How Organizations Use ChatGPT Enterprise

Summary & Methodology Analysis

The researchers constructed a longitudinal dataset, an organization-week panel, by linking ChatGPT Enterprise adoption logs with Compustat financial records for the period spanning January 1, 2024, to March 31, 2026. This data enabled an analysis of how enterprise-scale deployment varies by firm size and R&D investment. To analyze functional usage, the authors used an automated classifier to map user messages into a 60-category task taxonomy, capturing the distribution of work tasks performed via the tool across various job roles and seniority levels.

The analysis tracks usage intensity defined as messages per active user. While the paper leverages models like gpt-5-mini and references Codex for context, the primary contribution focuses on empirical measurement of organizational behavior rather than model architecture. The study provides clear evidence of scaling, noting that aggregate output tokens produced by enterprise customers increased approximately sevenfold over the final nine months of the study period. Usage patterns are assessed across organizational hierarchies, normalizing job titles into broader classes and manager categories to identify who is driving adoption.

There are several limitations to these findings. The financial analysis is restricted to U.S.-based public companies, which excludes potential insights from startups and smaller private firms. Furthermore, the dataset captures only active users rather than the full workforce, and the job title information is noted as incomplete. Crucially, the researchers clarify that the current task classification system maps usage intent but does not measure actual productivity effects or bottom-line organizational outcomes, leaving the direct impact on firm performance outside the scope of the current analysis.