that contrast, not reuses of the outcome as input. The self-citations (Athey et al. 2025 for a prediction approach; Athey and Palikot 2022 for LinkedIn outcome magnitudes) are methodological or contextual and are not load-bearing; no uniqueness theorem is invoked. The interview-quality comparison rests mainly on Claude scores, but a blinded human-scored subset independently corroborates the direction and is reported as significant, so the claim does not reduce to an LLM judging itself. The one genuinely circular element is the skill-misreporting analysis in Section 5.3: 'misreporting' is defined as the AI interviewer rating a resume-listed skill as Not familiar, so the finding that one in five candidates 'inflate' skills merely restates the AI's own labels; the paper admits a human benchmark is still needed. Separately, the primary ATE is threatened by non-exchangeability of the two Z=1 groups (35 finalists drawn from 286 AI-passers vs. ~11,000 resume-screened candidates) and by the unadjusted result's t≈1.69 (p≈0.09), but those are causal-identification and significance-reporting problems, not circular derivation.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on several non-experimental choices: hand-set AI pass thresholds, a chosen resume-score cutoff for the control comparison, and the assumption that the AI's ratings are ground truth for misreporting. Random assignment supports the descriptive funnel comparison, but not the causal 'attributable to AI' reading, because the final-stage groups are selected at very different rates with different recruiter information. No new entities are introduced; the AI Score is a hand-weighted index of the AI's own ratings.

free parameters (4)
  • AI interview pass thresholds = Mid-level in all required skills; proctoring score at least 70; soft-skill rating at least Mid-level; at least 2 years…
    Section 4.1: these hand-set thresholds define who 'passed the AI interview' (286 candidates), which determines the pool from which the 35 treatment finalists were drawn and hence the headline 54% pass rate.
  • Top Resume Score cutoff = Resume Score of 98.75 (1,400 control candidates)
    Section 4.2 and footnote 4: the expanded LinkedIn comparison defines the control group as candidates with the top resume score; the 5.9pp estimate depends on this chosen cutoff.
  • AI Score point weights = Senior=3, Mid-level=2, Junior=1, Not familiar=0
    Section 4.1.1: the composite AI Score feeding the rank-reorder analysis ('who benefits from AI') is a hand-chosen weighting of the three skill ratings.
  • Randomization ratio = 70/30 treatment to control
    Section 4.1: the unequal split was chosen to offset anticipated treatment attrition; it shapes the precision and the composition of all downstream comparisons.
assumptions (5)
  • domain assumption LinkedIn-reported new employment is a valid proxy for actual job finding
    Section 4.1 and footnote 2 rely on prior work (Athey and Palikot 2022) that LinkedIn updates reflect new jobs; no correction is made for profiles that are not updated, so the outcome contains unknown measurement error.
  • domain assumption Final-stage interviewers were effectively blind to pipeline, and their pass/fail verdicts measure candidate quality
    Section 4.1: the primary outcome's validity requires the described blinding to hold; the paper does not specify exactly what dossier materials the interviewer saw, so whether the AI report was visible is unverified.
  • ad hoc to paper The AI's proficiency labels are accurate ground truth for detecting skill misreporting
    Section 5.3: the 21% misreporting claim flags anyone the AI rates 'Not familiar'; no human benchmark is provided, making the treatment system the validator of its own screening value.
  • domain assumption Random assignment produced balanced groups
    Section 4.1 and Appendix B.1: the paper claims covariate balance, but the appendix SEs are computed without sample-size division, and corrected math shows a significant Resume Score imbalance (t about 5), so this premise is not verified by the paper's own tables.
  • standard math Normal-approximation inference is valid with 70 selected finalists
    Table 2: all three estimators report p-values from asymptotic machinery applied to 70 observations at the final stage; the GRF-based AIPW is particularly fragile at this sample size, and the 95% CI for the headline estimate includes zero.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Better Together: Quantifying the Benefits of AI-Assisted Recruitment." pith.science (2026). https://pith.science/paper/YSGVIFUS

@misc{pith2026250708029,
  author       = {Pith},
  title        = {Pith review of: Better Together: Quantifying the Benefits of AI-Assisted Recruitment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YSGVIFUS}},
  note         = {Machine review of arXiv:2507.08029}
}
read the original abstract

Artificial intelligence (AI) is increasingly used in recruitment, yet empirical evidence quantifying its impact on hiring efficiency and candidate selection remains limited. We randomly assign 37,000 applicants for a junior-developer position to either a traditional recruitment process (resume screening followed by human selection) or an AI-assisted recruitment pipeline incorporating an initial AI-driven structured video interview before human evaluation. Candidates advancing from either track faced the same final-stage human interview, with interviewers blind to the earlier selection method. In the AI-assisted pipeline, 54% of candidates passed the final interview compared with 34% from the traditional pipeline, yielding an average treatment effect of 20 percentage points (SE 12 pp.). Five months later, we collected LinkedIn profiles of top applicants from both groups and found that 18% (SE 1.1%) of applicants from the traditional track found new jobs compared with 23% (SE 2.3%) from the AI group, resulting in a 5.9 pp. (SE 2.6 pp.) difference in the probability of finding new employment between groups. The AI system tended to select younger applicants with less experience and fewer advanced credentials. We analyze AI-generated interview transcripts to examine the selection criteria and conversational dynamics. Our findings contribute to understanding how AI technologies affect decision making in recruitment and talent acquisition while highlighting some of their potential implications.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 52 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution isbn issn journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in "" FUNCTION format.date year ...

  3. [3]

    Human Resource Management Review 29(1):1--13, ://dx.doi.org/10.1016/j.hrmr.2018.02.009

    Acikgoz Y (2019) Employee recruitment and job search: Towards a multi-level integration. Human Resource Management Review 29(1):1--13, ://dx.doi.org/10.1016/j.hrmr.2018.02.009

  4. [4]

    Agrawal A, Gans J, Goldfarb A (2022) Power and prediction: The disruptive economics of artificial intelligence (Harvard Business Press)

  5. [5]

    An H, Acquaye C, Wang C, Li Z, Rudinger R (2024) Do large language models discriminate in hiring decisions on the basis of race, ethnicity, and gender? arXiv preprint arXiv:2406.10486

  6. [6]

    PNAS nexus 4(3):pgaf089

    An J, Huang D, Lin C, Tai M (2025) Measuring gender and racial biases in large language models: Intersectional evidence from automated resume evaluation. PNAS nexus 4(3):pgaf089

  7. [7]

    Journal of Econometrics 105945

    Athey S, Keleher N, Spiess J (2025) Machine learning who to nudge: causal vs predictive targeting in a field experiment on student financial aid renewal. Journal of Econometrics 105945

  8. [8]

    arXiv preprint arXiv:2211.09968

    Athey S, Palikot E (2022) Effective and scalable programs to facilitate labor market transitions for women in technology. arXiv preprint arXiv:2211.09968

Show all 63 references
  1. [9]

    The Quarterly Journal of Economics 123(1):219--277

    Autor DH, Scarborough D (2008) Does job testing harm minority workers? evidence from retail establishments. The Quarterly Journal of Economics 123(1):219--277

  2. [10]

    Available at SSRN: https://ssrn.com/abstract=4378074

    Avery M, Leibbrandt A, Vecci J (2023) Does artificial intelligence help or hurt gender diversity? evidence from two field experiments on recruitment in tech, working paper, Monash University and University of Gothenburg. Available at SSRN: https://ssrn.com/abstract=4378074

  3. [11]

    Springer Handbook of Automation, 1451--1458 (Springer)

    Balasubramanian P, Balakrishna D (2023) Case study: Infosys--talent management processes automation with ai. Springer Handbook of Automation, 1451--1458 (Springer)

  4. [12]

    Review of Economic Studies 92(3):1438--1496

    Birinci S, See K, Wee SL (2025) Job applications and labour market flows. Review of Economic Studies 92(3):1438--1496

  5. [13]

    Black JS, van Esch P (2020) Ai-enabled recruiting: What is it and how should a manager use it? Business Horizons 63(2):215--226, ://dx.doi.org/10.1016/j.bushor.2019.12.001, artificial Intelligence and Machine Learning

  6. [14]

    Upturn, December 7

    Bogen M, Rieke A (2018) Help wanted: An examination of hiring algorithms, equity, and bias. Upturn, December 7

  7. [15]

    Technical report, Stanford Institute for Human‐Centered Artificial Intelligence, arXiv:2108.07258

    Bommasani R, Hudson DA, Adeli E, Altman R, et al (2021) On the opportunities and risks of foundation models. Technical report, Stanford Institute for Human‐Centered Artificial Intelligence, arXiv:2108.07258

  8. [16]

    Human resource management review 18(3):103--118

    Breaugh JA (2008) Employee recruitment: Current knowledge and important areas for future research. Human resource management review 18(3):103--118

  9. [17]

    Advances in Neural Information Processing Systems 33:1877--1901, ://dx.doi.org/10.48550/arXiv.2005.14165

    Brown TB, Mann B, Ryder N, Subbiah M, Kaplan J, Dhariwal P, Neelakantan A, Shyam P, Sastry G, Askell A, Agarwal S, Krueger G, Henighan T, Child R, Ramesh A, Ziegler DM, Wu J, Winter C, Hesse C, Chen M, Sigler E, Litwin M, Gray S, Chess B, Clark J, Berner C, McCandlish S, Radfo...

  10. [18]

    Brownstein E (2024) Goldman sachs sees a record number of intern applicants for summer 2024 ://www.businessinsider.com/goldman-internship-just-got-harder-summer-analyst-acceptance-rate-2024

  11. [19]

    The Quarterly Journal of Economics 140(2):889--942

    Brynjolfsson E, Li D, Raymond L (2025) Generative ai at work. The Quarterly Journal of Economics 140(2):889--942

  12. [20]

    Science 358(6370):1530--1534

    Brynjolfsson E, Mitchell T (2017) What can machine learning do? workforce implications. Science 358(6370):1530--1534

  13. [21]

    Computers & Education 105:1--13

    Cai Z, Fan X, Du J (2017) Gender and attitudes toward technology use: A meta-analysis. Computers & Education 105:1--13

  14. [22]

    Personnel Psychology 50(3):655--702, ://dx.doi.org/10.1111/j.1744-6570.1997.tb00709.x

    Campion MA, Palmer DK, Campion JE (1997) A review of structure in the selection interview. Personnel Psychology 50(3):655--702, ://dx.doi.org/10.1111/j.1744-6570.1997.tb00709.x

  15. [23]

    ACM Computing Surveys (CSUR) 55(3):1--38

    Cao L (2022) Ai in finance: challenges, techniques, and opportunities. ACM Computing Surveys (CSUR) 55(3):1--38

  16. [24]

    Journal of Marketing Research 56(5):809--825, ://dx.doi.org/10.1177/0022243719827932

    Castelo N, Bos MW, Lehmann DR (2019) Task‐dependent algorithm aversion. Journal of Marketing Research 56(5):809--825, ://dx.doi.org/10.1177/0022243719827932

  17. [25]

    Marketing Science 44(1):30--53

    Chakraborty I, Chiong K, Dover H, Sudhir K (2025) Can ai and ai-hybrids detect persuasion skills? salesforce hiring with conversational video interviews. Marketing Science 44(1):30--53

  18. [26]

    International journal of selection and assessment 11(2-3):113--120

    Chapman DS, Webster J (2003) The use of technologies in the recruiting, screening, and selection processes for job candidates. International journal of selection and assessment 11(2-3):113--120

  19. [27]

    Columbia Business School, Columbia University 29

    Cowgill B (2020) Bias and productivity in humans and algorithms: Theory and evidence from resume screening. Columbia Business School, Columbia University 29

  20. [28]

    (2018) Artificial intelligence for the real world

    Davenport TH, Ronanki R, et al. (2018) Artificial intelligence for the real world. Harvard business review 96(1):108--116

  21. [29]

    (2025) The cybernetic teammate: a field experiment on generative ai reshaping teamwork and expertise

    Dell'Acqua F, Ayoubi C, Lifshitz H, Sadun R, Mollick E, Mollick L, Han Y, Goldman J, Nair H, Taub S, et al. (2025) The cybernetic teammate: a field experiment on generative ai reshaping teamwork and expertise. Technical report, National Bureau of Economic Research

  22. [30]

    Journal of experimental psychology: General 144(1):114

    Dietvorst BJ, Simmons JP, Massey C (2015) Algorithm aversion: people erroneously avoid algorithms after seeing them err. Journal of experimental psychology: General 144(1):114

  23. [31]

    Management science 64(3):1155--1170

    Dietvorst BJ, Simmons JP, Massey C (2018) Overcoming algorithm aversion: People will use imperfect algorithms if they can (even slightly) modify them. Management science 64(3):1155--1170

  24. [32]

    Journal of the Academy of Marketing Science 51(4):802--822

    Fotheringham D, Wiles MA (2023) The effect of implementing chatbot customer service on stock returns: An event study analysis. Journal of the Academy of Marketing Science 51(4):802--822

  25. [33]

    Business News Daily ://www.businessnewsdaily.com/12003-employee-lies-on-resume.html

    Freedman M (2024) People lie on their resumes to get jobs. Business News Daily ://www.businessnewsdaily.com/12003-employee-lies-on-resume.html

  26. [34]

    how leaders can improve hiring practices to uncover missed talent pools, close skills gaps, and improve diversity

    Fuller J, Raman M, Sage-Gavin E, Hines K (2021) Hidden workers: Untapped talent. how leaders can improve hiring practices to uncover missed talent pools, close skills gaps, and improve diversity

  27. [35]

    Journal of Applied Psychology 102(7):1148

    Hardy III JH, Gibson C, Sloan M, Carr A (2017) Are applicants more likely to quit longer assessments? examining the effect of assessment length on applicant attrition behavior. Journal of Applied Psychology 102(7):1148

  28. [36]

    Journal of Business and Psychology 34(1):87--106

    Henle CA, Dineen BR, Duffy MK (2019) Assessing intentional resume deception: Development and nomological network of a resume fraud measure. Journal of Business and Psychology 34(1):87--106

  29. [37]

    Journal of Applied Psychology 107(8):1323

    Hickman L, Bosch N, Ng V, Saef R, Tay L, Woo SE (2022) Automated video interview personality assessments: Reliability, validity, and generalizability investigations. Journal of Applied Psychology 107(8):1323

  30. [38]

    Quarterly Journal of Economics 133(2):765--800

    Hoffman M, Kahn LB, Li D (2017) Discretion in hiring. Quarterly Journal of Economics 133(2):765--800

  31. [39]

    Manuscript, MIT

    Horton J, Vasserman S (2021) Job-seekers send too many applications: Experimental evidence and a partial solution. Manuscript, MIT

  32. [40]

    Horton JJ, Vasserman NS, Stanford G, Watt NM (2024) Reducing congestion in labor markets: A case study in simple market design

  33. [41]

    Working Paper 30612, National Bureau of Economic Research

    Kanazawa K, Kawaguchi D, Shigeoka H, Watanabe Y (2022) AI , skill, and productivity: The case of taxi drivers. Working Paper 30612, National Bureau of Economic Research

  34. [42]

    Koenecke A, Choi ASG, Mei KX, Schellmann H, Sloane M (2024) Careless whisper: Speech-to-text hallucination harms. Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency ( FAccT '24) , 1672--1681 (Association for Computing Machinery), ://dx.doi.org...

  35. [43]

    Management Science

    Koren M (2024) The gatekeeper effect: The implications of pre-screening, self-selection, and bias for hiring processes. Management Science

  36. [44]

    Journal of Personality and Social Psychology 77(6):1121--1134

    Kruger J, Dunning D (1999) Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology 77(6):1121--1134

  37. [45]

    Personnel Psychology 67(1):241--293, ://dx.doi.org/10.1111/peps.12052

    Levashina J, Hartwell CJ, Morgeson FP, Campion MA (2014) The structured employment interview: Narrative and quantitative review of the research literature. Personnel Psychology 67(1):241--293, ://dx.doi.org/10.1111/peps.12052

  38. [46]

    Explainable and interpretable models in computer vision and machine learning 197--253

    Liem CC, Langer M, Demetriou A, Hiemstra AM, Sukma Wicaksana A, Born MP, K \"o nig CJ (2018) Psychology meets machine learning: Interdisciplinary perspectives on algorithmic job candidate screening. Explainable and interpretable models in computer vision and machine learning 197--253

  39. [47]

    Management Science

    Liu M, Tang X, Xia S, Zhang S, Zhu Y, Meng Q (2023) Algorithm aversion: Evidence from ridesharing drivers. Management Science

  40. [48]

    Clark GL, Feldman MP, Gertler MS, W \'o jcik D, eds., The New Oxford Handbook of Economic Geography, 407--435 (Oxford University Press)

    Manning S, Larsen MM, Kannothra CG (2017) Global sourcing of business processes: History, effects, and future trends. Clark GL, Feldman MP, Gertler MS, W \'o jcik D, eds., The New Oxford Handbook of Economic Geography, 407--435 (Oxford University Press)

  41. [49]

    Journal of Economics and Management Strategy 33, ://onlinelibrary.wiley.com/doi/full/10.1111/jems.12576

    McElheran K, Li JF, Brynjolfsson E, Kroff Z, Dinlersoz E, Foster L, Zolas N (2024) Ai adoption in america: Who, what, and where. Journal of Economics and Management Strategy 33, ://onlinelibrary.wiley.com/doi/full/10.1111/jems.12576

  42. [50]

    arXiv preprint arXiv:2405.19699 ://arxiv.org/abs/2405.19699

    Mujtaba DF, Mahapatra NR (2024) Fairness in ai-driven recruitment: Challenges, metrics, methods, and future directions. arXiv preprint arXiv:2405.19699 ://arxiv.org/abs/2405.19699

  43. [51]

    IEEE Transactions on Affective Computing 9(2):191--204

    Naim I, Tanveer MI, Gildea D, Hoque ME (2016) Automated analysis and prediction of job interview performance. IEEE Transactions on Affective Computing 9(2):191--204

  44. [52]

    Journal of Vocational Behavior 122:103471, ://dx.doi.org/10.1016/j.jvb.2020.103471

    Newman DA, Stein JT, Kuenzi M (2020) Privacy calculus and applicant reactions to data‐intensive hiring practices. Journal of Vocational Behavior 122:103471, ://dx.doi.org/10.1016/j.jvb.2020.103471

  45. [53]

    Proceedings of the 3rd ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization, 1--15

    Salinas A, Shah P, Huang Y, McCormack R, Morstatter F (2023) The unequal opportunities of large language models: Examining demographic biases in job recommendations by chatgpt and llama. Proceedings of the 3rd ACM Conference on Equity and Access in Algorithms, Mechanisms, and ...

  46. [54]

    Psychological Bulletin 124(2):262--274

    Schmidt FL, Hunter JE (1998) The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings. Psychological Bulletin 124(2):262--274

  47. [55]

    Schwarcz D, Manning S, Barry P, Cleveland DR, Prescott J, Rich B (2025) Ai-powered lawyering: Ai reasoning models, retrieval augmented generation, and the future of legal practice

  48. [56]

    https://resumelab.com/career-advice/lying-to-get-a-job, accessed 5 July 2025

    Szczepanek A (2023) The truth about lying to get a job [ survey report ] . https://resumelab.com/career-advice/lying-to-get-a-job, accessed 5 July 2025

  49. [57]

    California Management Review 61(4):15--42

    Tambe P, Cappelli P, Yakubovich V (2019) Artificial intelligence in human resources management: Challenges and a path forward. California Management Review 61(4):15--42

  50. [58]

    Journal of the Association for Information Systems 17(5):328--376, ://dx.doi.org/10.17705/1jais.00428

    Venkatesh V, Thong JYL, Xu X (2016) Unified theory of acceptance and use of technology: A synthesis and the road ahead. Journal of the Association for Information Systems 17(5):328--376, ://dx.doi.org/10.17705/1jais.00428

  51. [59]

    Academy of Management Discoveries 4(4):404--409

    Von Krogh G (2018) Artificial intelligence in organizations: New opportunities for phenomenon-based theorizing. Academy of Management Discoveries 4(4):404--409

  52. [60]

    Journal of the American Statistical Association 113(523):1228--1242

    Wager S, Athey S (2018) Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association 113(523):1228--1242

  53. [61]

    Management Science

    Wang H, Zhang Y, Lu T (2025) The power of disagreement: A field experiment to investigate human-algorithm collaboration in loan evaluations. Management Science

  54. [62]

    Human Resource Management 63(4):555--580

    Wingate TG, Bourdage JS (2024) What are interviews for? a qualitative study of employment interview goals and design. Human Resource Management 63(4):555--580

  55. [63]

    Wu S, Wenzel K, Li J, Li Q, Pradhan A, Kushalnagar R, Lea C, Koenecke A, Vogler C, Hasegawa-Johnson M, Su NM, Ratner NB (2025) Speech ai for all: Promoting accessibility, fairness, inclusivity, and equity. Extended Abstracts of the 2025 CHI Conference on Human Factors in Compu...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.