TL;DR
The short version
OpenAI's first country-by-country usage release says ChatGPT is shifting from a question box to a work tool. At work, people are more than twice as likely to use it to produce something — edit, code, analyse — than to ask for information. Adoption is broadening fastest across Latin America, Oceania and Africa, and among the over-35s.
The catch is structural. The dataset covers consumer accounts only — Free, Go, Plus and Pro — so a piece headlined about work is measuring personal accounts used for work, not enterprise deployment. And every figure is OpenAI grading its own homework: self-reported, classifier-labelled, with differential-privacy noise added before release. Read it as a direction of travel, not an audited fact.
Built on OpenAI's Economic Research report From asking to doing (6 Aug 2026), with an independent check against Pew Research.
From asking to doing
OpenAI's Q2 2026 Signals release gives the shift a name. At work, producing output — editing, coding, analysis — is more than twice as likely as information-seeking. Outside work, "asking" stays the largest category: people reach for the tool to find things out, not to make things.
AI is moving from something people ask to something they put to work.
OpenAI
The operator signal isn't that usage grew — it's where it converts to output. Task execution concentrates at work; casual use stays exploratory. Follow the doing when you decide what to build and where to measure.
The scope trap: this is shadow AI, not enterprise
The dataset covers Free, Go, Plus and Pro accounts only — explicitly not organisation-managed ones. So a piece headlined about work is measuring personal accounts used for work. That is arguably the more interesting finding — shadow adoption is real and large — but it is not a claim about enterprise deployment.
How to read vendor-graded adoption data
Every figure is OpenAI reporting on its own product. It is the sole data holder; content categories come from automated classifiers whose accuracy is unpublished; differential-privacy noise is deliberately added before release. The correct tag for these numbers is self-reported, not verified.
Watch the units and the metric. The global 35+ rise is stated as "5% higher share" while country figures use percentage points — a 5% relative rise is not five points. And "the adoption gap is closing" rides on a per-capita rank metric that is zero-sum: for Peru to rise, another country must fall.
Pew Research, Jun 2025 — the nearest independent yardstick, measured by survey rather than vendor telemetry.
What actually broadened
The direction of travel is broadening, geographically and by age. Per-capita adoption rose fastest in Latin America, Oceania and Africa — Peru, Uruguay and Costa Rica gained the most rank. The 35+ share of messages rose in nearly every country, with France and Czechia up more than ten points year over year. Multimedia is the fastest-growing use case, at 7.8% of global messages since ChatGPT Images 2.0 shipped in April 2026.
Directionally, this is consistent with independent surveys. The difference is method: OpenAI is counting its own messages, where a labour economist would measure displacement and a pollster would ask people directly. Same trend, three readings — keep more than one.
