OpenAI released new Signals data on October 26, 2026, mapping consumer ChatGPT use for work across 165 activity categories and nearly 100 countries. The research covers quarterly data from the first quarter of 2025 through the third quarter of 2026. Enterprise messages and Codex use are excluded.
The categories come from O*NET’s Intermediate Work Activities framework, which groups tasks across occupations. This gives researchers a common vocabulary to compare usage patterns with other employment data. It does not turn messages into a count of jobs automated.
What the concentration figure shows
For 74 countries observed in every quarter, OpenAI reports that the ten most common activities accounted for an average of 51.1% of each country’s classified work messages in Q3 2026, compared with 56.0% in Q1 2025.
Subtracting the two figures shows a 4.9 percentage-point decline. That is a calculation from OpenAI’s published summary, not an independent analysis of the raw CSV files. It also describes an average of country-level shares, rather than a single pooled share of every message worldwide.
A lower share for the top ten means that the classified message mix became less concentrated on that measure. It does not show that total usage fell. A percentage can decline while the number of messages rises, and this figure alone doesn’t provide the message counts needed to answer that question.
The sample needs to stay visible in every conclusion
The download and methodology page and data dictionary are the references for interpreting the published fields. OpenAI describes automated classification, aggregated statistics, suppression of small groups, and statistical noise to protect individual messages.
| What the release can inform | What it cannot establish by itself |
|---|---|
| The mix of classified consumer work messages | The full mix of enterprise or Codex work. |
| Differences between reported countries and quarters | A causal explanation for those differences. |
| Connections to occupational task categories | Productivity gains or jobs displaced. |
For readers following AI adoption in businesses, the missing enterprise population is consequential. Consumer use may overlap with work, but it cannot measure companies’ adoption rates. Business conclusions need additional evidence about the organizations and outcomes discussed.
This report reviews the announcement and methodology. Numerical summaries are attributed to OpenAI Signals; the underlying archive has not been independently analyzed here.
