OpenAI finds ChatGPT tasks crossing occupational boundaries

OpenAI, the company behind ChatGPT, has analyzed how American users use the service at work. The study is based on a random sample of over 800,000 work-related messages from people in eight occupational groups.

The researchers compared each message's main task with job tasks in the US occupational database O*NET. They classified 16.8 percent of all work-related messages as information outside the user's own professional field. When general tasks such as emails, summaries and scheduling were excluded, the percentage was 43.5 percent.

An example is a marketer asking for help debugging a website. Another is a salesperson who reviews customer data without first handing the work over to an analyst. OpenAI calls the pattern task crossover, meaning that a task that usually belongs to one professional role appears in another.

The pattern was clearest in certain groups. Among the profession-specific messages, 53 percent of marketers' questions were about tasks outside their own field. Marketing and technical tasks also appeared frequently in the messages of other professions.

However, the study does not say that the users did the tasks well, saved time or managed without a specialist. It measures messages, not completed projects or work hours. The selection applies to American ChatGPT users and is not representative of the entire US or Swedish labor market.

OpenAI also writes that a model classified the messages' data without the researchers reading the messages themselves. The results are therefore based both on self-reported occupational roles and an automatic classification that may miss nuances.

In a separate global analysis, the international labor organization ILO has assessed that generative AI more often changes parts of a job than replaces entire professions. The ILO's study measures potential exposure, not actual ChatGPT use. Together, the evidence points to the job assignment being a more useful level than the professional title when a business is planning for AI.

Därför spelar det roll

For a smaller Swedish business, AI can make it possible to try simpler analysis, technology or marketing without an immediate handover. It does not change who has responsibility or specialist competence. When a task moves across a professional boundary, the organization also needs to decide who reviews the results.

Det här kan du göra

1. List which AI-supported tasks are already outside the employee's usual area of expertise.
2. Set a clear review point for legal, financial, security and other tasks where errors can have major consequences.
3. Measure actual quality and time consumption in your own work. The OpenAI study shows usage patterns, not results.