ChatGPT users are tackling tasks that belong to other professions on a large scale

Who does what at work when a chatbot is within reach? OpenAI examined that question in its report "Work at the Frontier: How AI is Expanding What People Do at Work."

OpenAI analyzed more than 800,000 work-related messages from American ChatGPT users. The messages were matched to the job titles users themselves reported on ChatGPT Business accounts.

Tasks across professions

According to the report, as relayed by CFOtech Asia, 16.8 percent of all work-related messages involved tasks that historically belong to a different profession. OpenAI calls it "task crossover."

If you count only the messages that are specific to a profession, the share is far higher: 43.5 percent involved tasks from another profession. General office tasks such as writing, summarizing and scheduling meetings are not included.

The differences between professional groups are large. The share of profession-specific requests outside the user's own field was:

- 77 percent for customer service and customer experience - 75 percent for design - 69 percent for HR - 56 percent for law - 53 percent for marketing - 40 percent for sales and for finance - 28 percent for engineers, the lowest

In teams of 2-5 people, the share of messages with tasks from other professions was 18.9 percent. In organizations with more than 100 employees, it was 16.3 percent.

Marketing and engineering tasks were the most transferable across roles. Financial calculations and technical troubleshooting were among the three most frequent tasks outside a user's own field.

Flexible boundaries

OpenAI chief economist Ronnie Chatterji says, according to Unite.AI: "Boundaries between jobs are likely already becoming more flexible due to AI."

OpenAI points out that repeated task transfer can expand roles without the job title changing, and that the point is to redesign workflows rather than to measure lost jobs. According to CFOtech Asia, the study sees usage data as an early window into changes that may come before formal changes to job descriptions.

Caveats

The study cannot determine whether AI makes it possible to take on new tasks or simply helps with tasks people are already responsible for. It also has no data on quality, speed or hiring decisions, and it measures what users ask about, not what value comes out of it.

The figures cover only American ChatGPT Business users and are OpenAI's own analysis. It is not clear from the sources whether the full report is open to independent review: Unite.AI writes that it had not yet been published, while CFOtech describes it as published.