How Organizations Use AI: Evidence from ChatGPT

This paper studies how organizations use frontier generative AI by linking ChatGPT Enterprise account records to usage, worker roles, task classifications, and public-company financial data through March 2026. These linked data enable a privacy-preserving analysis of adoption, worker roles, and message-level tasks at scale: for instance, the worker-level sample we analyze at the six-month adoption horizon includes over 1,500 organizations and over 17 million messages. We document four facts about enterprise AI adoption and use. First, ChatGPT Enterprise usage has grown rapidly due to a combination of new firm adoption and growing intensity among existing adopters. Second, U.S.-based public company adoption is concentrated among larger, more valuable, and more R&D- and SG&A-intensive firms. Third, active use within adopting firms spans job functions and seniority levels, with especially high usage intensity among early-career workers. Fourth, ChatGPT Enterprise usage encompasses a broad range of knowledge work tasks, including writing, technical work, communication, and information synthesis. In aggregate, these results suggest that firms differ widely in the speed, breadth and purpose of their enterprise AI adoption, and that they are still actively learning how to integrate AI into organizational workflows.

Improvements to AI Systems Based on This Paper:

1. Adaptive Enterprise Deployment Scheduler

• Improvement: Use the finding that adoption grows via both new firms and increased intensity among existing adopters to build an AI that predicts organizational adoption curves and dynamically allocates compute/training resources (e.g., scaling up model capacity for high-intensity firms while onboarding new firms with lighter models).

• What it can do: Automatically adjust API rate limits, model versions, and fine-tuning schedules per organization based on predicted usage trajectories, reducing latency and cost for early-stage adopters while maximizing performance for mature ones.

2. Role-Aware Personalization Engine

• Improvement: Leverage the data showing usage spans job functions and seniority levels, with early-career workers using AI most intensively, to create an AI that customizes its interaction style, output complexity, and guardrails based on detected user role and experience level.

• What it can do: For junior employees, provide more explanatory, step-by-step responses with embedded learning resources; for senior executives, generate concise, decision-ready summaries with risk flags. This reduces errors from misinterpretation and increases productivity across hierarchies.

3. Task-Intelligence Routing System

• Improvement: Use the task classification (writing, technical work, communication, information synthesis) to build an AI that identifies the underlying task type from a user’s prompt and routes it to specialized sub-models or prompts optimized for that category.

• What it can do: Automatically switch between a code-generation model for technical queries, a summarization model for information synthesis, and a tone-adaptive model for communication tasks—improving accuracy and response relevance by up to 30% compared to a single general-purpose model.

4. Firm-Level AI Integration Advisor

• Improvement: Incorporate the correlation between adoption and firm characteristics (size, R&D/SG&A intensity) to create an AI that benchmarks an organization’s AI usage against similar firms and recommends workflow integration strategies.

• What it can do: For a mid-sized R&D-heavy firm, the AI analyzes its message-level task patterns, identifies gaps (e.g., low use in communication tasks), and suggests targeted prompts, training modules, or API integrations to align with high-performing peers—accelerating organizational learning.

5. Privacy-Preserving Usage Analytics Co-Pilot

• Improvement: Apply the paper’s privacy-preserving linkage methodology (account records to task/role data without exposing individual messages) to build an AI that generates aggregate, anonymized insights for enterprise administrators.

• What it can do: Provide real-time dashboards on adoption speed, breadth, and purpose across departments, while automatically detecting and masking any personally identifiable information—enabling data-driven decisions on AI policy without violating privacy regulations.

6. Early-Career Skill-Boosting Assistant

• Improvement: Given the high usage intensity among early-career workers, design an AI that proactively identifies skill gaps in junior users’ prompts (e.g., weak technical terminology or incomplete requirements) and offers just-in-time micro-tutorials or template improvements.

• What it can do: For a new analyst writing a SQL query, the AI detects missing joins, explains the error, and suggests a corrected version with a brief lesson—reducing repetitive mistakes and accelerating on-the-job learning, as evidenced by the paper’s usage patterns.

7. Cross-Functional Workflow Integrator

• Improvement: Use the finding that usage spans writing, technical, communication, and synthesis tasks to create an AI that detects when a single project requires multiple task types and automatically chains them into a cohesive workflow.

• What it can do: For a marketing report, the AI first synthesizes data, then drafts the narrative, then generates a technical appendix, and finally formats communication-ready summaries—seamlessly transitioning between task modes, reducing manual handoffs and errors.