OpenAI Releases First Enterprise AI Usage Data: Large Companies Lead, New Employees Most Active

In mid-August, OpenAI released two reports: the working paper 'How Organizations Use AI' and the companion report 'Enterprise Signals', marking the first time real usage data answers 'who uses AI in enterprises and for what'. The paper matches ChatGPT Enterprise account records with employee job titles, tasks, and public financial data of listed companies. The sample covers over 1,500 organizations and 17 million messages; task classification covers 973 organizations and nearly 8.7 million messages, matched with financial data from 417 U.S. listed companies. All results are anonymized and aggregated; researchers did not manually review any enterprise messages.

Comparing U.S. listed companies that adopted versus those that did not adopt ChatGPT Enterprise, the paper finds adopters have median revenue of $2.275 billion versus $209.6 million for non-adopters, a roughly 10x difference; total assets are $4.394 billion versus $668 million; and employee counts are 2,934 versus 424. Even after controlling for industry and size, companies with higher revenue per employee are more likely to adopt, while those with higher fixed assets per employee are less likely, indicating that 'asset-light, knowledge-heavy' companies lead adoption.

Scale effects are significant: companies in the top 25% of revenue have a 6.9 percentage point higher adoption probability, top 5% have 9.8 points higher, and top 5% within industry have 11.3 points higher. Additionally, companies with higher accumulated selling, general, and administrative expenses, R&D, and software spending per employee are more likely to adopt, confirming that AI requires 'supporting capabilities' to be effectively utilized.

The paper breaks down active users by job function and seniority six months after adoption. Engineering and technology professionals account for about 11% of weekly active users, executives about 9%, and finance and marketing about 5% each, with usage spanning all functions. By seniority, managers and directors account for about 24%, general employees about 15%, senior contributors about 14%, executives about 10%, and newcomers about 7%.

The key is intensity: relative to the average active user, entry-level employees and interns send 8 to 9 more messages per week than the average, while executives send significantly fewer; analysts and marketing/communications roles are also heavy users. This pattern holds within the same company, indicating that younger employees are the most frequent users in any AI-adopting organization.

The paper categorizes enterprise messages into 60 task types, finding extremely broad usage: over half of active users use AI for document and technical writing, nearly half for technical and digital work, with many others in communication, research, planning, and data analysis. Task focus varies by industry: financial and insurance companies have more finance and tax tasks, while entertainment and retail have more sales and marketing tasks.

The companion report shows that by June 2026, Codex accounted for 64% of total output tokens from enterprise customers, indicating agents are taking on substantive tasks. Since February, weekly active Codex users in legal, sales, recruiting, and marketing departments grew 108x, 41x, 41x, and 26x respectively, while engineering grew only 5x.

The gap between frontier enterprises (top 10% by monthly output tokens per user) and typical enterprises (middle 10%) widened from 2.6x in January 2026 to 8.3x in June, reaching 11.7x in the information technology sector. Frontier enterprises have 21% of active users using plugins, compared to 9% for typical enterprises, and 95% internally at OpenAI. The difference lies in whether AI is integrated into company tools, data, and processes.

The paper notes that AI adoption shows a 'deeper with use' pattern, but the real differentiator is not 'knowing how to use AI' but 'being able to get AI to complete a checkable, deliverable piece of work'. If large, intangible-asset-rich companies adopt and use AI deeply first, AI diffusion may entrench or even amplify differences between companies.

For managers, training and incentive targets need to be redefined; the ones who truly need productivity gains may be the young employees who are already wrestling with AI daily. The technology barrier is rapidly disappearing, while the organizational capability barrier is just emerging. The key is answering 'where to use it, how to use it, and who uses it'.