OpenAI just published data that should make any small business owner pay attention. The company looked at more than 800,000 work related ChatGPT messages and found that when people ask for help with something tied to a specific job, close to half the time it is not their own job. OpenAI calls this task crossover, and it is the first report in a new research series the company plans to keep running.
The company sorted messages using O*NET, a US government database that lists what tasks belong to which occupation. It threw out generic requests like writing an email or building a schedule, since almost everyone does those regardless of role. What was left told a clear story: 43.5 percent of the specialist requests belonged to a different profession than the person asking.
Some jobs give away far more work than they take in. Customer service staff, designers, and HR people are the biggest borrowers, often asking ChatGPT to do financial calculations, fix a website problem, or draft something that would normally sit with a specialist. Marketing and engineering sit on the other side. Their tasks show up most often in other people's messages, meaning those two fields are quietly becoming the default skills everyone leans on.
The effect is bigger at small businesses, where there is no in house lawyer, no marketing department, and no IT desk to call. A contractor drafting a contract clause, a shop owner running numbers for a promotion, a retailer troubleshooting a broken checkout page: these are jobs that used to mean hiring someone or paying an outside firm. Now they often just mean opening ChatGPT.
This is not a neutral academic exercise. OpenAI released this data at the same time it launched a new agent product called ChatGPT Work and a program aimed squarely at small businesses, both selling the idea that ChatGPT can absorb work that used to require a specialist hire. The research and the product roadmap point the same direction on purpose.
The real question for any business owner is not whether this is happening, it clearly already is, but whether it is happening safely. A marketing person running a financial calculation with ChatGPT saves time. A marketing person reviewing a contract clause with ChatGPT and skipping a lawyer is a different kind of risk, because a wrong answer there does not show up until it costs money. OpenAI itself admits the data cannot tell whether this is genuinely new work being taken on, or just existing duties finally getting written down.
For smaller companies, the practical move is not to fight this trend but to decide where it is fine to let staff borrow AI generated expertise and where it still needs a real specialist to sign off. Job descriptions at most companies have not caught up to what people are already doing with AI at their desks. That gap is where the risk sits, quietly, until something goes wrong.