Employees disclose generative-AI use through three different, equally sufficient pathways of personal attitudes, workplace norms, task features and tool characteristics; disclosure is driven by configurations of factors, not a single cause, so measurement and transparency policies must account for contextual trade-offs.

Main Finding

Using data from 1,011 U.S. employees and a configurational analytic approach, the study finds that there are three distinct, consistently sufficient combinations of individual-, workplace-, task-, and tool-related factors that lead employees to disclose their use of generative AI at work. These configurations represent equifinal pathways—different routes that are each sufficient for disclosure—and reveal complementarity among factors and trade-offs that employees weigh when deciding whether to disclose AI use.

Key Points

• Sample and scope: Employees in the United States (N = 1,011); focus on disclosure of generative-AI use in the workplace.
• Analytic approach: A configurational method (identifying combinations of conditions rather than isolating single-variable net effects) was used to uncover patterns that predict disclosure.
• Result structure: Three alternative configurations were identified; each configuration is a different sufficient pathway to disclosure (equifinality).
• Multilevel factors: Disclosure is explained by configurations spanning four domains — individual-level, workplace-level, task-related, and tool-related factors — rather than by any single factor alone.
• Complementarity and trade-offs: The findings highlight complementarity effects (certain factors reinforce each other) and show that employees trade off competing considerations when deciding to disclose AI use.
• Practical consequence: The observed patterns provide actionable insights for designing workplace AI-transparency policies in contexts of growing human–AI interaction.

Data & Methods

• Data: Survey/empirical data from 1,011 U.S.-based employees (reported sample size).
• Methodology: Configurational analysis (identification of sufficient configurations). This approach focuses on how specific combinations of conditions jointly produce the outcome (disclosure), and accommodates equifinality—multiple, different combinations can each be sufficient.
• Conditions considered: Factors at four levels were incorporated: individual (e.g., employee attitudes/knowledge), workplace (e.g., norms, policies), task (e.g., task characteristics), and tool (e.g., properties of the AI tool). (The study emphasizes patterns across these domains rather than isolated main effects.)

Implications for AI Economics

• Measurement and inference: Economists studying AI adoption should distinguish between actual use and disclosed use—disclosure is endogenous and driven by multifactored configurations, so surveys or administrative counts based on disclosure may mis-measure true AI use if disclosure incentives vary across contexts.
• Heterogeneous adoption effects: Because disclosure follows multiple sufficient pathways, the labor-market effects of AI (productivity, task reallocation, wage impacts) likely vary across workplaces and workers depending on the configuration of incentives, norms, and task characteristics.
• Policy and regulation design: Effective transparency rules must account for complementarity across domains (tool design, workplace policy, task assignment) and the trade-offs employees face (e.g., reputational concerns, accountability, productivity gains). One-size-fits-all disclosure mandates may have uneven effects or unintended consequences.
• Organizational investment decisions: Employers designing AI governance—training, disclosure protocols, auditing—should consider that different mixes of organizational policies, task redesign, and tool features can each produce disclosure, so interventions can be tailored to achieve transparency goals with fewer adverse trade-offs.
• Future research: Economic models of AI diffusion and labor market adjustment should incorporate configurational heterogeneity and the possibility that disclosure behavior is shaped by multi-dimensional complementarities rather than single-factor drivers.