The trouble with revolutions is that they rarely send a formal RSVP. They arrive mid-conversation, rearrange the furniture of daily life, and casually make themselves indispensable before you’ve had time to finish your morning coffee. Such is the case with ChatGPT.
According to the authors of a new National Bureau of Economic Research working paper, How People Use ChatGPT,, within three years of launch, one in ten of the world’s adults is using it weekly. That’s 700 million people, every seven days, sending 18 billion messages. It is also roughly 29,000 prompts a second, if you're counting, and rest assured, someone always is.
The paper also marks a quiet but historic first: the inaugural economics study based on internal ChatGPT message data. Yes, the real thing.
And before you reach for your privacy panic button, take heart. No researcher ever saw a user message. Not one. Everything was classified by algorithm, passed through a privacy-preserving pipeline
And what did they find?
Generative AI is not creeping across borders. It is vaulting them. Lagos logs on before Los Angeles has finished updating its software. The pace of adoption outstrips anything in consumer technology history. Faster than smartphones. Faster than social media. Faster even than search which was the last great upheaval to how we think.
Part whisperer, part workhorse, entirely disruptive. ChatGPT doesn’t simply fetch your facts, it teases out your intent, reframes your indecision, edits your memos, brainstorms your business plans, and occasionally, very politely, tells you you’re being ridiculous.
The compulsion to query, to edit, to reframe, to ask “what if?” is no mere feature. It’s a fundamental shift in how we think, how we write, and how we decide.
Work vs. Play — The Surprising Tilt
When ChatGPT made its grand public debut in late 2022, the assumption among the clipboard-and-chart crowd was that this would be the new office darling. The hype was positively caffeinated: analysts would analyse faster, consultants would conjure memos with newfound flair, engineers would summon code like modern-day sorcerers. The office, they declared, was the natural habitat of generative AI; well-lit, deadline-driven, and decorously corporate.
But the story, like many office forecasts, went off-script.
While work-related queries certainly did their dutiful rise, they were soon elbowed aside by something far messier, far more human, and also entirely unfit for a quarterly earnings report. By June 2024, a narrow majority of messages (53%) were non-work related. Just a year later, that figure had ballooned to 73%. The curve is not flattening, it’s pirouetting.
“Non-work messages have grown faster and now represent more than 70% of all consumer ChatGPT messages,” the paper reports, with the sort of composed understatement one uses when announcing the kitchen has just been repurposed as a nightclub. And no, this is not simply because latecomers arrived, treating the chatbot like a digital plaything. Even the early adopters, these productivity purists, have slowly but surely shifted their usage from spreadsheets to story time.
The tilt reveals two truths, and like all good truths, they’re equal parts revealing and inconvenient.
First: people are not just looking for efficiency. They are looking for help. Real help. The sort that answers “what’s for dinner?”, coaches a teenager through a breakup, suggests a cardio plan with exactly zero jumping jacks, or invents a bedtime tale involving a narwhal named Steve. It’s tutor, therapist, sous-chef, and party planner all served in an eerily calm tone and flawless grammar.
Second: this is not unprecedented. Email, that most beige of digital tools, started out in boardrooms before becoming the medium of midnight rants and passive-aggressive family updates. Smartphones promised mobile productivity but wound up as portals to TikTok. And now ChatGPT, supposedly the herald of the workplace 2.0, is becoming a domestic appliance long before it becomes a professional standard.
The implications are, in a word, awkward. If consumers feel more chemistry with ChatGPT at home than in the boardroom, will the much-touted productivity boom arrive fashionably late? Or will the habits shaped in kitchens and classrooms, the offhand query, the rewritten birthday card, the quick fix for an existential crisis, slip into office life by stealth?
Either way, the distinction between work and play is beginning to blur. ChatGPT is now the whisperer in both spheres. And like any good whisperer, it knows just what to say—whether you're drafting a pitch deck or naming your sourdough starter.
The Big Three: Writing, Guidance, Information
Strip away the billions of daily prompts—the birthday limericks, the desperate meal plans, the furious attempts to rewrite corporate speak into something resembling English—and you’ll find three pillars quietly holding up the ChatGPT empire: Writing, Practical Guidance, and Seeking Information. These aren’t just common, they account for nearly 80% of all conversations. In the realm of generative AI, it turns out, most roads lead to words, wisdom, or Wikipedia-with-a-personality.
Writing, naturally, looms largest in the workplace. “Writing is the most common use case at work, accounting for 40% of work-related messages on average in June 2025,” the researchers find.
But here's the twist, the unexpected swerve that makes this more than just a tale of automated prose: it’s not brilliance users are after, it’s better versions of themselves.
Two-thirds of writing requests aren’t asking for original drafts. They’re asking for help cleaning up the human mess. People paste in their clunky attempts—emails marinated in passive voice, bios with all the warmth of a weather report, and ask ChatGPT to fix them. Shorter. Smoother. Smarter. The bot isn’t writing the next great novel. It’s editing your meeting notes. It’s become the world’s editor-in-chief, dutifully sanding the splinters off our collective syntax.
Practical Guidance isn’t far behind. Roughly 29% of usage is devoted to tutoring, teaching, how-to advice, or creative brainstorming. This is not the domain of dilettantes, it’s an intellectual gig economy. Students beg for help prepping exams. Professionals ask for frameworks that don’t scream “last-minute.” Aspiring entrepreneurs outline product ideas with the hopeful energy of a garage on fire. And around 10% of all messages are tutoring requests alone, a quiet but telling reminder that education is where generative AI is taking root most deeply, and most personally.
Then, there is Seeking Information, an old friend in new shoes. It’s surged from 14% of usage in mid-2024 to 24% just a year later. This is where ChatGPT starts elbowing into Google’s territory, and not politely. Users want to know things: about people, products, places. But they don’t want ten blue links, they want the answer, served warm.
Ask about Boston Marathon qualifying times, and you’ll get a table. Ask how to train for it with your dodgy knee and three free mornings a week, and you’ll get a plan stitched to your life like a bespoke suit. This is the distinction that matters. One is information retrieval. The other is tailored guidance. The latter, as it happens, is winning.
Yes, there are other categories such as technical Help, Multimedia, Self-Expression, but they’re lingering politely in the margins. Technical queries have declined to just 5% of usage as coders migrate to more specialist tools (a quiet revolution in itself). Multimedia, despite its brief fame during the 2025 image generation boom, accounts for only 7%. And Self-Expression, everything from romantic advice to roleplay scenarios, makes up less than 3%. The heart may want what it wants, but it apparently wants it less than a bullet-pointed action plan.
This hierarchy already seems etched in stone. ChatGPT is neither the new Google or the new Photoshop. It is the red pen, the mentor, and the encyclopaedia, bundled into one digital whisper. In offices, it plays spellcheck and style guide. At home, it’s the tutor, the troubleshooter, the patient explainer when no one else has time. And everywhere, it’s that curious little echo chamber, turning our rough questions into sharper answers—while leaving us wondering how we ever lived without it.
Coding Isn’t King
One of the more amusing features of modern mythmaking is the conviction that large language models are destined to become the cherished sidekicks of every software developer on Earth, digital copilots whispering perfect code into their weary ears. It’s a lovely fantasy. Very cinematic. Also, wildly inaccurate. At least in the case of ChatGPT.
“Only 4.2% of ChatGPT messages are related to computer programming,” the paper finds, with the dry understatement of someone pointing out the emperor’s rather drafty wardrobe. In contrast, Handa et al. (2025) found that a full 33% of Claude conversations involve coding, a figure that suggests Anthropic’s model has become the darling of devs and debugging obsessives everywhere.
But here, in the noisier, broader, and decidedly more human ecosystem of ChatGPT, code is a niche pursuit. The real compulsion isn’t logic, it’s language.
People aren’t lining up to compile Python scripts. They’re here to wrestle with the blank page. To punch up an email. To clarify a thought that arrived garbled. To turn a jumble of notes into something that might pass as a memo. This is less co-pilot for programmers, more ghostwriter for the knowledge class, a kind of literary valet for anyone with too many meetings and not enough verbs.
ChatGPT isn't writing your software. It's rewriting your sentences.
Decision Support: The Hidden Killer App
If writing is ChatGPT’s marquee act and the obvious starlet in a digital Broadway revue, then decision support is its understudy, quietly stealing the show. It isn’t flashy. It doesn’t hum with the same performative flair. But make no mistake: this is the role that may change the script of knowledge work altogether.
The paper introduces a deceptively simple taxonomy: Asking, Doing, Expressing. Of these, Asking dominates. Nearly half of all messages are requests for information or advice meant to guide a decision.
At work, Asking becomes even more consequential. These queries aren’t just growing the fastest, they’re also rated highest in user satisfaction. Professionals increasingly lean on ChatGPT not simply to write things down, but to think things through. The chatbot has begun to play the role of management consultant at scale: omnipresent, unbothered, and terrifyingly prompt.
And here, if you listen closely, is the rustle of economic significance.
ChatGPT’s real productivity boost likely comes not from polished prose, but from upgraded judgment. Decision support, particularly in knowledge-intensive fields, is where human value lives, and where AI is now setting up shop. Better decisions, made faster, with fewer blind spots: that’s not just a workflow improvement. It’s a competitive advantage.
In practice, this translates to millions of analysts, project leads, middle managers and quietly overwhelmed execs using the bot to compare strategies, test assumptions, and whisper-check their instincts. ChatGPT isn’t just typing for them. It’s thinking with them.
And in knowledge work, there’s nothing more valuable than a thinking partner who never sleeps, never blinks, and never once asks for credit.
A Tool for Everyone, Not Just Elites
For a technology born in Silicon Valley and trained on the detritus of the internet, ChatGPT has achieved something quite rare: it has managed to escape the gravitational pull of elitism. The demographic story, if not quite revolutionary, is certainly subversive.
At launch, around 80% of users bore typically masculine names—a tech stereotype so familiar it practically wore a hoodie, and launched a podcast. But by June 2025, the script had flipped. “In the first half of 2025, we see the share of active users with typically feminine and typically masculine names reach near-parity. By June 2025 we observe active users are more likely to have typically feminine names,” the authors note.
This is not just rare. It’s almost unheard of. Very few technologies have seen the gender gap not only close but reverse in three short years. It’s not just that women caught up, it’s that they overtook. And one suspects they didn’t need a manual to do it.
Age, too, tells its own tale. Nearly half of all messages come from users under 26. Part teacher, part coach, part creative partner, it’s become the third party in millions of coming-of-age conversations, advising on everything from essay structure to emotional resilience.
Geography, meanwhile, skews global. Usage has grown fastest in low- and middle-income countries, where necessity sharpens the appetite for digital help. Education? That skews up, for instance, those with degrees are more likely to use ChatGPT for work. But the real pattern is broader.
Generative AI, once feared as the plaything of the technocratic elite, is proving itself something far more disruptive: a tool for those with the least access to traditional resources. The young, the global, the overextended, the underestimated. They are not just using the tool, they are shaping its future.
Evolution thrives at the margins, as my father used to insist; once again, it is the margins, not the centre, that show us where the frontier really lies.
Universal Work Patterns
For a tool heralded as the harbinger of personalised intelligence, ChatGPT seems to be turning the global workforce into a study in symmetry. Across continents, job titles, and education levels, people are using it in ways that look… eerily similar.
The researchers, not content with anecdotes, mapped millions of work-related conversations against O*NET, the U.S. Department of Labor’s delightfully exhaustive taxonomy of occupational tasks. What they uncovered was less a kaleidoscope of digital behaviour than a recurring motif.
Eighty-one percent of work messages fall into just two buckets;
- Obtaining, documenting, and interpreting information.
- Decision-making, advising, solving problems, and thinking creatively.
“The majority of ChatGPT usage at work appears to be focused on two broad functions: 1) obtaining, documenting, and interpreting information; and 2) making decisions, giving advice, solving problems, and thinking creatively,” the authors reveal.
That resonance across roles is, frankly, remarkable. In business and management, the dominant task is writing, accounting for 52% of work-related messages. In technical fields, it’s problem-solving and code assistance. Educators and health professionals lean heavily on drafting, documentation, and the gentle art of explaining things clearly to people who’d rather be anywhere else.
And yet, despite the surface variety, the shape is the same. ChatGPT isn’t just a niche tool for the code-wranglers or the email-wearied. It’s become a thinking partner and a writing partner, a quiet co-pilot in the white-collar cockpit.
This uniformity matters more than it lets on. It signals that generative AI is less a clever gadget than a general-purpose technology in the classic sense. A clerk in Nairobi and a consultant in New York may ask different questions, but both are turning the same cognitive crank: gather, interpret, decide, record. Repeat.
For executives, this is the bit that should make them spill their coffee.
If every knowledge worker is leaning on the same whisperer for help with core mental labour, then strategic advantage stops being about who has more data, and starts being about who asks the better questions. The firms that win won’t be the ones with the biggest servers. They’ll be the ones with the sharpest prompters.
Which brings us, inevitably, to culture.
The most in-demand workplace skills such as structured problem-solving, clarity in writing, creativity are precisely what ChatGPT magnifies. Workers who can wield it with fluency will surge ahead. Those who fumble may find themselves on the slow track, no matter how many KPIs they’ve met.
So, no, ChatGPT isn’t here to make you redundant. It’s here to make you, curiously, just like everyone else: asking better questions, writing better answers, and making decisions with a little more bite.
And somewhere in all that sameness, the clever will find their edge.
The Economic Upshot
Economists are not known for understatement, and yet even they seem a little breathless this time. The numbers, frankly, are staggering.
Collis and Brynjolfsson (2025) estimate that ChatGPT delivered a consumer surplus of at least $97 billion in the U.S. alone in 2024. Yes, billion, with a B, and without including a single chatbot-generated haiku. Globally, the figure is even higher, though at that point one suspects the commas blur.
Where the gains shine brightest is in knowledge-intensive work. Not the factory floor, but the think tank. Not the wrench, but the whiteboard. Here, ChatGPT doesn’t replace labour, it enhances judgment. It’s not stealing your job. It’s helping you figure out how to do it slightly better, slightly faster, and with fewer meetings.
The economic value of ChatGPT lies not in what it does, but in how it helps humans decide what to do. That single shift, from execution to cognition, is where the real treasure lies buried.
This is not a minor distinction. Much of the public debate around automation still swirls around substitution: machines muscling out humans like grim little robocapitalists. But the evidence here suggests something subtler, and more promising. Augmentation. Machines not replacing decisions, but accelerating them. Sharpening them. Polishing the indecisive bits until they shine.
This reframing matters, especially for policymakers and executives, the very people most likely to treat AI as a headline threat rather than a systemic upgrade. The compulsion to fear AI as a job destroyer may yet blind them to its deeper promise: lifting the overall quality of human decision-making across the workforce.
Handle with care
ChatGPT has threaded itself into our daily rituals with the soft insistence of a tool we now reach for without thinking. It helps us write better, decide faster, learn quicker, and, in some cases, feel a little less alone in the digital tangle. The advantages are undeniable. It smooths out the mess of modern knowledge work with the efficiency of a ghostwriter and the charm of a therapist who’s read your emails.
But its power lies not just in what it does, but in how easily it becomes part of how we think. That ease is both its greatest feature and its quietest risk.
Because when a tool helps you decide, it also shapes what you see as a decision. When it finishes your sentence, it might just redirect your thought. And when hundreds of millions rely on the same whisperer, the real question becomes: whose voice, exactly, is doing the whispering?
Yes, the bot is brilliant. But brilliance, unsupervised, has a habit of burning through things.
So by all means, use it. But don’t fall asleep with it driving.