Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Most ChatGPT use is non-work, focused on writing tasks.
- Claude is used more for automation, especially coding.
- AI adoption is uneven, with wealthier regions benefiting first.
Two of the biggest AI companies have revealed how people use their models.
Also: The best AI chatbots of 2025
OpenAI released a research paper analyzing millions of ChatGPT conversations, while Anthropic published its third Economic Index report on Claude usage. Both have plenty of charts and stats, but also a few interesting details about who’s using AI and why. Together, they provide a fascinating look at how AI is creeping into our work, school, and everyday life.
(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
Most ChatGPT chats are for asking about non-work stuff
OpenAI’s new study, How People Use ChatGPT, analyzed a random sample of conversations from May 2024 to June 2025. The first surprise: Non-work use has overtaken work use. In June 2024, about 47% of ChatGPT messages were work-related, but by June 2025, that dropped to 27% — even as daily messages grew from roughly 451 million to 2.6 billion over the same period.
The top categories now? Practical guidance (which includes tutoring, teaching, and how-tos), seeking information (about people, current events, products, and recipes), and writing (emails, documents, and other communications). These collectively account for “nearly 80% of all conversations,” according to OpenAI’s research. Coding is 4.2% of usage.
Also: The best AI for coding (including a new winner – and what not to use)
So, while the world (and your boss) may have imagined people using ChatGPT to fly through tedious work tasks or for computer programming, the majority are actually turning to it for help, advice, learning, and writing.
OpenAI’s researchers grouped user interactions into three buckets: Asking (seeking information or guidance), Doing (producing outputs or completing tasks), and Expressing (sharing opinions or feelings). They estimate about 49% of messages are Asking, 40% are Doing, and 11% are Expressing. As of July 2025, 56% of work messages were in the Doing category.
Also: Is ChatGPT Plus still worth $20 when the free version offers so much?
In fact, when ChatGPT is used at work, over 40% of that use is for writing — and more than two-thirds of that is editing text people wrote themselves. So when people turn to ChatGPT on the job, it’s more for proofreading than for fully generating text from scratch.
Most Claude chats are automation directives (for coding)
Anthropic’s Economic Index looks at how Claude is used by both consumers and companies. The report divides tasks into two categories: Automation (where Claude completes work with little input — either directive, with minimal interaction, or feedback loops, where users relay outcomes), and Augmentation (where humans and AI work together through learning, iteration, and validation).
The report analyzed randomly sampled transcripts and found that directive conversations — where a user asks Claude to complete a task with minimal interaction — rose sharply from 27% (in December 2024 to January 2025) to 39% (February to March 2025).
Also: The work tasks people use Claude AI for most, according to Anthropic
Since last December, people seem to be delegating entire tasks to Claude with little back-and-forth — telling it to do something and trusting the output. For the first time, Anthropic reports “automation (49.1%) has become more common than augmentation (47%).” It says this could be due to “AI winning users’ confidence, and becoming increasingly responsible for completing sophisticated work.”
In enterprise settings, specifically, automation is common. Anthropic says 77% of its API customer conversations have “automation patterns,” with the “vast majority” being directive. So, while about half of everyday users use Claude.ai to complete work with minimal interaction, businesses are relying on Claude even more to put tasks on autopilot — especially coding.
Anthropic said about 44% of API traffic is related to coding (computer and mathematical tasks), compared to 36% on Claude.ai. It said Claude.ai usage is higher for educational and writing tasks.
Who is using AI and where?
ChatGPT usage is booming worldwide, with 700 million weekly users as of July 2025. Over the past year, growth has been relatively faster in low- and middle-income countries, the report notes. While about 80% of early ChatGPT users were men, that share has since dropped to 48% (as of June 2025), with more active users now having “typically feminine” first names.
The user base also skews young, with nearly half of all messages from adults coming from people under 26. Researchers also said educated and highly paid professionals are “substantially more” likely to use ChatGPT for work.
Also: The one way millennials beat Gen Z in AI adoption
As for Claude usage, Anthropic said the US accounts for the highest share (21.6%). However, Israel leads when it comes to “global per capita Claude usage.” The nation has an “AI Usage Index” (AUI) score of 7, which means people in Israel use Claude about seven times more often than you’d expect based on the size of the country’s working-age population.
In the US, California is the leading state in terms of usage, making up 25.3% of all Claude activity, mostly for IT tasks. But when you adjust for population size, the per-capita-usage leader becomes Washington, DC, with an AUI score of 3.82. People there apparently use Claude nearly four times more often than their share of the US working-age population would suggest.
And just why are DC folks using Claude so much? “Document editing, information provision, and job applications,” Anthropic says.
Also: Most developers use AI in their daily workflows – but they don’t trust it
Raising concerns about inequality, Anthropic warned uneven AI adoption could mirror the history of electricity and the internet, where wealthy countries pulled ahead of developing ones. The report noted a “1% higher GDP per capita was associated with a 0.7% higher Anthropic AI Usage Index,” suggesting richer regions may already be capturing most of the productivity gains.
“People in higher-income countries are more likely to use Claude, more likely to seek collaboration rather than automation, and more likely to pursue a breadth of uses beyond coding,” Anthropic says. Countries with lower AI adoption per capita concentrate overwhelmingly on coding tasks, too — with over half of all usage in India, compared to roughly a third globally.
Common themes
Both OpenAI’s and Anthropic’s reports look at how people interact with AI: asking versus doing.
For example, Anthropic focused on whether Claude was used for automation or augmentation and found that users prefer automation and giving directives with little input, rather than collaborating with the model. OpenAI, meanwhile, broke user interactions into asking, doing, or expressing categories, and found that users tend to use it for asking.
Also: The fastest growing AI chatbot lately? It’s not ChatGPT or Gemini
Both studies also looked at the kinds of tasks people use AI for, such as writing or coding. While ChatGPT and Claude are still used for writing, ChatGPT users often ask the AI to edit non-work-related text they wrote themselves. Claude, meanwhile, is used more for automation, especially coding, with businesses directing it to complete tasks.
Add it all up, and you get an interesting snapshot of AI so far and how it differs between competitors.