Workers who had used AI for a task from another occupation went back to it the next month 23.6% of the time, against 8.4% for comparable workers not seen using it. OpenAI says that if such borrowed tasks become regular duties, jobs could widen while their titles stay the same.
The finding comes from OpenAI’s Economic Research group and its second Work at the Frontier report. The first documented what the company calls task crossover, in which people use AI for work that has historically belonged to a different occupation; the new one asks which of those borrowed activities become part of workers’ routines.
The analysis draws on more than 1.5 million work-related ChatGPT messages from April through July 2026, with about 6,200 workers tracked consistently across those months. Among their uses of AI for occupation-specific tasks, repeat use for tasks from other occupations rose from 13.1% in April to 25.9% in July. OpenAI interprets that increase as workers making those activities part of their routines. However, each month also adds newly tried tasks to the pool that can count as repeat use in later months.
The workers’ occupations come from the role or department they entered during ChatGPT Business onboarding; the messages are sampled from those users’ conversations, excluding training-disabled data and messages that could not be classified, and were pooled and anonymized, with researchers never reading individual messages.
The return-rate comparison comes from matched observations, each revisited a month later: workers seen using a cross-occupation task in the prior month were measured against comparable workers with no such use recorded. Tasks within an occupation and general tasks show gaps of a similar size.
Return rates also varied widely by task. Workers went back to discussing goods or services with customers 54% of the time; for explaining financial information the figure was about 15%. OpenAI suggests the spread may show where AI fits naturally into repeated work, or may reflect workplace norms, caution, or how serious workers believe a mistake would be.
From the repeat use it found, OpenAI sketches one route by which AI could reshape jobs: a worker tries an activity outside their role, finds that AI helps with it, and keeps returning to it. If those borrowed tasks turn into regular duties, a job could take in a wider mix of work while keeping its title.