ChatGPT: Why do most of your users ask for help writing - prose, not code?
- Reference: 1758003229
- News link: https://www.theregister.co.uk/2025/09/16/chatgpt_usage_study/
- Source link:
The [1]study [PDF], penned by researchers from OpenAI and the USA’s National Bureau of Labor Research, used a random selection of messages sent to ChatGPT by users signed up for Free, Plus, and Pro plans between May 2024 and June 2025. Those are plans OpenAI aims at individuals, not businesses. That may help to explain why the study found steady growth in work-related messages but even faster growth in messages of a personal nature, which have grown from 53 percent to more than 70 percent of all usage.
If you’ve used ChatGPT during the study period and fear the authors read your input, the paper explains that the researchers devised a method that meant no human read your work. They also excluded material from users who opted out or deleted their accounts, or were under 18 at the time they used the bot.
[2]
The study found most queries to ChatGPT concern either “Practical Guidance”, “Seeking Information”, or “Writing”.
[3]
[4]
An example the paper offers for “Seeking Information” is asking ChatGPT to find the qualifying times by age and gender for the Boston Marathon, while “Practical Guidance” involves asking ChatGPT to devise a fitness program that will help a user prepare to run in the event.
“Writing” covers requests for “automated production of emails, documents and other communications, but also editing, critiquing, summarizing, and translating text provided by the user.” The study finds writing accounts for 40 percent of ChatGPT usage at work, and that almost two thirds of such requests ask ChatGPT to edit, critique, or translate text rather than create it from scratch.
[5]
Most of you still appear to write your own material. As we always do here at The Register .
The authors also considered the outcome users wanted to achieve when using ChatGPT, and categorized those in three ways:
Asking for information or clarification to inform a decision
Doing - a request for ChatGPT to produce some output or perform a particular task
Expressing sharing a view or feeling without seeking any information or action
The study found about 49 percent of messages analyzed involved Asking, 40 percent were Doing tasks, and the remainder involved Expressing.
Those numbers shift in messages related to work, 56 percent of which involve Doing – and nearly three-quarters of those are Writing tasks.
The authors found the prevalence of writing tasks notable for two reasons.
[6]
“First, writing is a task that is common to nearly all white-collar jobs, and good written communication skills are among the top ‘soft’ skills demanded by employers,” they wrote. “Second, one distinctive feature of generative AI, relative to other information technologies, is its ability to produce long-form outputs such as writing and software code.”
Yet just 4.2 percent of input considered in this study concerned computer programming, a rate the authors compared to another study that found a third of work-related requests sent to Anthropic’s Claude chatbot involve coding.
[7]OpenAI reorg at risk as Attorneys General push AI safety
[8]OpenAI eats jobs, then offers to help you find a new one at Walmart
[9]Microsoft unveils home-made ML models amid OpenAI negotiations
[10]ChatGPT hates LA Chargers fans
The researchers then mapped the content of messages sent to ChatGPT against the Occupational Information Network (O*NET) survey of job characteristics supported by the U.S. Department of Labor.
“We find that about 81 percent of work-related messages are associated with two broad work activities: 1) obtaining, documenting, and interpreting information; and 2) making decisions, giving advice, solving problems, and thinking creatively,” they wrote, adding that this pattern is very similar across different occupations.
“Overall, we find that information-seeking and decision support are the most common ChatGPT use cases in most jobs,” the paper states. “This is consistent with the fact that almost half of all ChatGPT usage is either Practical Guidance or Seeking Information.”
Other findings include:
A big shift in the gender of users. In the months after ChatGPT’s debut, 80 percent of active users’ names were masculine. That number fell to 48 percent as of June 2025;
More than half of all messages sent by adults came from users aged under 26;
Educated users and users in highly paid professional occupations are substantially more likely to use ChatGPT for work.
The paper concludes “users currently appear to derive value from using ChatGPT as an advisor or research assistant, not just a technology that performs job tasks directly” and that the bot “likely improves worker output by providing decision support, which is especially important in knowledge-intensive jobs where productivity is increasing in the quality of decision-making.”
That's a welcome finding for OpenAI given the company hopes to do things like [11]spend $300 billion on Oracle Cloud. ®
Get our [12]Tech Resources
[1] https://www.nber.org/system/files/working_papers/w34255/w34255.pdf
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aMk1PHiMPM_-k_gOVw9mcgAAAMk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aMk1PHiMPM_-k_gOVw9mcgAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aMk1PHiMPM_-k_gOVw9mcgAAAMk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aMk1PHiMPM_-k_gOVw9mcgAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aMk1PHiMPM_-k_gOVw9mcgAAAMk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2025/09/05/openai_reorg_at_risk/
[8] https://www.theregister.com/2025/09/05/openai_jobs_board/
[9] https://www.theregister.com/2025/08/28/microsoft_unveils_housemade_models_amid/
[10] https://www.theregister.com/2025/08/27/chatgpt_has_a_problem_with/
[11] https://www.theregister.com/2025/09/11/openai_reportedly_on_the_hook/
[12] https://whitepapers.theregister.com/
Because people can't be bothered to read or write anything more than a couple of paragraphs nowadays.
The last time I was asked for a report (cybersecurity) by my employers, I was accused of using AI - which I'm vehemently opposed to doing - purely because it was: long, comprehensive, detailed, correct and covered everything they could think of to cover. Apparently "nobody does that".
And it's a trend I've seen grow even before AI... people just can't be arsed to read comments or documents that are longer than a couple of paragraphs unless they're forced to, or they do it as part of a larger group, etc. Hell, I meet increasing numbers of people who have never read a book, not even a short Terry Pratchett or similar.
But do you know what? Sometimes you NEED more than the basics in a document. I literally write up contracts and binding agreements as part of my job and everyone just assumes I'm copy-pasting or using AI to do so. No, I'm not. Because I actually *read* those things and I know what they need to say, rather than just relying on "Oh, you know what I mean, though".
There are also related reasons why people don't understand politics, the economy, what they're voting for, how credit cards work, how debt-collection works, etc... it's because they've never read stuff.
TL;DR: People are dumb.
Sorry, you lost me after the second paragraph....
I've had coworkers sigh when I send an email that is longer than a tweet or text. And I just piss them off in Slack. #@$%#@ wall of text. Then they get even by scheduling a Teams meeting.
What's with the 'nowadays' :)
My boss 30 years ago - a good guy - had a reputation for not reading beyond the first sentence of an email. We would litter more complex emails with comments such as 'still here, boss?' "Hang in there boss" "Nearly done, boss" etc.
He wasn't unusual in that respect.
Once upon a time, "if I've made the effort to write this then you can make the effort to read it" was a thing. Not any more it seems. Now we're destroying the planet so ChatGPT can write long flowery e-mails and documents that nobody reads, or maybe so they can plug them into ChatGPT again to get a summary.
[1]ChatGPT at both ends
One employee says “AI turns this single bullet point into a long email I can pretend I wrote”.
Another says, “AI makes a single bullet point out of this long email I can pretend I read.”
[1] https://marketoonist.com/2023/03/ai-written-ai-read.html
Why would you use ChatGPT for anything important?
Study that used actual input to OpenAI’s chatbot finds personal use surging
This is why you never use hosted AI for sensitive information.
the paper explains that the researchers devised a method that meant no human read your work.
Oh, it's OK then... except the problem is their storing the input, not that someone is reading it. It's all fine until a curious employee decides to sift through it, they get hacked, or they get strong-armed for partial ownership and access to all stored data by the Trump administration. Because you KNOW they are going to need federal approval for something.
code snippets is where it shines
last Sunday I got it to write me some C# to decode some web scraped JSON , which it did instantly and worked straight away.
The previous time with much the same problem I spent an afternoon banging my head on the desk before resorting to going crawling to AI for help.
Devised a method that meant no human read your work.
Well, duh.
They just fed all your inputs into the LLM training maw. As they do anything they can get their hands on.
The only "method" they might have needed to "devise" was to take a pre-trained base and feed it all your chat logs as "specialised data" - the same thing the LLM companies try to flog you to create chatbots "personalised to your business practices".
At least this time they are being honest about whether a human ever gets involved to vet the inputs for, you know, accuracy/usefulness/sanity.
Seems...
like a business writing course might be cheaper and more planet friendly.
Or more radically, seriously invest more in schools, teaching and teachers.