OpenAI finds ChatGPT use outside a worker's own job keeps recurring

OpenAI's economic research group published a new study on how workers use ChatGPT for tasks that fall outside their own occupation, and whether that behavior sticks. The team analyzed more than 1.5 million work related ChatGPT messages sent between April and July 2026, building on an earlier report that had already documented what OpenAI calls task crossover: workers using AI for activities normally associated with a different job than their own.

The new question was whether that crossover is a one time experiment or something workers return to. Among roughly 6,200 workers tracked consistently across the four months, the share of a worker's occupation specific AI activity made up of previously used cross-occupation tasks rose from 13.1% in April to 25.9% in July. A separate matched-sample analysis looked at whether a worker who used a given cross-occupation task in one month used it again the next month: they did 23.6% of the time, versus 8.4% for comparable workers who had not used that task the previous month. OpenAI reports similar gaps for tasks inside a worker's own occupation and for general tasks.

How often workers returned to a task varied a lot by type. Among cross-occupation tasks, workers came back the next month to discussing goods or services with customers 54% of the time, to advertising or promotional writing 44% of the time, and to creating marketing materials 37% of the time. By contrast, they returned to explaining financial information only about 15% of the time. The average next-month return rate across all cross-occupation tasks was 18.5%.

OpenAI also found workers prompt AI differently depending on whether a task sits inside or outside their occupation. For outside-occupation tasks, prompts are shorter on average, and workers are less likely to ask for explanations, how-to guidance, a specific response format, or advice. They are, however, more likely to supply examples or background material, and more likely to ask AI to check or verify something. OpenAI's reading is that workers are using AI to borrow expertise: bringing a problem plus context from their own job, such as a document or a colleague's input, and asking AI to apply knowledge from a field that is not theirs.

OpenAI frames this as evidence for a possible path of AI driven job change that happens before job titles do: a worker tries an activity outside their traditional role, finds AI useful, and starts folding it into regular work. If enough of those activities become standing responsibilities, a job's actual content can broaden even while its title stays the same. OpenAI says it will keep studying these shifts to track how AI is changing the division of labor.

Key facts

OpenAI analyzed more than 1.5 million work-related ChatGPT messages from April through July 2026 to study cross-occupation AI use.
Among about 6,200 workers tracked across the period, previously used cross-occupation tasks grew from 13.1% to 25.9% of their occupation-specific AI activity.
Workers returned to a cross-occupation task used the prior month 23.6% of the time, versus 8.4% for workers with no prior use of that task.
Return rates varied by task: 54% for discussing goods or services with customers, 44% for advertising writing, 37% for marketing materials, but only 15% for explaining financial information.
OpenAI argues the pattern shows AI can broaden a job's actual responsibilities well before its official title changes.

Why it matters

This is OpenAI's own economic research team trying to document how AI actually changes work, rather than speculating about it. The core finding, that cross-occupation AI use is not a novelty but something workers build into recurring habits, is evidence for a specific mechanism of job change: responsibilities can expand gradually through repeated AI-assisted use of tasks outside a person's formal role, well before any job title or job description is updated.

Who it affects

The data covers ChatGPT Business users whose occupation is identified from role or department information given at onboarding, roughly 6,200 of them tracked consistently over four months, plus the wider 1.5 million message sample. The source does not name specific companies, industries, or individual workers, so the findings speak to workplace AI users in aggregate rather than to any named organization.

How to use it

For an employer or manager, the practical signal is that workers already use AI to reach past their job description, and that once a borrowed task proves useful they keep using it rather than dropping it. That argues for treating work design (how tasks are actually divided and assigned) as part of an AI rollout plan, not an afterthought to giving people access to the tools, since the return-rate gap by task type (54% for customer-facing tasks versus 15% for financial explanation) suggests some categories of borrowed work catch on far more easily than others.

How solid is it

The analysis is based on a large sample, over 1.5 million messages and a tracked cohort of about 6,200 workers over four months, which is a meaningfully longer and more systematic look than a single snapshot survey. It is still OpenAI's own internal research on its own product's usage logs, occupations are self-reported at onboarding rather than independently verified, and OpenAI itself frames some of the explanations for why return rates differ by task as speculative ("may reflect... could also reflect").

Risks and caveats

The study does not identify which companies, industries, or specific workers were involved, and it does not give the size, geography, or demographic makeup of the underlying sample beyond the roughly 6,200-worker figure. Because occupation labels come from self-reported onboarding data rather than verified job records, some of the observed crossover could reflect imprecise occupation categories rather than workers genuinely operating outside their real role. The claim that jobs could broaden while titles stay the same is presented by OpenAI as a possible interpretation, not a proven causal mechanism.

“AI may reshape jobs well before their titles change: workers use AI for activities associated with another occupation and return to some of them over time.”

— OpenAI, in the report