A new working paper from the National Bureau of Economic Research recently dropped, and if you've ever wondered how other people use ChatGPT (and whether it's just you using it to rewrite your Slack messages), then this blog is for you.
It's called "How People Use ChatGPT", and it's the first economics paper to analyse anonymised message-level data from ChatGPT itself - 18 billion messages from 700 million users!
Me and ChatGPT have summarised some key trends.
Is Everyone Using it for Work?
Not as much as I expected.
Back in mid-2024, about half of all ChatGPT use was work-related. Now? Only 27%. Non-work use has taken over - rapidly.
That might sound surprising, especially if your own use is mostly productivity-based. But when you dig into what people actually do, it makes sense:
- "What's a good skincare routine?"
- "How do I teach fractions to an 8-year-old?"
- "Give me 5 fantasy football team names."
These aren't work queries. But they're useful. They're part of what the researchers call Practical Guidance - the most common use case of all.
And actually, this all makes generative AI feel easy and useful. It's a tool. People use it for life stuff. Like we do with Google. Or Excel. Or WhatsApp.
Writing is Still King (Especially at Work)
When people do use ChatGPT for work, they overwhelmingly use it for writing. And not just write me a press release - most work writing is actually:
- Editing drafts
- Rewriting awkward bits
- Translating content
- Summarising or clarifying something they already wrote
This makes a lot of sense to me as it's how I use it a lot. It's less "AI writes for me" and more "AI makes what I wrote better". Especially if you're working in a comms-heavy role (hello, teachers, marketers, consultants...).
Most People are Asking, Not Doing
One of the cleverest parts of the study is how they classified user intent:
- Asking (e.g. "How do I format a CSV?"")
- Doing (e.g. "Write a bash script to format a CSV")
- Expressing (e.g. "I hate CSVs")
Turns out, most people are Asking. And Asking messages consistently get higher satisfaction scores.
The researchers suggest this shows that people are treating ChatGPT like an advisor, not just a task doer. And if you've ever asked it to help you brainstorm a title, rewrite a paragraph, or understand a new concept, that'll probably ring true.
Which is quite the opposite of the "agent hype" seen all over social media.
Who's Using ChatGPT?
The gender gap is narrowing fast (thank god). It started off skewed male, but now usage is slightly higher among people with typically feminine names.
Under-25s dominate usage, but growth is happening across all age groups.
It's not just a "rich country" thing - usage has grown faster in lower-income countries over the last year.
People in professional jobs are much more likely to use it for work.
So yes, if it feels like you're seeing more and more students, early-career folks, and creatives talking about ChatGPT - you are. But this is usually how technology adoption works (and highlights the need for education among those of us not in GenZ).
A Few Grains of Salt
The study is rigorous and transparent, but it's also limited to users on Free, Plus, and Pro plans - not Business or Enterprise. So we're not seeing what's happening inside large teams or corporate workflows. Which is potentially where a lot of the "work" stuff is happens.
And while the paper is full of stats and charts, it doesn't capture the quality of conversations or their long-term impact. For example: does using ChatGPT to write emails make you a better writer over time? Or just faster?
I think that's one that needs a longer study.
I'll Leave You With a Thought
This paper reinforces something we see again and again at Taught by Humans: there's no one right way to use AI tools. People use them differently, depending on their goals, comfort level, and context.
And that's actually great news.
Because it means there's room for each of us to figure out what works, to explore, to question, to get better at prompting - or to just get help brainstorming dinner ideas.
Either way, the data shows that humans are still at the centre of how AI is used. And that feels like the right place to be.
Want to read the full paper? It's here, and surprisingly readable.