AI Adoption Across a Multinational Workforce: Sociotechnical Conditions for GenAI Acceptance in Human Resources
Citation. Ali, D., Rodríguez Velázquez, M. J., Ribeiro, M. H., Liao, V., & Papakyriakopoulos, O. (2026). AI Adoption Across a Multinational Workforce: Sociotechnical Conditions for GenAI Acceptance in Human Resources. arXiv. https://doi.org/10.48550/arXiv.2606.17887
Summary. This mixed-methods preprint examines a live transition from a legacy HR search system to a GenAI-supported system at one multinational technology company. Adoption was selective rather than a simple migration: employees chose among the new system, legacy search, source documents, colleagues, and HR according to situational fit, interaction skill, confidence in answers, and the underlying quality of organizational knowledge. The study argues that formal access and aggregate use can conceal unequal practical access, especially where device access, language, local work conditions, or organizational familiarity differ.
Key claims
- In a survey of 25 employees, 76% used the GenAI system daily or weekly, while 56% still used the legacy system daily or weekly; adoption operated as ongoing route selection across parallel information channels rather than complete replacement.
- Situational fit varied by device access, work environment, language, tenure, and familiarity with organizational terminology; the authors argue that aggregate usage can obscure exclusion when workers with the greatest friction are absent from the logs.
- Effective GenAI search required workers to phrase and rephrase questions, interpret summaries, find sources, and decide when to escalate; mean query length changed only from 1.87 tokens in legacy search to 2.11 tokens in GenAI search, while one- or two-token queries remained about 83% in both systems.
- Users calibrated reliance through source checking, comparison with legacy systems, and fallback to people or HR, but verification was uneven: 13 of 25 respondents said they never or rarely clicked sources in GenAI answers.
- The tool’s reliability depended on knowledge infrastructure—current and localized HR content, metadata, tagging, taxonomy, guidance, and escalation routes—not on the GenAI layer alone.
Evidence & limitations
- Sequential mixed-methods single-organization case: search logs from the legacy and GenAI HR systems, a descriptive survey (n=25), and ten semi-structured interviews with project-team members and end users.
- The survey was small and uneven across countries, roles, and experience; the authors treat it as descriptive and triangulate it with logs and interviews.
- Direct participants did not systematically include some groups central to the paper’s inclusion argument, including shop-floor, shift-based, and less digitally embedded workers; evidence about their fit is partly indirect.
- Public arXiv provenance was verified: the Drive PDF exactly matches the public arXiv PDF by SHA-256. Crossref exact-title search returned no clear published-version match; this remains an arXiv preprint.