Much of the research into AI and work asks which tasks within an existing job a model can perform. OpenAI’s latest study examines a different question: whether workers are using ChatGPT to attempt work historically associated with another occupation.
Tasks associated with another occupation accounted for almost 17% of more than 800,000 work-related messages in OpenAI’s analysis of U.S. ChatGPT users. After broadly shared activities such as writing, summarizing and scheduling were removed, cross-occupation work represented over 40% of occupation-specific messages. OpenAI calls the pattern “task crossover.”
OpenAI used ChatGPT to classify each selected message into one primary O*NET detailed work activity, using up to nine earlier messages from the conversation as context. The resulting label represented the message’s main activity rather than every task it might contain.
The finding points to a possible change in how work is divided, rather than evidence that jobs or specialist roles are disappearing. The analysis records what users asked ChatGPT to help with. It does not show whether the output was used, effective or reviewed by someone with relevant expertise.
Different measurements of task breadth
A separate Google ATLAS report found a related but differently measured pattern. Gemini activity appeared across occupations accounting for more than 88% of U.S. employment, but the median occupation crossed Google’s usage threshold for only 21% of its O*NET tasks.
Google measured how widely AI activity appeared within occupational task lists, while OpenAI measured whether individual messages mapped to work historically associated with another occupation. The results therefore capture different kinds of breadth and are not directly comparable.
How specific tasks cross professional boundaries
The analysis covered users in customer experience, design, engineering, finance, human resources, legal, marketing and sales. Cross-occupation tasks formed a majority of occupation-specific messages in five groups, reaching 77% for customer-experience users, 75% for designers, 69% for human-resources users, 56% for legal users and 53% for marketers.
Several activities traveled consistently across those occupational boundaries. Calculating financial data and troubleshooting computer systems ranked among the three most common outside tasks in each of the seven other groups. Developing promotional materials was the most common marketing-related activity among nonmarketing users, accounting for 25% of their messages assigned to marketing work.
Marketing and engineering work spread differently. Marketing tasks accounted for 8.9% of messages among workers in other fields, the highest outward share in the sample. Engineering tasks made up over 7% of messages from other occupations, even though engineers themselves devoted a relatively smaller almost 19% of their messages to work associated with another field.
The role of workspace size and internal resources
The crossover rate also varied with workspace size, but only for part of the user population. Among users in the middle 50% of message volume, cross-occupation work accounted for 18.9% of messages in workspaces with two to five seats and 16.3% in workspaces with 101 or more. No comparable monotonic decline appeared among users in the top quartile by message volume.
Workspace seats do not represent total company employment, however. OpenAI acknowledged that the measure may reflect differences in industry, specialization, organizational maturity, occupation or access to colleagues. The results therefore cannot establish that small-business employees cross occupational boundaries more often than workers at larger companies.
OpenAI offered one possible explanation: workers with fewer internal resources may ask ChatGPT for help with work that would otherwise involve a colleague or specialist. The report nevertheless cautioned that attempting a task does not remove the need for expertise, stating that “specialists remain critical for expert-level judgment and review.”
Individual crossover versus firm-level deployment
A separate Center for Economic Studies working paper by Census Bureau and University of Maryland researchers found that AI use within firms remained concentrated. Among firms reporting AI use in at least one business function, 57% used it in one to three functions. Among firms reporting employee use of generative AI in at least one work-related task, nearly 65% used it in one to three task categories.
Of firms reporting any AI-related task effect, 66% reported augmentation alone. The working paper used firm-level survey responses, including employers’ reports of worker-task use, rather than observed individual ChatGPT messages.