Every week, 700 million people fire up ChatGPT and send more than 18 billion messages. That’s about 10% of the world’s adults, collectively talking to a chatbot at a rate of 29,000 messages per second.
The question is: what on earth are they talking about?
OpenAI and a team of economists recently released a fascinating paper that digs into exactly that. It’s the first time we’ve seen a systematic breakdown of how people actually use ChatGPT in the wild.
There’s one important caveat though: the study only looks at consumer accounts (Free, Plus, and Pro). No Teams, no Enterprise, no API. That means all the numbers you’re about to see skew toward personal usage rather than business use.
But even with that limitation, the trends are clear. And when you combine the consumer data with what we know about enterprise usage, a bigger story emerges about how AI is reshaping both work and daily life.
Work vs. Non-Work: AI Moves Into Daily Life
In mid-2024, about half of consumer ChatGPT messages were work-related. Fast forward a year and non-work usage dominates. 73% of messages are about personal life, curiosity, or hobbies.
Some of this is skew: Enterprise data isn’t in here, and yes, plenty of serious work happens on corporate accounts. But I don’t think that fully explains it. There is a real trend of people bringing ChatGPT into their everyday lives.
That tracks with my own journey. Back in 2020, when I first used the GPT-3 API, it was strictly work. I was building a startup on top of it back then so it was all about product development, copywriting, business experiments.
When ChatGPT launched, I still had a “work-only” account. Over time, I started asking it for personal things too. Today? I’m about 50-50. And that’s exactly what the data shows at scale.
The paper also shows that each cohort of users increases their usage over time. Early adopters send more messages than newer ones, but even the new cohorts ramp up the longer they stick around.
That also reflects my personal experience. The more I played with ChatGPT, the more I discovered new ways to use it, from drafting a proposal to planning a weekend trip. It went from a tool I used for certain activities to something I turn to almost immediately for any activity.
The Big Three Use Cases
When you zoom out, almost 80% of all usage falls into three buckets:
- Practical Guidance (29%): tutoring, how-to advice, creative ideation.
- Seeking Information (24%, up from 14%): essentially, ChatGPT-search.
- Writing (24%, down from 36%): drafting, editing, translating, summarizing.
What’s fascinating here is the growth of seeking information. The move from Google to ChatGPT is real. People are asking it for information, advice, even recommendations for specific products. Personally, I’ve used it for everything from planning a trip to Barcelona to asking why so many Japanese restaurants feature a waving maneki-neko cat statue.
There’s also a very big opportunity here in the education space. If we break it down further, 10.2% of all ChatGPT messages are tutoring and teaching requests. That’s one in every ten conversations, making ChatGPT one of the world’s largest educational platforms.
Now, when you look at work-related queries only, writing is still king: 40% of all work-related usage is writing. And that makes sense. Everyone deals with emails and business communications.
Interestingly, two-thirds of writing requests are edits to user-provided text (“improve this email”) rather than net-new generation (“write a blog post for me”). AI is acting more as a co-writer and editor than a ghostwriter.
Where’s Coding?
One surprise: only 4.2% of consumer ChatGPT usage is programming-related. Compare that to Claude, where 30%+ of conversations are coding.
But that doesn’t mean coding with AI isn’t happening. It’s just happening elsewhere (in the API, in GitHub Copilot, in Cursor, in Claude Code). Developers don’t want to pop into a chatbot window; they want AI integrated into their IDEs and workflows.
So the consumer product underrepresents coding’s real importance.
Self-Expression: Smaller Than Expected
Another surprise: “self-expression” (role play, relationships, therapy-like use) is only 4.3% of usage. That’s far smaller than some surveys had suggested.
Part of me wonders if some of these conversations were misclassified. But if the data’s accurate, I’m actually glad. We already know AI has a sycophancy problem. The last thing we need is people turning it into their therapist en masse.
Further on in the research, there’s more evidence to indicate this: self-expression had the highest satisfaction scores of any category. The good-to-bad ratio was almost 8:1, way higher than writing or coding. People seem happiest when using it for therapy.
Asking vs. Doing
The researchers also classified queries into three intents:
- Asking: seeking info or advice (“What’s a good health plan?”).
- Doing: asking ChatGPT to produce an output (“Rewrite this email”).
- Expressing: sharing feelings or views (“I’m feeling stressed”).
Across consumer usage:
- 49% Asking
- 40% Doing
- 11% Expressing
Here’s what’s interesting: Asking is growing faster than Doing, and Asking gets higher satisfaction.
Why? Because asking for advice or information is pretty straightforward. There’s not a lot that can go wrong if you ask the AI what the capital of Canada is.
But when people ask ChatGPT to do something, they often don’t provide enough context for a great output. In writing, for example, “write me a blog post on fitness” usually gives you generic AI slop. Having worked with multiple companies and trained professionals on how to use ChatGPT, I often see them try to get an output without adding any context or prompting the AI well.
But, as models get better at handling sparse instructions, and as people get better at prompting, Doing will likely grow. Especially with OpenAI layering on more agentic capabilities. Today, ChatGPT is an advisor. Tomorrow, it will be a doer too.
Who’s Using ChatGPT?
Some demographic shifts worth noting:
- Age: Nearly half of usage comes from people under 26.
- Gender: Early adopters were 80% male; now, usage is slightly female-majority.
- Geography: Fastest growth is in low- and middle-income countries.
- Education/Occupation: More educated professionals use it for work; managers lean on it for writing, technical users for debugging/problem-solving.
That international growth story is remarkable. We’re witnessing the birth of the first truly global intelligence amplification tool. A software developer in Lagos now has access to the same AI coding assistant as someone in San Francisco.
For businesses, this matters. Tomorrow’s workforce is AI-native, global, and diverse. Employees (and customers) are going to bring consumer AI habits into the workplace whether enterprises are ready or not.
ChatGPT as Decision Support
When you look at work-related usage specifically, the majority of queries cluster around two functions:
- Obtaining, documenting, and interpreting information
- Making decisions, giving advice, solving problems, and thinking creatively
This is the essence of decision support. And in my consulting work, it’s where I see the biggest ROI. Companies want automation, but the biggest unlock is AI that helps people make smarter, faster decisions.
The Big Picture
So what does all this tell us?
For consumers: ChatGPT is increasingly a part of daily life, not just work.
For businesses: Don’t just track “what consumers are doing with AI.” Track how those habits bleed into the workplace. Adoption starts at home, then shows up in the office.
For the future: AI at work will center on decision support, not pure automation. The companies that understand this earliest will unlock the most value.
The intelligence revolution is already here, 29,000 messages per second at a time. The question is whether your organization is ready for what comes next.