Resume-ing Control: (Mis)Perceptions of Agency Around GenAI Use in Recruiting Workflows

1. Problem Statement

The paper addresses a critical gap in the understanding of human agency within high-stakes decision-making processes, specifically in recruitment. While policy and academic discourse advocate for "AI-assisted" rather than "AI-replaced" decision-making, there is limited empirical evidence on how professionals actually perceive and experience their control when integrating Generative AI (GenAI) into their workflows.

The core problem identified is a misalignment between perceived autonomy and actual influence:

• Recruiters believe they retain final authority, viewing AI as a tool for low-level tasks.
• In reality, GenAI acts as an "invisible architect," structuring the decision space (defining job criteria, creating rubrics, summarizing data) before human judgment is even applied.
• There is a lack of understanding regarding whether AI adoption is a self-directed choice or a systemic necessity driven by organizational mandates and an "AI arms race" between recruiters and candidates.

2. Methodology

The study employs a mixed-methods approach combining qualitative interviews with a functional analysis of tools.

• Participants: 22 recruiting professionals across eight industry sectors (e.g., technology, healthcare, manufacturing) in the US.
• Data Collection:
  • Semi-structured Interviews: Conducted over Zoom (Sept–Dec 2025), lasting ~60 minutes. The protocol utilized the Critical Incident Technique to reconstruct recent hiring events and minimize recall bias.
  • Normalization Framing: To mitigate social desirability bias, the interviewers used framing techniques to encourage candid disclosure about unofficial or "shadow" AI use.
• Analysis:
  • Qualitative: Reflexive thematic analysis using Atlas.ti. The researchers used a hybrid coding approach (inductive and deductive), refining a codebook from 81 initial codes to 54 final codes. Inter-rater reliability (Krippendorff's Alpha) was 0.80.
  • Functional Analysis: A systematic categorization of the GenAI tools mentioned by participants (Appendix A) based on recruitment pipeline stage and functionality.
• Ethics: The study was IRB-exempt. Participants were anonymized, and data was de-identified.

3. Key Contributions

The paper makes three primary contributions to the fields of Human-Computer Interaction (HCI) and Algorithmic Decision-Making:

1. The "Invisible Architect" Concept: It redefines the role of GenAI in hiring not as a decision-maker, but as a structural force that defines the framework of decisions (job descriptions, interview questions, rubrics) and processes the input data (resume summaries, interview transcripts). This influence occurs subtly, often before the human recruiter realizes a decision is being shaped.
2. Systemic Loss of Agency: It challenges the notion of "choice" in AI adoption, demonstrating that recruiters often lack agency regarding whether to use AI due to top-down corporate mandates, efficiency pressures, and the need to counteract candidate-side AI manipulation.
3. The "Boiling the Ocean" Effect: It identifies a paradox where massive investment in GenAI yields marginal efficiency gains while introducing significant downstream costs, including workforce deskilling, increased bias, and higher employee turnover due to poor candidate matching.

4. Key Results & Findings

RQ1: Perceptions of Agency

• Perceived Control: Recruiters overwhelmingly assert they make the final decisions and view AI as a "filter" or "helper" for low-level tasks.
• Reality of Influence: Despite these claims, GenAI dictates the decision space. For example, AI generates the interview rubrics and summarizes candidate performance. Recruiters often defer to these outputs without verification (e.g., "blindly" moving AI-approved candidates to the next stage), effectively ceding control to the algorithm's framing.

RQ2: Adoption as Choice vs. Necessity

• Forced Adoption: Adoption is rarely a voluntary optimization choice. It is driven by:
  • Top-down Mandates: Corporate goals requiring AI integration for productivity.
  • The AI Arms Race: As candidates use AI to create "perfect" resumes, recruiters feel compelled to use AI to filter them out, creating a feedback loop of automation.
  • Existential Pressure: Fear of obsolescence ("if you're not using it, you're gonna die").

RQ3: Implications of Decision-Making Changes

• Deskilling: There is a trend of workforce deskilling, particularly among junior recruiters. Reliance on AI for research, note-taking, and rubric creation erodes domain expertise and the ability to make deliberate, reasoned decisions. Junior staff express fear of being unable to perform tasks without AI.
• Retreat to "Gut Instinct" & Bias: Faced with an overload of AI-polished applications that lack signal, recruiters are retreating to "gut checks" and "vibe checks" (culture fit) to reclaim agency. This reintroduces human bias (which AI was supposed to mitigate) and relies on subjective assessments that are harder to audit.
• Low Return on Investment (ROI): Despite claims of efficiency, participants report:
  • Hidden Labor: Significant time spent prompting, verifying, and correcting AI outputs.
  • Quality Degradation: Higher turnover rates because AI-matched candidates often lack genuine fit or skills.
  • The "Boiling the Ocean": The industry is investing exhaustive effort for marginal gains, resulting in a net increase in workload and a decrease in hiring quality.

5. Significance and Implications

The paper offers critical insights for policy, organizational practice, and future research:

• Regulatory Gaps: Current regulations (e.g., GDPR, Title VII) focus on the final output of automated decisions. However, because GenAI structures the entire decision infrastructure (rubrics, data processing), current legal frameworks may fail to capture how agency is altered. The "human in the loop" may be an illusion if the loop's parameters are set by AI.
• Organizational Policy: Organizations must move beyond "prompt engineering" training. They need to implement intentional friction (e.g., requiring recruiters to articulate their logic before seeing AI summaries) to prevent deskilling and maintain critical oversight.
• Research Directions: There is an urgent need to develop tools to audit intermediate model outputs (e.g., generated interview questions) rather than just final recommendations. Furthermore, research must address how to maintain human awareness of AI influence as it becomes increasingly subtle and intertwined with human cognition.

Conclusion: The paper concludes that while GenAI promises efficiency, it currently functions as an invisible architect that subtly erodes human agency, leads to professional deskilling, and fails to deliver on its promise of better hiring outcomes, creating a "boiling the ocean" scenario for the recruitment industry.