OpenAI Finds Enterprise AI Use Is Deepening, but Adoption and Revenue Gains Remain Uneven

Enterprise use of ChatGPT has expanded rapidly across firms, workers, and business functions, according to OpenAI’s analysis of ChatGPT Enterprise records through March 2026. The study links usage data with worker roles, task classifications, and public-company financial information, covering more than 1,500 organizations and 17 million messages at the six-month adoption point.

Adoption is spreading beyond isolated experiments into recurring work. Employees use AI for writing, research, coding, analysis, customer-facing tasks, and internal knowledge work, while the broader ChatGPT user base still spends roughly 70% to 75% of its interactions on non-work topics. That split puts enterprise growth in context: workplace usage represents a smaller share of total activity, but it is becoming more structured and operational.

The leading 10% of companies by monthly AI usage now generate 8.3 times more output tokens per active user than typical firms. OpenAI’s report places that ratio at 2.6 times in January, showing a sharp increase in the depth of use over 5 months.

The gap reflects more than employee enthusiasm. Frontier companies adopt tools that connect models to company context, software, and repeatable workflows at higher rates. Plugin usage reached 21% among frontier firms, compared with 9% among other companies, while Skills adoption stood at 19% versus 3%. OpenAI’s internal plugin adoption rate, cited as 95%, provides a benchmark for how much further enterprise customers can go. The pattern indicates that the competitive divide is forming around implementation, not simply access to a chatbot.

Enterprise AI is moving from assistance toward execution, with agents handling parts of real business processes rather than only drafting responses or answering questions. Codex accounted for 64% of output tokens from enterprise customers in June, according to the corporate adoption analysis, making software-related work a major driver of advanced usage.

Codex weekly active users rose 108 times in legal, 41 times in sales, 41 times in recruiting, and 26 times in marketing. Those increases show how agentic tools are spreading into departments that traditionally relied on document review, prospect research, candidate screening, and campaign production. The change also raises the bar for governance: organizations must manage permissions, sensitive data, review requirements, and accountability when AI systems operate inside workflows.

The research does not establish a direct link between heavier ChatGPT use and higher revenue per employee. Fortune highlighted that finding as the most consequential detail in the enterprise study, because adoption depth has advanced faster than clear proof of financial returns.

Usage metrics still reveal meaningful organizational differences, but output tokens measure activity rather than business value. A company can generate more model output through productive automation, duplicated work, or poorly controlled experimentation. Revenue per employee also reflects pricing, industry mix, acquisitions, staffing, and broader economic conditions. Enterprise leaders therefore need measures tied to cycle time, error rates, throughput, customer outcomes, and cost rather than treating token volume as a return-on-investment score.

The spread of AI into routine workflows makes governance a day-to-day operating issue. Centific points to decision support, content refinement, multilingual work, compliance, and global deployment as areas where organizations need clear controls around quality, privacy, and accountability.

The organizational divide reinforces that need. Firms that connect AI to internal context and tools gain more capability, but those connections also create larger exposure if permissions, retention rules, or human review are poorly designed. Multilingual deployment adds another layer, including uneven performance across languages and regions. Companies adopting Codex or workflow-connected agents need documented approval boundaries, audit trails, escalation paths, and tests for sensitive use cases before scaling.

The next phase will be defined by whether companies convert broad access into repeatable, measurable work. OpenAI’s findings show a widening separation between firms experimenting with AI and firms embedding it in software, departmental processes, and internal knowledge systems.

That separation creates pressure on companies to identify high-value workflows, train employees, and connect approved tools to reliable business data. It also gives vendors a reason to compete on agent controls, integrations, analytics, and security rather than model access alone. The evidence remains early, but the direction is clear: enterprise AI adoption is becoming deeper and more uneven, while financial results still require careful measurement.