Back

OpenAI Details ChatGPT Work Use Across Task Categories

The share of classified ChatGPT work messages represented by each country’s 10 most common activities fell between early 2025 and the third quarter of 2026, according to OpenAI. Across 74 countries observed in every quarter, the average share was 51.1% in Q3 2026, down from 56.0% in Q1 2025.

The figures form part of an OpenAI Signals release published on October 26, 2026. OpenAI said the analysis concerns work-related messages from individual ChatGPT plans. Enterprise messages and use of Codex are outside its coverage.

The wider dataset spans nearly 100 countries, with observations recorded quarterly from Q1 2025 through Q3 2026, according to the company. It covers 165 categories called Intermediate Work Activities, which describe tasks shared across occupations. Examples range from creating teaching materials and marketing content to programming and business-data analysis.

For its classification framework, OpenAI uses O*NET, which organises tasks associated with particular occupations into broader categories of work. The company said this provides a basis for comparing the activities classified in ChatGPT messages with information about occupations and employment.

OpenAI also stated a limitation on comparisons with the employment-weighted mix of tasks in mapped U.S. jobs. About 43% of that mix belongs to other activity categories that the release does not break out individually. Those categories include materials handling, vehicle operation and the upkeep of health or medical records. According to OpenAI, such activities appeared too inconsistently in the original U.S. ChatGPT sample to be reported separately.

The company described an automated process for classifying messages rather than having researchers read individual conversations. Under the method outlined in the release, identifying information is removed before automated systems assign messages to categories.

OpenAI said the published figures combine observations into aggregate statistics, while results for small groups are withheld. It also said statistical noise is added as a safeguard for individual messages. These measures are part of the processing and privacy methods described for this release.

* This article is based on publicly available information at the time of writing.