Some people try AI for a task outside their usual job responsibilities, then use it for the same task again the next month. A study by OpenAI of more than 1.5 million work-related ChatGPT messages from April through July 2026 observed this pattern of repeat use. It shows that AI is being used not only to help with familiar work but also to attempt tasks that have typically fallen outside a person’s role. Source
This shift matters to workers who need to handle tasks outside their area of expertise and to organizations examining how employees use AI. The key takeaway is not that using AI for out-of-role tasks improves performance. Rather, some users returned to those tasks, and the information and requests they gave AI differed from those for tasks within their roles. Source
What changed: From trying a task to repeating it
Among roughly 6,200 people whose AI activity by occupation was observed continuously from April through July, the share of out-of-role tasks they had used AI for before rose from 13.1% in April to 25.9% in July. The study describes this as consistent with the possibility that some tasks became part of an ongoing workflow rather than remaining one-off experiments. It does not mean every out-of-role task became routine work, but instances of use that did not end after a single attempt increased. Source
A separate look at a sample one month later also found a difference. Workers who had used AI for an out-of-role task in the previous month used it for the same task again at a rate of 23.6%. Among comparison workers with no observed use for that task in the previous month, 8.4% used it. These figures help illustrate the pattern of returning to a task someone has tried before, but they are not evidence that AI improved workers’ abilities. Source
Where should you start with your own work?
First, when thinking about tasks you have recently used AI for, distinguish between your usual responsibilities and work outside your role. The practical implication of this study is not to increase out-of-role tasks indiscriminately. It is to examine which tasks you used AI for, what background and examples you provided, and what you asked it to review. Organizations can look not only at whether employees have access to tools but also at how these requests carry through into actual workflows. Source
Next, distinguish between tasks tried once and tasks revisited the following month. In the study, the next-month repeat-use rate was 54% for discussing products and services with customers, 44% for writing advertising or promotional copy, and 37% for creating marketing materials. By contrast, the rate was about 15% for explaining financial information, while the average across out-of-role tasks was 18.5%. With such large differences between tasks, frequent use in another field is not enough to conclude that the same approach will take hold in your work. Source
How did requests differ?
In the study, workers entered shorter prompts on average when using AI for out-of-role tasks than for tasks within their roles. They were relatively less likely to ask for explanations, procedural guidance, a specific format, or advice. Instead, they were more likely to provide examples or background information and ask AI to review or check a result. The researchers interpreted this as a possible way of drawing on knowledge from another field and applying it to a current problem. Source
So if you try an out-of-role task, the usage pattern observed in the study suggests looking beyond what you want AI to do: consider what context and examples you provide and what you want it to check. This is a suggested application of an observational finding, however. The study did not establish that shorter prompts or any particular way of providing examples produces better results. Source
For organizations, look at workflows as well as tools
Giving employees access to AI alone makes it difficult to know which tasks become part of actual work. Organizations can examine the background and examples employees provide for out-of-role tasks, what they ask AI to review, and which tasks recur the following month. The study suggests that reducing friction between identifying a problem and making progress on it with AI may matter. It does not, however, provide a proven implementation process. Source
What to keep in mind when interpreting the numbers
These results do not represent AI use by all workers. The study classified occupations using role and department information provided during ChatGPT Business onboarding and analyzed a subset of conversations that could be classified. The researchers said they did not read individual messages, instead analyzing aggregated, anonymized data. They excluded data for which training use was disabled and messages without classification information. The repeat use observed here should therefore not be generalized to the entire labor market or taken as proof that a particular way of using AI improves outcomes. Source
The starting point for your own assessment is simple: if you have used AI for a task outside your role, distinguish whether it was a one-time attempt or something you returned to, what information you provided, and what you checked. Repeat use is a signal worth noticing, but it cannot, on its own, show that the task was done better. Source