People are using ChatGPT at work less as a coding and computer helper and more as a sounding board for decisions, ideas and advice. That is the main finding of new data from OpenAI Economic Research, published on 26 October 2026 as part of its Signals programme, which tracks what workers ask the chatbot to do across nearly 100 countries.
The release maps work-related messages to 165 activities drawn from O*NET, the occupational database sponsored by the US Department of Labor. It covers seven quarters, from the first quarter of 2025 to the third quarter of 2026, and the data is free for anyone to download under a Creative Commons licence.
What people use ChatGPT at work for now
In the United States, the most common work activity in the first quarter of 2026 was "making decisions and solving problems", at 21.6% of work requests, according to figures reported by PPC Land from OpenAI's charts. "Thinking creatively" came second with 15.6%, followed by "documenting or recording information" at 11.8%.
Germany, the other country OpenAI broke out in detail, looks similar. Decision-making led there too, at 19.4% of work requests.
The bigger story is what changed over a year. Between the first quarter of 2025 and the first quarter of 2026, US shares moved like this:
- Thinking creatively rose from 11.8% to 15.6%
- Providing consultation and advice to others rose from 5.5% to 9.0%
- Documenting information rose from 10.6% to 11.8%
- Interacting with computers fell from 11.7% to 6.7%
- Getting information fell from 11.6% to 9.9%
In Germany, advice-giving grew even faster, from 7.1% to 11.7%, and moved from seventh to third place among the labelled activities.
Why computer tasks are shrinking
The drop in "interacting with computers", which O*NET defines as using hardware and software, including writing programs, is the most striking shift. It fell by five percentage points in the US and by 4.3 points in Germany.
There is a likely reason that has little to do with people abandoning technical work. The dataset only covers individual ChatGPT plans. It leaves out enterprise accounts and Codex, OpenAI's dedicated coding agent, which had 1.6 million weekly active users by February 2026. As programmers move to tools built for code, their requests disappear from this sample. A similar pattern showed up at Anthropic, which reported earlier this year that coding on Claude was shifting from its consumer app toward its API.
So the data does not prove that people are writing less code with AI. It shows that the chatbot window itself is being used for a different mix of jobs.
Work is spreading across more tasks
OpenAI also measured how concentrated work use is. Across 74 countries observed in every quarter, the ten most common activities made up an average of 56.0% of work requests in early 2025. By the third quarter of 2026 that share had fallen to 51.1%.
In plain terms, people are bringing a wider range of tasks to the tool. OpenAI also says countries have grown more different from each other over time, but that measure moved only from 22.2% to 23.0%, a small change that PPC Land flagged as thin support for the claim.
How ChatGPT use compares with real jobs
For the US, OpenAI compared its numbers with how work is distributed across the whole economy. The gaps are large. Planning and decision tasks make up 35.8% of ChatGPT work requests but only 9.3% of economy-wide activity. Communicating and interacting with people is the reverse: 7.0% of ChatGPT requests against 12.8% of real work.
Much of the economy barely appears at all. Activities such as moving materials, operating vehicles and maintaining health records fall into an "other" bucket that accounts for 43.1% of economy-wide work, because they did not show up often enough in ChatGPT conversations to be counted.
What the data leaves out
Readers should treat the numbers as shares, not volumes. The release does not say how many work requests were sent, how many people sent them, or how much total work use grew.
OpenAI says researchers do not read conversations. Automated systems classify messages after identifying information is removed, small groups are withheld and statistical noise is added. The company has not said which model does the classifying or how it decides that a message is work-related. The noise also makes figures for smaller countries less reliable, which matters for readers in Africa: the release does not break out any African country in its headline charts.
What this means for you
For workers, the shift is a useful signal about where these tools are most often trusted. The fastest-growing uses are judgement-heavy: weighing options, drafting ideas, giving advice. Those are also the tasks where an AI's errors are hardest to spot, so checking its reasoning matters more, not less.
For employers, the gap between ChatGPT use and the real shape of work shows where AI is not yet helping. Jobs built around physical tasks and face-to-face communication remain largely untouched.
For researchers and journalists, the open data is the main prize. Anyone can now compare countries and track change quarter by quarter, including whether African markets follow the same path as the US and Germany once enough data exists.
This report sits alongside other recent ChatGPT changes, including the expansion of ChatGPT ads to 89 countries, among them Nigeria and South Africa.
