Intro / Hook
Companies aren't just buying ChatGPT and leaving it sitting there. OpenAI's new research found that ChatGPT Enterprise output grew roughly sevenfold between June 2025 and March 2026. And this wasn't simply because more companies signed up. Among firms that had already adopted by June 2025, output still grew roughly fourfold. Something is happening after AI enters the workplace: people keep finding more reasons to use it. But the really interesting part isn't the growth number. It's who is using ChatGPT most intensively, what they're actually using it for, and which companies were first through the door. Because the answers don't look much like the simple story of executives buying an AI tool to automate a few tasks. They suggest something broader — AI gradually working its way through the organisation.
What Happened
The study is called 'How Organizations Use AI: Evidence from ChatGPT', and importantly, it isn't based mainly on employees being asked how they think they use AI. The researchers analyse administrative data from ChatGPT Enterprise, OpenAI's centrally administered workplace product. The underlying panel covers organisations adopting Enterprise between January 2024 and March 2026. The worker-characteristics sample contains 1,764 organisations and more than 17.4 million messages, while a separate task-classification sample covers 973 organisations and about 8.7 million classified messages. Individual enterprise messages weren't manually read by researchers. Their content was classified using automated systems and the results were reported in aggregate. That gives us a different view of workplace AI: not simply corporate AI strategies or survey responses, but patterns in how the product is actually being used.
Why It Matters
This is why the strongest finding isn't one flashy ChatGPT use case. It's the absence of one. More than half of active users in the task sample perform documentation or technical-writing tasks, nearly half perform technical digital work, and substantial numbers use the system for communication, information, research, planning, sales and marketing, data analysis, legal and regulatory work, and finance. Different industries and job roles naturally lean towards different applications, but writing, technical work and communication appear across much of the organisation. That makes generative AI look less like a specialised software tool and more like a general layer sitting underneath knowledge work. But the paper ends on an important warning: rapid adoption should not be confused with immediate productivity transformation. Companies are still learning where AI belongs. And in the long run, that learning process — not the number of corporate ChatGPT accounts — may be the part that actually changes how businesses work.
What the Details Show
Here's where the picture gets more surprising. Six months after organisational adoption, active ChatGPT Enterprise users appear across the hierarchy. At the average firm, managers and directors account for about 24% of observed weekly active users, individual contributors and professionals around 15%, senior individual contributors and principals around 14%, executives around 10%, and early-career workers and trainees around 7%. But those percentages don't tell us who is using it most intensively. Among people who are active, early-career workers and trainees send roughly eight to nine more messages per week than the average active user inside the same company. Managers, directors and executives send fewer. Analysts and marketing and communications workers also show relatively high message intensity. That doesn't mean junior workers receive more economic value from AI — the researchers explicitly warn that message volume isn't a measure of productivity. But it does tell us where some of the deepest day-to-day interaction with the technology is happening.
Reading Between the Lines
That distinction between buying AI and actually absorbing it into a company may be the most important idea in the whole paper. The researchers argue that adoption is only the beginning. Organisations still have to discover useful applications, encourage employees to use them and integrate the technology into existing workflows. And the companies adopting earliest aren't a random cross-section of business. Among the U.S. public companies studied, ChatGPT Enterprise adopters were generally larger, more valuable and more heavily invested in organisational capabilities and R&D before adoption. That raises an uncomfortable possibility. Making powerful AI widely available doesn't necessarily mean every company benefits equally. The model might be available to everyone, but the ability to reorganise work around it may become an advantage in its own right.
What We Do Not Know
And there are some important limits to these numbers. The study measures activity inside ChatGPT Enterprise. It doesn't capture employees using personal ChatGPT accounts, other AI providers, API-based products or internally developed AI systems. The worker data also don't reveal the complete workforce composition of every organisation, so saying that a percentage of active users belong to a particular role isn't the same as measuring the adoption rate of that profession. Most importantly, this study measures usage, not outcomes. It doesn't tell us whether sending more messages makes someone more productive, whether a generated document saved an hour or created more work, whether a company redesigned jobs because of AI, or whether ChatGPT ultimately increased profits. And the relationship between adoption and financially stronger firms is an association — not evidence that ChatGPT caused those firms to perform better.
What Happens Next
The next stage of this story isn't simply whether more companies buy access to generative AI. It's whether all this individual activity eventually becomes something more durable: redesigned workflows, organisational knowledge and measurable improvements in output. The paper treats generative AI as a potential general-purpose technology, and technologies like that rarely transform an economy the moment they're installed. Companies have to discover where the technology works, build processes around it and sometimes reorganise the way work gets done. The rapid growth in ChatGPT Enterprise usage shows that use is deepening quickly. But the harder question is which organisations actually learn to convert that activity into value. If companies with stronger existing capabilities integrate AI more effectively, the first major economic effect of workplace AI might not be a universal productivity boom. It could be a growing gap between companies that merely have access to AI and companies that learn how to use it well.