OpenAI Grades Its Own ChatGPT Homework with Signals Data

Here's the number that should make you pause: six months after signing up, ChatGPT users send 50% more messages per day than they did on day one. That's not novelty wearing off. That's habit formation — the exact metric every consumer platform from Instagram to DoorDash has spent a decade optimizing for.

OpenAI just published Signals, a self-reported dataset tracking how people actually use ChatGPT across countries and demographics, built on classification of roughly 300,000 consumer messages per month since July 2024. The headline framing is that the world is moving from asking ChatGPT things to doing things with it. In workplace settings, users are now more than twice as likely to use the tool for completing tasks — writing, coding, analysis — than they are outside of work.

That's a meaningful shift, assuming you trust the messenger.

The multimedia surprise nobody predicted

The fastest-growing category globally isn't code generation or business analysis. It's multimedia — image generation, image understanding, that whole bucket — now sitting at 7.8% of all messages and cracking 10% in Brazil and Colombia. That's a genuinely interesting data point buried under OpenAI's productivity narrative. People aren't just using ChatGPT to write memos. They're using it as a visual tool, and Latin America is leading that charge, not Silicon Valley.

Who's actually driving growth

The geographic story is the most credible part of this release, mostly because it's the hardest to spin:

• Adoption is accelerating fastest in Africa, Latin America, and Oceania — regions with historically low per-capita usage
• Users over 35 are rising in nearly every country tracked
• France and Czechia saw that over-35 cohort jump more than 10 percentage points in a single year
• Growth is fastest in lower-HDI countries, not the wealthy early-adopter markets that got all the attention in 2023

A separate Signals update pegs Q1 2026 as the moment ChatGPT went properly mainstream, name-checking Japan, Mexico, Tanzania, Brazil, Costa Rica, and Myanmar as countries climbing the usage rankings. That's not the adoption curve anyone predicted when GPT-4 launched to a wave of San Francisco hype.

What Nobody Is Talking About

Every single data point in this report comes from OpenAI classifying its own users' messages. There is no external audit here. No independent researcher validated these categories. "Asking" versus "doing" versus "expressing" is OpenAI's taxonomy, applied by OpenAI, published by OpenAI, to make OpenAI's product look like it's winning the transition from toy to tool.

That's not a small caveat. It's the whole ballgame. If a huge chunk of "doing" work has already migrated to API-based agents and enterprise copilots that don't show up in consumer chat logs, this dataset is measuring last year's use case while missing this year's growth market entirely.

None of this means the underlying trend is fake. The over-35 growth, the multimedia surge, the Latin American momentum — these patterns are plausible and align with what independent app-store data has shown for months. But treating a company's internal analytics as neutral research is how tech journalism got played during the dot-com boom, and it's how it's getting played again now.

For developers, the actionable signal is simpler than OpenAI's press release: build for task completion with verification, not chat. Localize aggressively — more than half of active users now work primarily in non-English languages. And build for phones in Lagos and São Paulo, not just laptops in Palo Alto. That part of the story, at least, checks out.