Work is now 32.2% of ChatGPT requests, and writing is no longer the top use

What OpenAI's numbers show

As PPC Land reported, OpenAI released version 3.0 of a dataset called OpenAI Signals on Monday, October 26. It tracks what consumers ask ChatGPT, month by month, from July 2024 to September 2026. Over those 27 months, requests tied to work dropped from 51.6% to 32.2%.

Writing, once the main reason people opened the app, went from 34.1% of requests to 21.4%. The biggest single topic since March 2025 has been what OpenAI calls practical guidance, meaning everyday how-to help and advice. It edged up from 26.1% to 28.1%.

Looking at all requests together, non-work use rose from 48.4% to 67.8%. In plain terms, about two in three conversations are now not about a job.

How the counting works

No person reads the conversations. An automated classifier, which is itself an AI language model, labels each message. It uses 23 detailed labels that are rolled up into seven broad topics. Other classifiers judge whether a message is about work, school or neither, and what the user is trying to do.

Each month's sample is 1,000,000 messages from adult consumer users. That is up from 300,000 in earlier versions, and earlier months were recalculated. So the new series does not match older OpenAI figures exactly.

OpenAI adds deliberate statistical noise, a standard privacy technique called differential privacy, and hides very small groups. The company is open about the gaps. Enterprise accounts, deleted and opted-out accounts, and Codex (its coding agent) are all left out. The README itself warns that the figures likely understate business use.

Where the shift shows up

Two topics moved the most. Technical help, which covers programming, maths and data analysis, fell from 13.5% of requests to 4.6%. PPC Land suggests much of that work has moved to tools the data does not cover, such as Codex, rather than disappearing. A rival's data points the same way. Anthropic's research found coding moving away from the Claude.ai chat window.

Multimedia, mostly image creation, grew fastest, from 2.0% to 8.8%. PPC Land ties the jump to image generation arriving inside GPT-4o on March 25, 2025. The share rose from 5.4% that March to 9.4% the next month. Self-expression also more than doubled, from 4.0% to 8.9%.

The way people phrase things changed too. PPC Land's calculations show "expressing" (statements that neither ask a question nor request a task) rising from about 11.7% of requests to 29.1%. Asking fell from about 45.7% to 36.9%, and asking ChatGPT to do something fell from about 42.7% to 33.9%. People increasingly just tell ChatGPT things.

Who is typing now

The user base has aged. The 18 to 24 group made up 39.3% of requests in July 2024 and 30.6% by September 2026. The 25 to 34 group, at 34.3%, has led every month since December 2025. Every older band grew, with those aged 65 and over rising from 0.9% to 2.7%.

Older users want somewhat different things. Among people aged 65 and over, multimedia reached 10.4% of requests, the highest of any age band. Those in their late fifties and early sixties leaned toward seeking information.

Gender is estimated from first names, not declared by users. By that measure, feminine names accounted for 51.8% of requests in September 2026, up from 42.2% in July 2024.

Countries differ widely. In the third quarter of 2024, work made up more than half of requests in 44 of 137 countries. Two years later that was true in none. Singapore had the highest work share, at 42.9%, and Iraq the lowest, at 12.4%.

Why advertisers are watching

PPC Land's interest is commercial. ChatGPT now sells ads, but only on its Free and Go plans. Around three in four Free-plan requests, and 57.6% of Go-plan requests, were not work-related in September 2026. Those are the people advertisers can reach.

An older, more mixed audience that asks for advice and images looks more like a mass medium than a student and office tool. But the rise in personal, "expressing" conversations has a catch. OpenAI's ad rules keep ads away from topics such as health, mental health and politics. The data does not say how much of the personal talk falls into those categories.

Digiday reported in May that one advertiser had spent $2,500 of a $250,000 commitment over four weeks. OpenAI executives blamed careful filtering. If more conversations are personal, the pool of eligible moments could grow more slowly than the number of users.