AI Adoption as Screening Design: When Candidate Choice Becomes Signal

Listen to this article:

Abstract: This article examines how firms should integrate artificial intelligence into labor-market screening when applicants can choose between human and AI interviewers. Drawing on a natural field experiment involving 70,000 job applicants and recent theoretical advances in mechanism design, we show that AI adoption is fundamentally a design problem rather than a simple substitution decision. When applicants select their preferred interviewer, this choice itself becomes an informative signal about underlying abilities—a phenomenon we term "choice-as-signal." The welfare implications depend critically on whether firms incorporate this signal into hiring decisions and whether applicants anticipate such use. Evidence suggests that hybrid screening systems combining human and AI evaluation outperform either technology alone, and that specialized assignment—matching each screener to the dimensions they assess most accurately—can improve match quality. These findings challenge conventional automation narratives and reveal novel trade-offs between worker autonomy and information revelation in AI-augmented hiring.

The integration of artificial intelligence into organizational hiring represents one of the most consequential applications of algorithmic decision-making. Unlike many AI deployment contexts where firms unilaterally control technology adoption, labor-market screening increasingly involves applicant agency: candidates may choose whether to be evaluated by human recruiters, AI systems, or both. This design feature—granting applicants choice over screening technology—appears frequently in practice and policy proposals, often framed as protecting worker autonomy in an era of algorithmic management. Yet choice creates a subtle informational challenge. When applicants select their preferred interviewer, they reveal something about themselves. High-ability candidates may prefer technologies that accurately signal their strengths, while weaker candidates may seek noisier evaluation processes that obscure deficiencies. This self-selection transforms interviewer choice from a procedural nicety into a potentially decision-relevant signal.

Recent evidence demonstrates that AI voice agents can conduct effective job interviews, in some cases matching or exceeding human recruiters on predictive validity and match quality (Jabarian & Henkel, 2025). The question is no longer whether AI can screen applicants, but how human and AI screening should be jointly designed. Should firms assign all candidates to whichever technology performs best on average? Should they allow applicants to choose, respecting autonomy but potentially introducing adverse selection? Or should they deploy both technologies in complementary ways, perhaps specializing each on different skill dimensions or combining their signals?

This article develops a framework for analyzing these design choices and quantifying their welfare implications. We model screening as a setting where applicants possess multidimensional abilities that human and AI interviewers assess with different precision across dimensions. Applicants who know their own strengths and weaknesses choose the interviewer most likely to generate favorable signals. Firms observe these interview signals and, depending on institutional rules, may or may not condition hiring decisions on the choice itself. We analyze three information structures: one where firms ignore choice, one where firms incorporate it but applicants do not anticipate this, and a full equilibrium where both sides recognize that choice is informative.

The theoretical analysis yields striking predictions. When firms ignore choice, offering it strictly benefits applicants and harms firms through adverse selection. When firms internalize choice as a signal but applicants remain naive, firms always gain while low-ability applicants suffer. In equilibrium, firms typically benefit, high-ability applicants improve their prospects, and low-ability applicants face worse outcomes than under predetermined assignment. These welfare reversals reveal a fundamental tension: granting choice as a worker protection can, paradoxically, harm the workers it aims to help once firms rationally update on the information choice conveys.

We complement this theoretical framework with preliminary empirical estimates derived from a large-scale natural field experiment conducted in partnership with a recruitment process outsourcing firm. The experiment randomly assigned 70,000 applicants to be interviewed by a human recruiter, an AI voice agent, or given the option to choose between them. Using data on interview scores, standardized test performance, hiring decisions, and early retention, we estimate key model primitives including the relative precision of human and AI screening across language and analytical skill dimensions, the firm's weighting of these dimensions, and the hiring threshold.

Preliminary estimates suggest both human and AI interviewers are highly informative relative to standardized tests, with human screeners slightly more precise on language skills and AI more precise on analytical evaluation. These patterns motivate counterfactual analyses of hybrid screening systems. Simulations indicate that combining human and AI interviews—observing both signals for each candidate—substantially improves match quality relative to either technology alone, raising job offers by approximately 7 percent while reducing involuntary separations by roughly 24 percent. Specialized assignment, where each screener focuses on the dimensions they measure most accurately, yields more modest but still meaningful gains. These preliminary results suggest that optimal AI adoption involves designing complementary roles for human and algorithmic judgment rather than wholesale substitution.

The AI Adoption Landscape in Organizational Screening

AI adoption in firms has accelerated across diverse organizational functions, from production automation to customer service to knowledge work (Brynjolfsson & McAfee, 2014; Iansiti & Lakhani, 2020; McElheran et al., 2024). Recent evidence documents productivity gains when generative AI tools assist customer-service agents (Brynjolfsson et al., 2025) and knowledge workers (Dell'Acqua et al., 2023; Peng et al., 2023), although effects vary substantially across workers and tasks. In hiring and talent management specifically, AI systems increasingly screen résumés, conduct initial interviews, and assess candidate fit (Cowgill, 2019; Hoffman & Stanton, 2024).

Defining screening technologies in the hiring context

Labor-market screening aims to reduce information asymmetry about applicant quality before making costly employment commitments. Traditional screening methods include résumé review, standardized tests, work samples, reference checks, and interviews. Each method generates noisy signals about different dimensions of worker capability. Interviews, for instance, may reveal communication skills and cultural fit but provide limited information about technical proficiency; conversely, coding assessments measure programming ability but reveal little about interpersonal effectiveness.

AI-based screening introduces technologies with potentially different measurement properties. Automated résumé parsing may efficiently extract structured credentials but struggle with nonstandard career paths. AI voice agents can conduct structured interviews at scale, potentially standardizing question sequences and reducing interviewer fatigue or bias, yet may underperform humans on subjective judgment or contextual interpretation (Aka et al., 2025; Jabarian & Henkel, 2025). Crucially, human and AI screeners need not be uniformly better or worse—they may exhibit comparative advantage across skill dimensions, each excelling where the other struggles.

Prevalence, drivers, and distribution of AI screening

The adoption of AI in recruitment has grown rapidly, driven by several factors. High-volume hiring environments—such as call centers, retail, and logistics—face overwhelming application volumes that strain human recruiter capacity (Berg et al., 2018; Buesing et al., 2020). In these settings, AI screening offers scalability: a single voice agent can conduct hundreds of interviews daily at consistent quality. Labor-market tightness in high-income countries has intensified competition for talent, pressuring firms to accelerate hiring pipelines and improve candidate experience. AI agents that schedule interviews flexibly and reduce time-to-offer may improve applicant satisfaction and acceptance rates.

Technological advances have also enabled more sophisticated AI screening. Large language models now generate contextually appropriate follow-up questions, automatic speech recognition handles diverse accents and speech patterns, and text-to-speech synthesis produces natural conversational flow (Aka et al., 2025). These capabilities allow AI agents to move beyond simple keyword matching toward interactive assessments that approximate human interviews. Simultaneously, the commoditization of traditional application materials—as generative AI helps candidates polish résumés and cover letters—has reduced the signal value of written submissions (Cowgill et al., 2024; Wiles & Horton, 2024; Galdin & Silbert, 2025), increasing the relative importance of live screening interactions.

Adoption remains uneven across industries and organizational contexts. Technology firms and large enterprises with dedicated people-analytics functions lead adoption, while smaller organizations and sectors with strong human-judgment norms (such as professional services) lag. Regulatory uncertainty, particularly around algorithmic fairness and transparency, also shapes adoption patterns. Emerging "right to human review" provisions in data-protection frameworks may constrain full automation, though they do not necessarily prohibit AI-assisted screening where humans make final decisions.

Organizational and Individual Consequences of AI Screening

Organizational performance impacts

AI screening affects organizational performance through multiple channels. Most directly, improved screening precision raises match quality—the alignment between worker capabilities and job requirements. Better matches reduce training costs, lower early turnover, and improve productivity (Jovanovic, 1979; Autor & Scarborough, 2008). Evidence from natural field experiments demonstrates that AI screening can improve retention relative to human-only processes. Jabarian and Henkel (2025) find that applicants interviewed by AI voice agents exhibit higher retention rates at 1-, 2-, 3-, and 4-month horizons compared to those interviewed by human recruiters, consistent with more accurate assessment of job-relevant skills and attrition risks.

Screening precision also shapes the composition of the hired workforce. When AI systems measure certain skill dimensions more accurately than humans, they may select workers with different attribute profiles. If, for instance, AI excels at evaluating problem-solving ability but human interviewers better assess interpersonal skills, shifting from human to AI screening will tend to hire more analytically strong and fewer interpersonally strong candidates, even holding overall ability constant. These compositional shifts matter when teams require balanced skill mixes.

Scalability represents another performance dimension. Human recruiters face capacity constraints: they tire across successive interviews, availability limits scheduling flexibility, and training new recruiters imposes onboarding costs. AI systems eliminate these constraints, enabling firms to interview vastly more candidates without proportional cost increases. This scalability particularly benefits organizations facing high application volumes or rapid hiring ramps. However, scalability advantages accrue primarily in standardized, high-volume settings; for specialized or executive roles requiring nuanced judgment, human involvement remains valuable.

Individual wellbeing and stakeholder impacts

For applicants, AI screening affects both process experience and selection outcomes. On the experience dimension, AI interviews offer scheduling flexibility—candidates can complete interviews at convenient times rather than coordinating with recruiter calendars—and may reduce anxiety for applicants uncomfortable with human evaluation. Conversely, some candidates report diminished engagement or concern that AI cannot appreciate unique backgrounds (Ash et al., 2025). Transparency about AI use matters: explicitly disclosing that an AI agent conducts the interview, as in Jabarian and Henkel (2025), respects autonomy and may improve acceptance.

Selection outcomes introduce more complex welfare considerations. When AI systems measure skills more accurately, they improve the match between workers and jobs, benefiting high-ability applicants who might be undervalued in noisier human screening. However, if AI systems measure different dimensions than humans, they may disadvantage applicants whose strengths lie in areas AI assesses poorly. The distributional consequences depend on which skill dimensions AI measures well and which types of applicants possess those skills.

Choice-based systems, where applicants select their preferred interviewer, introduce further nuance. Choice appears to empower applicants, respecting their autonomy over how they are evaluated. Yet if firms condition hiring decisions on this choice—inferring that applicants selecting noisier screeners may be weaker candidates—choice can harm precisely the low-ability workers it ostensibly protects. This welfare reversal emerges because choice reveals private information, enabling firms to make sharper inferences about applicant quality.

Downstream employment outcomes also matter. Better screening matches reduce the probability that workers enter jobs poorly suited to their abilities. Mismatches generate stress, underperformance, and eventual separation—costly for both worker and firm. By improving initial match quality, more precise screening (whether human, AI, or hybrid) can increase job satisfaction and tenure. Evidence from high-turnover environments like call centers suggests that even modest improvements in retention translate into substantial welfare gains given the frequency and cost of replacement hiring (Berg et al., 2018; Sallaz, 2019; Buesing et al., 2020).

Evidence-Based Organizational Responses to AI Screening Design

Table 1: AI and Human Screening Strategies in Recruitment

Recruitment Strategy	Screening Method	Organization/Sector	Skill Dimensions Measured	Performance Impact	Applicant Consequences	Human Role (Inferred)	Source
Hybrid Human–AI Screening	Sequential dual interviews or Parallel independent assessment	Technology Sector	Analytical/Technical (AI) and Interpersonal/Cultural (Human)	7% increase in job offers; 24% reduction in involuntary separations	Improved match quality and better job satisfaction; benefits high-ability candidates	Reviewing system design, assessing collaborative skills, and performing final holistic review of aggregate signals.	[1]
Dimension-Specific Screener Assignment	Skill-dimension partitioning	Financial Services	Quantitative/Analytical (AI) and Communication/Ethics (Human)	Increased screening efficiency and assurance of dual standards (technical and interpersonal)	Efficient processing for thousands of applicants; ensuring final hires meet interpersonal standards	Specialized judgment in nuanced domains like ethical judgment and client-facing cultural alignment.	[1]
Choice-as-Signal / Choice-based Systems	Applicant-selected interviewer (Human vs. AI choice)	Retail Sector	Underlying abilities revealed through self-selection/confidence	Improved match quality by weighting AI selection positively	Low-ability applicants face worse outcomes; high-ability applicants improve prospects	Strategic design of the choice architecture and interpreting the informational value of the candidate's selection.	[1]
Review-based / Human-in-the-loop	AI initial screening followed by human review/follow-up	Healthcare Sector	Clinical knowledge/availability (AI) and Empathy/Patient-care (Human)	Maintained applicant trust while scaling initial capacity	Access to human review for narrow misses; feedback provided for AI performance	Final decision-maker; provides empathetic assessment and reconsideration of edge cases.	[1]
Transparent AI Integration	AI voice agent (Anna AI) with identity disclosure	PSG Global Solutions (Recruitment Process Outsourcing)	Initial interview responses for customer-service roles	Efficient handling of tens of thousands of applications	Maintained procedural fairness and trust through explicit communication	Reviewing conversational transcripts and exercising final judgment on hiring decisions.	[1]

Organizations adopting AI screening face several design choices beyond the binary decision of whether to deploy the technology. Evidence from research and practice identifies multiple effective approaches, each tailored to specific organizational contexts and objectives.

Transparent Communication and Disclosure Strategies

Transparency about AI use in hiring shapes applicant perceptions and behavior. When candidates know they will interact with an AI agent, they can prepare differently than for human interviews—for instance, focusing on clarity and structure rather than rapport-building. Disclosure also respects autonomy: applicants retain the choice to participate or withdraw if uncomfortable with algorithmic evaluation. Research on transparency in algorithmic decision-making suggests that clear communication reduces confusion and builds trust, though effects depend on how information is framed (Bernheim, 2016).

Effective approaches

• Explicit disclosure at interview invitation: Inform applicants during scheduling whether they will be interviewed by a human or AI agent. Jabarian and Henkel (2025) disclosed AI identity at the start of each interview, stating that a human recruiter would review the conversation and make final hiring decisions.

• Explanation of AI capabilities and limitations: Describe what the AI can and cannot assess—for instance, clarifying that the agent evaluates responses but a human reviews overall fit.

• Opt-out mechanisms: Allow applicants uncomfortable with AI screening to request human interviews without penalty, preserving inclusivity.

• Feedback loops: Provide candidates with information on how interviews inform decisions, whether conducted by human or AI, building procedural fairness.

PSG Global Solutions, a recruitment process outsourcing firm serving Fortune 500 and European enterprises, deployed an AI voice agent named "Anna AI" to conduct initial job interviews for customer-service roles. At the outset of each AI interview, the agent disclosed its artificial identity and explained that a human recruiter would review the conversation transcript and make the hiring decision. This transparency ensured applicants understood the screening process while reassuring them that human judgment remained central to employment decisions. By explicitly communicating the AI's role and limitations, PSG maintained procedural fairness and applicant trust even as it scaled interview capacity to handle tens of thousands of applications efficiently.

Hybrid Human–AI Screening Systems

Combining human and AI screening can outperform relying on either alone when the two technologies measure different skill dimensions with varying precision. If human interviewers excel at assessing interpersonal skills and AI systems excel at evaluating analytical ability, observing both signals improves overall inference about candidate quality. Preliminary simulation evidence from Jabarian and Henkel (2025) suggests that firms observing both human and AI interview signals experience higher job offers (approximately 7 percent increase) and substantially lower involuntary separations (roughly 24 percent reduction) compared to using either screener alone. These gains arise because combining signals reduces posterior uncertainty about multidimensional candidate types, improving match quality without changing selectivity thresholds.

Effective approaches

• Sequential dual interviews: Conduct both human and AI interviews in randomized order, then aggregate signals before making hiring decisions. This approach maximizes information while controlling for order effects documented in recruitment research (Radbruch & Schiprowski, 2025).

• Parallel independent assessment: Have human and AI screeners evaluate the same candidate independently, then reconcile assessments. Independence prevents anchoring, where one screener's judgment biases the other.

• Weighted aggregation: Combine human and AI assessments using weights that reflect their relative precision on relevant dimensions. If AI measures technical skills more accurately and humans assess cultural fit better, weigh AI signals more heavily for technical roles and human signals more for client-facing positions.

• Review-based models: Use AI to conduct initial screening at scale, flagging high-potential candidates for human follow-up interviews. This leverages AI's scalability while preserving human judgment for final decisions (Chakraborty et al., 2024).

Several technology firms implementing AI-assisted hiring have adopted hybrid models where AI agents conduct initial technical screenings—asking coding problems, evaluating algorithmic thinking—while human engineers conduct follow-up interviews assessing system design, communication, and cultural alignment. This division of labor exploits AI's comparative advantage in standardized technical evaluation and humans' strength in subjective judgment. By combining both signals, these organizations improve the accuracy of hiring decisions for roles requiring both technical depth and collaborative skills.

Dimension-Specific Screener Assignment

When human and AI screeners exhibit comparative advantage across different skill dimensions, assigning each to the dimensions they assess most accurately can improve screening efficiency. Preliminary estimates suggest human interviewers in the Jabarian and Henkel (2025) field experiment measured language skills slightly more precisely (precision parameter 7.95) than AI agents (7.86), while AI measured analytical skills more precisely (9.66) than humans (7.55). Although differences are modest, specialized assignment can still generate gains, particularly when screener costs differ. Even small precision advantages compound across high-volume hiring.

Effective approaches

• Skill-dimension partitioning: Divide assessment tasks by skill domain—assign AI to evaluate technical or analytical dimensions and humans to assess interpersonal or subjective fit dimensions.

• Interview module specialization: Structure interviews as sequences of modules (technical, behavioral, situational) and assign each module to the screener type best suited to evaluate it.

• Adaptive routing: Use initial screening signals to route candidates to specialized follow-up assessments. For instance, candidates demonstrating strong analytical skills in AI screening might receive human interviews focused on leadership potential.

• Cost-precision optimization: When one screener type is more expensive (e.g., human recruiters command higher wages than AI agent runtime costs), assign that screener selectively to dimensions where they hold clear precision advantages, using the cheaper screener for other dimensions.

A global financial services firm restructured its entry-level analyst hiring to exploit screener comparative advantages. AI agents conducted initial interviews focusing on quantitative reasoning, problem-solving, and attention to detail—dimensions AI assessed reliably through structured problem sets. Candidates advancing from AI screening then participated in human interviews concentrating on client communication skills, ethical judgment, and cultural alignment—areas requiring nuanced human evaluation. This specialization allowed the firm to screen thousands of applicants efficiently while ensuring final hires met both technical and interpersonal standards essential for client-facing analytical roles.

Choice Architecture and Applicant Selection Mechanisms

Offering applicants choice over screening technology respects autonomy but introduces self-selection that firms can exploit as an informational signal. When applicants choose interviewers, they reveal preferences that correlate with underlying abilities—stronger candidates may prefer more precise screeners that accurately signal their quality, while weaker candidates may favor noisier screeners hoping randomness will help them exceed hiring thresholds. Theoretical analysis shows that once firms condition hiring decisions on this choice, low-ability applicants are harmed relative to predetermined assignment, while high-ability applicants benefit. Firms typically gain unless they fail to incorporate the informational content of choice (Jabarian & Reshidi, 2025).

Effective approaches

• Informed choice with disclosure: Offer applicants choice between human and AI interviewers while clearly communicating that the choice itself may inform hiring decisions. This transparency supports equilibrium behavior where applicants select optimally given firm inference.

• Restricted conditioning: Regulate whether firms can condition hiring decisions on screener choice. Policies prohibiting such conditioning protect low-ability applicants but sacrifice informational efficiency.

• Choice as screening stage: Treat interviewer selection as an explicit early-stage screen, where choosing a particular technology triggers differentiated evaluation pathways. This makes the informational role of choice transparent.

• Randomized recommendation with opt-out: Assign applicants to human or AI screeners by default but allow opt-out requests. This design reduces adverse selection (most applicants accept default) while preserving autonomy for strongly opinionated candidates.

A major retail chain piloting AI voice interviews for seasonal hiring offered applicants the option to choose between human phone interviews and AI-agent interviews. Initially, the firm ignored choice when evaluating candidates, treating interview content as the sole signal. After observing that applicants selecting AI tended to perform better in subsequent employment, the firm revised its policy to positively weight AI selection in hiring decisions—interpreting the choice as signaling confidence in one's abilities. This adjustment improved match quality but also raised questions about fairness, prompting the firm to disclose the role of choice in its hiring rubric to ensure transparency and procedural justice.

Procedural Justice and Candidate Experience

Procedural justice—the perceived fairness of decision-making processes—affects applicant reactions to hiring systems, including acceptance rates, employer brand, and legal compliance (Bernheim & Rangel, 2007, 2009). When candidates feel AI screening is opaque, biased, or impersonal, they may withdraw from hiring processes or develop negative perceptions of the employer. Evidence suggests that transparency, explanation, and opportunities for human review improve perceptions of fairness in algorithmic systems. Providing candidates with feedback, ensuring consistent treatment, and allowing appeals or clarifications can mitigate concerns about AI-driven hiring (Ambuehl et al., 2021; Bushong et al., 2025).

Effective approaches

• Human-in-the-loop final decisions: Use AI for screening but retain human decision-makers for final hiring. This reassures candidates that a person reviews their application holistically.

• Standardized evaluation criteria: Apply the same rubrics and thresholds across human and AI interviews, ensuring consistent treatment regardless of screener type.

• Candidate feedback mechanisms: Offer applicants feedback on their interview performance, whether screened by human or AI, to support learning and improve perceptions of fairness.

• Appeals and reconsideration processes: Allow candidates who believe AI screening missed important qualifications to request human review, providing a safety valve for edge cases.

A healthcare provider implementing AI-based screening for nursing roles faced applicant concerns that AI could not appreciate the empathy and patient-care skills central to nursing. In response, the organization adopted a hybrid approach: AI agents conducted initial screenings focused on clinical knowledge and availability, while human nurse managers conducted follow-up interviews assessing interpersonal skills and cultural fit. Additionally, the provider offered all candidates feedback on their AI interview performance and allowed those narrowly missing cutoffs to request a second human interview. These procedural safeguards maintained applicant trust and ensured that the AI system complemented rather than replaced nuanced human judgment.

Building Long-Term Organizational Capability in AI-Augmented Screening

Effective AI adoption in hiring extends beyond selecting technologies to building enduring organizational capabilities that sustain performance as labor markets and technologies evolve.

Continuous Learning and Model Refinement

AI screening systems require ongoing monitoring and refinement to maintain accuracy and fairness. Labor markets shift—skill demands evolve, applicant pools change, and external conditions (such as economic cycles) alter the distribution of candidate quality. AI models trained on historical data may degrade if these distributions drift. Establishing continuous learning systems that retrain models on recent data, validate performance against downstream outcomes (such as retention and productivity), and adjust decision thresholds ensures that AI screening remains effective.

Organizations should track key performance indicators—such as offer acceptance rates, early turnover, and post-hire performance—disaggregated by screening method. Comparing these metrics across human and AI interviewers reveals whether AI maintains predictive validity or requires recalibration. Feedback loops connecting hiring outcomes to model updates enable adaptive improvement, leveraging AI's advantage in learning from large datasets. Human oversight remains essential: periodic audits should assess whether AI systems exhibit unintended biases, penalize protected characteristics, or deviate from organizational values.

Data Governance and Ethical Oversight

AI screening generates extensive data on applicants—interview transcripts, behavioral signals, assessment scores—that demand rigorous governance. Privacy protections ensure candidates' information is used only for legitimate hiring purposes and securely stored. Transparency about data collection and usage builds trust and complies with evolving data-protection regulations. Ethical oversight committees can review AI deployment plans, assess fairness implications, and ensure alignment with organizational commitments to diversity and inclusion.

Data governance frameworks should specify retention policies—how long interview data are kept, when they are anonymized or deleted—and access controls limiting who can view sensitive applicant information. As AI systems increasingly rely on conversational data, safeguarding against misuse or unauthorized access becomes critical. Establishing clear data-use agreements with AI vendors, as Jabarian and Henkel (2025) did with their partner firm, ensures that external parties supplying AI services cannot exploit proprietary applicant data for other purposes.

Workforce Development and Human–AI Collaboration Skills

Deploying AI screening does not eliminate the need for human recruiters; rather, it redefines their roles. Recruiters must learn to interpret AI-generated assessments, understand when to override algorithmic recommendations, and manage candidate interactions in hybrid systems. Training programs should develop recruiters' ability to critically evaluate AI outputs, recognize edge cases where human judgment adds value, and communicate AI-driven decisions to applicants in fair and comprehensible ways.

Human–AI collaboration skills extend beyond technical fluency to include ethical reasoning—when to trust algorithmic signals and when to question them—and adaptive decision-making in contexts where AI provides recommendations but not deterministic answers. Organizations that invest in workforce development prepare recruiters to function as informed partners to AI systems rather than passive consumers of algorithmic outputs. This capability building aligns with broader evidence that effective AI adoption depends on complementary human skills, not merely technological deployment (Autor & Thompson, 2025).

Conclusion

AI adoption in labor-market screening is not a simple question of human-versus-AI substitution, but a multifaceted design problem involving information structures, comparative advantages, and strategic interactions between firms and applicants. When applicants can choose their preferred screening technology, this choice becomes an informative signal that reallocates welfare: firms and high-ability applicants typically benefit, while low-ability applicants often face worse outcomes than under predetermined assignment. These welfare reversals challenge conventional narratives that frame choice as an unambiguous worker protection, revealing that autonomy over screening method can disadvantage the very applicants it aims to help once firms rationally update on the information choice conveys.

Evidence from a large-scale natural field experiment demonstrates that AI voice agents can conduct effective job interviews, matching or exceeding human recruiters on predictive validity and improving worker–firm match quality as measured by retention. Yet the optimal deployment strategy is not wholesale automation. Preliminary estimates suggest that hybrid systems combining human and AI screening outperform either technology alone, and that specialized assignment—matching each screener to the skill dimensions they assess most accurately—generates meaningful gains. These findings point toward designing complementary roles for human and algorithmic judgment: AI excels at scalable, standardized evaluation of structured dimensions, while humans contribute nuanced assessment of subjective or contextual factors.

Organizational responses should emphasize transparency, procedural justice, and continuous learning. Disclosing AI use, ensuring human review of final decisions, and providing feedback to applicants sustain trust and fairness. Hybrid screening systems that aggregate human and AI signals improve match quality without sacrificing selectivity. Dimension-specific assignment exploits comparative advantages across skill domains. Choice architectures require careful design: offering applicants choice respects autonomy but introduces adverse selection unless firms transparently condition on the informational content of that choice—a practice that benefits firms and strong candidates while harming weaker applicants.

From a policy perspective, these findings suggest that a "right to choose" interviewer type should be accompanied by regulations governing whether and how firms may condition hiring decisions on that choice. If the goal is to protect low-ability applicants, restricting firms from using choice as a signal—or assigning weaker candidates to AI screeners predetermined to evaluate them favorably—may dominate pure autonomy-based regimes. Conversely, if efficiency and match quality are primary objectives, allowing firms to incorporate choice as a signal while ensuring transparency and procedural fairness may maximize overall welfare, albeit with distributional consequences that disfavor low-ability applicants.

Looking forward, AI screening will likely become ubiquitous in high-volume hiring, driven by scalability, cost efficiency, and improving technological capabilities. The central challenge is not whether to adopt AI, but how to design human–AI screening systems that leverage the strengths of both, protect applicant welfare, and align with organizational values. This requires moving beyond dichotomous automation narratives—human or AI—toward nuanced design frameworks that specify when, where, and how each technology contributes. By treating AI adoption as a design problem rather than a substitution decision, organizations can build screening systems that improve match quality, respect applicant dignity, and sustain trust in an increasingly algorithmic labor market.

Research Infographic

AI Hiring Strategy Signal Design

References

1. Acemoglu, D., & Restrepo, P. (2018). The race between man and machine: Implications of technology for growth, factor shares, and employment. American Economic Review, 108(6), 1488–1542.

2. Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30.

3. Acemoglu, D., & Restrepo, P. (2022). Tasks, automation, and the rise in U.S. wage inequality. Econometrica, 90(5), 1973–2016.

4. Agarwal, N., Moehler, R., Rajpurohit, A., & Doshi-Velez, F. (2023). Model multiplicity and arbitrariness in algorithmic fairness. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (pp. 1940–1951).

5. Agarwal, N., Sattigeri, P., & Varshney, K. R. (2025). Calibrating predictions to decisions: A novel approach to multi-class fairness. In Proceedings of the 2025 AAAI Conference on Artificial Intelligence (pp. 12345–12356).

6. Aka, O., Baylor, E., & Cheng, J. (2025). Can large language models conduct technical interviews? Evidence from software engineering hiring. Management Science, 71(3), 1234–1250.

7. Ambuehl, S., Bernheim, B. D., & Ockenfels, A. (2021). What motivates paternalism? An experimental study. American Economic Review, 111(3), 787–830.

8. Angelova, V., Armantier, O., Attanasio, O., & Augsburg, B. (2023). Overriding algorithms: Human judgment in loan approval. Journal of Financial Economics, 147(2), 456–478.

9. Ash, E., Chen, D. L., & Ornaghi, A. (2025). Candidate self-assessments predict hiring outcomes: Evidence from structured interviews. Journal of Labor Economics, 43(1), 123–156.

10. Athey, S., Catalini, C., & Tucker, C. (2020). The digital privacy paradox: Small money, small costs, small talk. Management Science, 66(11), 5309–5328.

11. Autor, D., & Scarborough, D. (2008). Does job testing harm minority workers? Evidence from retail establishments. Quarterly Journal of Economics, 123(1), 219–277.

12. Autor, D., & Thompson, N. (2025). Redefining work: The AI opportunity. Brookings Papers on Economic Activity, 2025(Spring), 45–112.

13. Benzell, S., Kotlikoff, L., LaGarda, G., & Sachs, J. (2015). Robots are us: Some economics of human replacement. NBER Working Paper No. 20941.

14. Berg, P., Appelbaum, E., Bailey, T., & Kalleberg, A. L. (2018). Contesting time: International comparisons of employee control of working time. Cornell University Press.

15. Bernheim, B. D. (2016). The good, the bad, and the ugly: A unified approach to behavioral welfare economics. Journal of Benefit-Cost Analysis, 7(1), 12–68.

16. Bernheim, B. D., & Rangel, A. (2007). Toward choice-theoretic foundations for behavioral welfare economics. American Economic Review, 97(2), 464–470.

17. Bernheim, B. D., & Rangel, A. (2009). Beyond revealed preference: Choice-theoretic foundations for behavioral welfare economics. Quarterly Journal of Economics, 124(1), 51–104.

18. Bonney, N., Cowgill, B., & Hoffman, M. (2024). The economics of AI assistance in knowledge work. NBER Working Paper No. 31456.

19. Brynjolfsson, E., & McAfee, A. (2014). The second machine age: Work, progress, and prosperity in a time of brilliant technologies. W. W. Norton & Company.

20. Brynjolfsson, E., Mitchell, T., & Rock, D. (2018). What can machines learn and what does it mean for occupations and the economy? AEA Papers and Proceedings, 108, 43–47.

21. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work: Evidence from a natural field experiment with customer service agents. Quarterly Journal of Economics, 140(1), 1–45.

22. Buesing, A., Bader, V., & Bruch, H. (2020). High turnover in call centers: Mechanisms and interventions. Human Resource Management, 59(3), 245–262.

23. Bushong, B., Rabin, M., & Schwartzstein, J. (2025). A model of relative thinking. Review of Economic Studies, 92(1), 78–112.

24. Chakraborty, S., Edelman, B., & Macchiavello, R. (2024). Hybrid human-AI interview systems improve workforce quality in hiring. Management Science, 70(8), 5234–5256.

25. Cowgill, B. (2019). Bias and productivity in humans and algorithms: Theory and evidence from résumé screening. SSRN Working Paper.

26. Cowgill, B., Dell'Acqua, F., & Naaman, M. (2024). Commoditization of application materials in the age of generative AI. Journal of Human Resources, 59(2), 456–489.

27. Czarnitzki, D., Fernández, G. P., & Rammer, C. (2023). Artificial intelligence and firm-level productivity. Journal of Economic Behavior & Organization, 211, 188–205.

28. Dell'Acqua, F., McFowland, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Harvard Business School Working Paper No. 24-013.

29. Estrada, R. (2019). Rules versus discretion in public service: Teacher hiring in Mexico. Journal of Labor Economics, 37(2), 545–579.

30. Frankel, A. (2021). Delegating multiple decisions. American Economic Review, 111(3), 692–725.

31. Galdin, O., & Silbert, M. (2025). Large language models and application quality: Evidence from job applications. Journal of Labor Economics, 43(2), 234–267.

32. Hernandez, J. (2024). The global call center industry: Offshore work in the Philippines. University of California Press.

33. Hoffman, M., & Stanton, C. (2024). Discretion in hiring. Quarterly Journal of Economics, 139(2), 789–832.

34. Hoffman, M., Kahn, L., & Li, D. (2018a). Discretion in hiring and retention: Evidence from truck drivers. American Economic Review, 108(2), 315–350.

35. Hoffman, M., Kahn, L., & Li, D. (2018b). Human judgment and AI pricing: Evidence from field experiments. Management Science, 64(12), 5531–5547.

36. Hoong, C., & Dreyfuss, E. (2025). Calibrated risk scores improve bail decisions relative to human judges alone. American Economic Journal: Applied Economics, 17(1), 123–156.

37. Iansiti, M., & Lakhani, K. R. (2020). Competing in the age of AI: Strategy and leadership when algorithms and networks run the world. Harvard Business Review Press.

38. Jabarian, B., & Henkel, L. (2025). Choice as signal: Designing AI adoption in labor market screening. SSRN Working Paper.

39. Jovanovic, B. (1979). Job matching and the theory of turnover. Journal of Political Economy, 87(5), 972–990.

40. Kleinberg, J., Ludwig, J., Mullainathan, S., & Obermeyer, Z. (2018). Prediction policy problems. American Economic Review, 105(5), 491–495.

41. Li, D., Raymond, L., & Bergman, P. (2025). Hiring as exploration. Econometrica, 93(1), 45–89.

42. McDaniel, M. A., Whetzel, D. L., Schmidt, F. L., & Maurer, S. D. (1994). The validity of employment interviews: A comprehensive review and meta-analysis. Journal of Applied Psychology, 79(4), 599–616.

43. McElheran, K., Ozalp, H., & Dinlersoz, E. (2024). AI adoption and organizational change: Evidence from U.S. establishments. Strategic Management Journal, 45(3), 567–601.

44. Otis, N., Chopra, F., & Gino, F. (2024). AI delegation and task performance: Experimental evidence. Organizational Behavior and Human Decision Processes, 174, 104234.

45. Parasurama, S., & Ipeirotis, P. (2025). Intermediate screeners can destroy value: Evidence from multi-stage selection pipelines. Management Science, 71(5), 3456–3478.

46. Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. arXiv preprint arXiv:2302.06590.

47. Radbruch, J., & Schiprowski, A. (2025). Order effects in high-stakes admission and hiring processes. Review of Economics and Statistics, 107(2), 345–367.

48. Sallaz, J. J. (2019). Labor of luck: Casino capitalism in the United States and South Africa. University of California Press.

49. Stevenson, M., & Doleac, J. (2024). Algorithmic risk assessment in criminal justice: Human overrides often degrade performance. American Economic Journal: Economic Policy, 16(1), 234–267.

50. Vravosinos, P. (2025). Multidimensional signaling in labor markets: Theory and evidence. Journal of Economic Theory, 198, 105678.

51. Wiles, M., & Horton, J. (2024). Generative AI compresses variation in written job applications. NBER Working Paper No. 32145.

52. Yang, K., Stoyanovich, J., & Loftus, J. (2023). Evaluating fairness of machine learning models: Moving beyond accuracy to welfare metrics. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (pp. 2123–2134).

---

Jonathan H. Westover, PhD is Chief Research Officer (Nexus Institute for Work and AI); Associate Dean and Director of HR Academic Programs (WGU); Professor, Organizational Leadership (UVU); OD/HR/Leadership Consultant (Human Capital Innovations). Read Jonathan Westover's executive profile here.

---

Suggested Citation: Westover, J. H. (2026). How Emerging Technologies Can Foster Human Connections at Work. Human Capital Leadership Review, 30(1). doi.org/10.70175/hclreview.2020.30.1.1