August 9, 2026. OpenAI published its first country-by-country ChatGPT usage data on August 6, and the finding being quoted everywhere is that at work people are more than twice as likely to use ChatGPT to complete a task or create something than they are outside work. Then there is a one-line note about the dataset that changes what that finding is a measurement of.
The numbers, and the note
- Doing beats asking at work. Across the globe, per OpenAI's Signals release, people are more than twice as likely to use ChatGPT to produce an output or perform a task at work than outside it. Outside work, asking remains the largest category.
- Multimedia is the fastest growing use case. It reached 7.8 percent of messages globally following the April 2026 release of ChatGPT Images 2.0, and more than one in ten messages in countries including Brazil and Colombia.
- The adoption gap is closing. In Q2 2026 rankings, Peru, Uruguay and Costa Rica rose the most, with parts of Latin America, Oceania and Africa growing faster in per-capita usage than established early adopters.
- Older users are the growth story in Europe. The share of messages from self-reported users aged 35 and over rose in nearly every country, up about 5 percentage points of share year over year, and by more than 10 points in France and Czechia.
- The dataset is personal accounts. OpenAI states the Signals dataset reflects messages sent within ChatGPT Free, Go, Plus and Pro accounts, the group of accounts generally managed by individuals rather than organizations.
The release was covered independently by EdTech Innovation Hub the same week.
What that footnote does to the headline
Enterprise and business accounts are excluded. So the work being measured here is work people are doing in accounts their employer does not manage. That does not make the finding weaker. It makes it a different and arguably more interesting one: this is the best public measurement to date of how much real output is being produced on unmanaged accounts, and the answer is that it is more than twice as task-oriented as personal use.
One more caveat OpenAI prints and most summaries drop: the age breakdown covers only users who self-reported their age on the platform, so it describes a subset, not the whole base.
What it means for operators
If you sell to SMBs, this is a procurement argument rather than a scare story. The work is already happening. The question a business owner should be answering is not whether their team uses AI, but whether the outputs their team is producing with it are reviewable, repeatable and owned by the company rather than sitting in someone's personal chat history. That is the gap between a person using a chatbot and an automated process the business owns, and it is usually the most persuasive thing to put in front of a prospect who thinks they have already solved this. For agencies, the multimedia number is the other signal worth acting on: image work is the fastest growing category on the platform and most agency service menus still do not have a line item for it.
Frequently Asked Questions
OpenAI states that the Signals dataset reflects messages sent within ChatGPT Free, Go, Plus and Pro accounts, which it describes as the group of accounts generally managed by individuals rather than organizations. Business and enterprise accounts are not part of this dataset, so the work usage it measures is happening on personally managed accounts.
Doing means using ChatGPT to produce an output or perform a task, such as editing, coding or running an analysis. Asking means seeking information or clarification. OpenAI reports that people are more than twice as likely to use ChatGPT for doing at work than outside work, while asking remains the largest category in personal use.
OpenAI reports multimedia generation, analysis and retrieval reached 7.8 percent of messages globally, making it the fastest growing use case since ChatGPT Images 2.0 shipped in April 2026. In some countries including Brazil and Colombia it exceeds one in ten messages, though it still trails practical guidance, writing and information seeking.
Worrying is less useful than deciding. The practical risks are that outputs are not reviewable, prompts and context are not retained by the company, and quality is not repeatable when the person leaves. The usual fix is not a ban but moving the two or three highest-volume tasks onto a managed account or an automated workflow the business controls, and leaving exploratory use alone.