On September 16, OpenAI published the second installment of Work at the Frontier, its research series on the labor market[1]. Drawing on more than 1.5 million work-related ChatGPT messages sent between April and July 2026, the report shows in numbers how often people who use AI for work outside their own occupation come back to that work later. It points to a future in which job titles stay the same while the range of work inside them quietly widens.

The First Report Found Task Crossover, This One Asks What Happens Next

The first installment, released in July, reported that 43.5 percent of occupation-specific work messages involved tasks historically associated with a different occupation[2]. Designers writing ad copy, salespeople troubleshooting software, and similar uses that cut across occupational lines.

The second installment asks the obvious follow-up question. Is crossing those lines a one-time experiment, or does it settle into the way people work? OpenAI tracked ChatGPT Business usage over four months and says the evidence leans toward the latter[1].

Prompts Look Different Outside a Worker's Own Field

One of the more interesting findings is that the same person asks AI for help differently depending on the kind of task. When asking about work outside their occupation, people write shorter prompts on average than they do for work inside it. They are also less likely to ask for explanations, how-to guidance, a specific response format, or advice[1].

What rises instead is the habit of supplying examples or background, and the habit of asking AI to check or verify something. OpenAI reads this as borrowing expertise rather than learning a new field: workers bring a problem and the relevant material, then ask AI to apply knowledge that belongs to another discipline[1]. Put another way, the question that used to go to a colleague now goes to the model.

Recurring Use Doubled in Four Months

Two sets of figures show how far this has settled in. The first covers roughly 6,200 workers observed consistently from April through July. Previously used cross-occupation tasks accounted for 13.1 percent of occupation-specific AI activity in April and grew to 25.9 percent by July[1].

The second is a one-month follow-up. Workers who had used AI for a cross-occupation task in the previous month returned to that same task 23.6 percent of the time, against 8.4 percent among comparable workers with no observed use of it the month before[1]. Similar gaps show up for within-occupation and general tasks.

Which Tasks People Come Back To

Return rates vary widely by task. The highest next-month figures were for discussing goods or services with customers at 54 percent, advertising or promotional writing at 44 percent, and creating marketing materials at 37 percent[1].

Explaining financial information, by contrast, drew a return rate of about 15 percent. Averaged across cross-occupation tasks, the next-month return rate was 18.5 percent[1]. OpenAI notes that these differences may reflect not only where AI fits naturally, but also workplace norms, caution, and how costly a mistake would be. Anyone who has hesitated before sending out a number has a feel for that gap.

How the Study Was Built and How Privacy Was Handled

The analysis covers sampled conversations from users whose occupation could be identified from the role or department information they provided during ChatGPT Business onboarding. Training-disabled data and messages without usable classifications were excluded[1].

OpenAI states that messages were aggregated and anonymized for analysis, and that researchers on the team never read individual user messages[1]. One caveat worth keeping in mind when reading the numbers: as workers try new tasks each month, the pool of tasks that can count as recurring also grows.

Work Changes Before the Org Chart Does

OpenAI concludes that how work is divided and carried out deserves a place alongside access to AI tools in any organization's AI strategy[1]. Someone tries an activity outside their usual role, finds AI useful for it, and keeps coming back. Repeat that often enough and the mix of activities inside a job broadens even though the title on the door has not moved.

The company has also published an AI Jobs Transition Framework, which classifies occupations by more than technical substitutability, weighing how far a human is still required and how demand might shift[3]. This latest study suggests the changes that framework anticipates are already visible in usage logs.

Summary

OpenAI's second report uses more than 1.5 million messages to argue that cross-occupation AI use is not a passing experiment but something being folded into daily routines. The share of such tasks rose from 13.1 percent to 25.9 percent over four months, and in customer conversations and promotional writing nearly half of workers return the following month. The content of a job is broadening well before the job posting is rewritten, which suggests the more urgent question is not how to hand out AI tools but where to draw the line on what gets handed off.

Source: https://openai.com/index/unlocking-new-ways-of-working/

Source: https://openai.com/index/how-ai-is-expanding-what-people-do-at-work/

Source: https://openai.com/index/modeling-ai-jobs-transition/