AI use is moving beyond formal job boundaries

Workers are using artificial intelligence to handle activities that have traditionally belonged to other occupations, according to new research from OpenAI Economic Research. The company describes this pattern as “task crossover” and says some of these activities are becoming recurring parts of workers’ AI-assisted routines.

The findings build on OpenAI’s earlier Work at the Frontier research, which documented employees using AI for work beyond their typical occupational roles. The latest study examines what happens after workers try those activities, focusing on whether they return to them over time.

OpenAI analyzed more than 1.5 million work-related ChatGPT messages sent between April and July 2026. The company says the messages were aggregated and anonymized, and that researchers did not read individual user messages during the analysis.

Cross-occupation activity increased over four months

Among approximately 6,200 workers observed consistently throughout the study period, previously used cross-occupation tasks grew from 13.1% of occupation-specific AI activity in April to 25.9% in July. OpenAI says the pattern is consistent with workers incorporating some tasks outside their formal roles into ongoing workflows rather than trying them only once.

A separate comparison also found a higher likelihood of repeated use when workers had already used a cross-occupation task. In matched one-month follow-up observations, workers returned to a task used in the previous month 23.6% of the time. Comparable workers with no observed use of that task in the prior month used it 8.4% of the time.

OpenAI says similar differences appeared for tasks within a worker’s occupation and for general tasks. The results therefore point to recurrence as a broader feature of AI-supported work, while showing a notable gap between workers who have already incorporated a task and those who have not.

Workers prompt AI differently outside their roles

The research found differences in how workers ask ChatGPT for help depending on whether a task falls within their usual occupational domain. For activities outside their roles, workers wrote shorter prompts on average and were less likely to request explanations, how-to guidance, a specific response format or advice.

At the same time, they were more likely to provide examples or background information and to ask the system to check or verify something. OpenAI says one interpretation is that workers are borrowing expertise rather than attempting to learn an entirely new field from scratch.

Under that interpretation, a worker supplies a problem and relevant context, such as a document, example or information from a colleague, and uses AI to apply knowledge associated with another field. The pattern suggests that AI can help employees cross boundaries between specialties without requiring them to fully adopt a new professional identity.

Some tasks fit recurring workflows better than others

The likelihood that workers returned to a cross-occupation task varied significantly by the type of activity. OpenAI reported a 54% next-month return rate for discussing goods or services with customers. By comparison, workers returned to explaining financial information about 15% of the time.

Across the cross-occupation tasks studied, the average next-month return rate was 18.5%. OpenAI says the differences may reflect how naturally an activity fits into a recurring workflow. They may also be shaped by workplace norms, caution and the perceived consequences of making an error.

The figures do not establish that AI has permanently reassigned those responsibilities or that every repeated activity becomes part of a worker’s formal job. They show that some workers are repeatedly using AI for tasks associated with other occupations during the period observed.

Job responsibilities may broaden before titles change

OpenAI says the results suggest a possible path for AI-driven job transformation: a worker experiments with an activity outside a traditional role, finds AI useful for it and begins returning to that activity as part of regular work. Over time, the mix of activities in a job could broaden even if the job title remains unchanged.

For organizations deciding how to adopt AI, the company argues that access to tools is only part of the implementation challenge. The way tasks are divided and carried out may also determine whether employees can use AI to move from identifying a problem to advancing the work.

OpenAI says work design should therefore have a place alongside access to AI tools in organizational AI strategies. The company plans to continue studying how these shifts affect the division of labor, workers, businesses and the broader economy.

Sources

• OpenAI NewsHow workers are unlocking new ways of working