AI Use Beyond Occupational Boundaries
OpenAI Economic Research analyzed over 1.5 million work-related ChatGPT messages from April to July 2026, revealing how workers utilize AI for tasks beyond their traditional job roles. The findings indicate that not only do workers experiment with tasks from other occupations, but they also frequently return to these tasks, integrating them into their workflows.
Differences in Prompting AI
The study highlights variations in how workers prompt ChatGPT based on whether the task aligns with their usual responsibilities. For tasks outside their primary roles, prompts are generally shorter and less likely to request detailed explanations, step-by-step guidance, specific formats, or advice. Conversely, when seeking assistance for cross-occupation tasks, workers tend to provide more context, examples, and often ask the AI to verify or check information.
Recurrence of Cross-Occupation Tasks
Among approximately 6,200 workers consistently observed during the study period, the proportion of AI activity related to previously used cross-occupation tasks increased from 13.1% in April to 25.9% in July. In follow-up analyses, workers returned to a cross-occupation task from the previous month 23.6% of the time, compared to 8.4% for those who had not previously engaged with that task.
Task-Specific Return Rates
Return rates for specific cross-occupation tasks varied significantly. For instance, discussing goods or services with customers had a return rate of 54%, while advertising or promotional writing saw a 44% return rate. Creating marketing materials had a 37% return rate, whereas explaining financial information was revisited only about 15% of the time. The overall average return rate across all cross-occupation tasks was 18.5%. These discrepancies may reflect how naturally AI fits into different workflows, as well as workplace norms and the perceived risks associated with errors.
Recommendations for AI Adoption
The report suggests that organizations should consider work design alongside access to AI tools when planning their AI adoption strategies. This approach may help reduce friction in identifying problems and advancing work processes with AI support.
Data Privacy and Research Integrity
OpenAI Economic Research ensured data integrity by excluding training-disabled data and messages lacking usable classifications. All analyzed messages were aggregated and anonymized, with researchers not accessing individual user messages. The dataset encompasses 1.5 million ChatGPT messages from April to July 2026, focusing on a consistent sample of about 6,200 workers for recurrence analysis.
Original source: OpenAI News