The second installment of OpenAI’s economic research arrived on September 16 with a finding that reframes what AI is doing inside companies. The researchers analyzed more than 1.5 million work-related ChatGPT messages from registered users in the United States between April and July, and found that employees are increasingly using the tool for tasks outside their own job descriptions.
The shift is measurable and steep. Among roughly 6,200 employees who took part in the study on an ongoing basis, tasks beyond their core occupation accounted for 13.1 percent of AI-assisted work in April. By July the share had climbed to 25.9 percent, nearly doubling over a single quarter. The researchers flagged the trend as one of the report’s central observations.
What employees ask the model to do changed along with what they were working on. When handling tasks outside their own field, workers were less likely to demand explanations, operating instructions, or specific formatting from the AI. They were more likely to supply background information and then check the output. The pattern suggests employees were borrowing the AI’s expertise to complete unfamiliar work rather than using it to speed up work they already understood.
The persistence of use varied sharply by task type. Customer communication, and advertising and promotional copywriting, returned at rates of 54 percent and 44 percent respectively, meaning workers came back to the AI for those jobs repeatedly. Legal research came back at just 10 percent, a sign that some tasks resist the delegated approach, whether because of risk, complexity, or a lack of trust in the output.
The report is the second in a series the company calls Frontier Work, an attempt to document how AI is changing employment at the level of individual tasks rather than headlines about jobs created or eliminated. The first report established a baseline; this one tracks the drift toward cross-occupation work.
The implication the authors draw is that job responsibilities may be widening without job titles changing. If cross-occupation tasks become a permanent part of an employee’s routine, the argument goes, the boundaries of a role expand even when the org chart does not move. A marketer who now drafts basic legal summaries, or an analyst who writes customer-facing copy, is quietly doing a different job than the one she was hired for.
The data is drawn from ChatGPT users who consented to have their messages studied, and the company has been careful to describe the findings as descriptive rather than predictive. Still, the direction is consistent with the broader argument OpenAI and its peers have been making: that AI changes the composition of work before it changes the count of workers.
Economists outside the company have long argued that technology tends to do exactly this, expanding the reach of individual workers rather than simply replacing them. The difference now is the speed with which the expansion is happening, and the fact that the instrument doing the expanding is a single general-purpose tool that any employee can open in a browser.
The study’s methodology gives the finding its weight. The messages came from ChatGPT users who had agreed to have their conversations examined, and the researchers focused on a core of roughly 6,200 workers who participated continuously across the four-month window. Following the same people over time is what allows the report to describe a shift in how they work, rather than a snapshot of what the tool was used for in a given week.
The term cross-occupation is precise in the report’s framing. A task counts as cross-occupation when it falls outside the occupation the worker reports as her own, whether that is a marketer drafting legal summaries or an engineer writing customer-facing copy. The rising share means workers are reaching outside their job descriptions more often, using the model as a generalist colleague rather than a specialist for their own field.
The finding lands inside a longer debate about technology and work. Economists have argued for decades that new tools expand the reach of individual workers before they change the count of them, and that the first measurable effect is often workers doing more kinds of tasks. The report is an attempt to measure that expansion directly, using the log of what people actually asked the model to do.
The report’s authors stopped short of forecasting where the trend goes from here. The jump from 13 percent to 26 percent in a single quarter could be a one-time adjustment as users discovered new capabilities, or the leading edge of a longer climb. OpenAI offered no conclusion on the point, leaving the question of how far cross-occupation work will spread to the next installment of the series.
The report does not forecast how far the share of cross-occupation tasks will rise. The jump from 13 percent to 26 percent in three months could be a one-time adjustment as users discovered new capabilities, or the start of a longer run. OpenAI offered no conclusion on that point, leaving the question of where the line stops for the next installment.
