How the World Uses Claude: Inside Anthropic’s Economic Index

The Claude Economic Index: Decoding Global AI Behavior to Spot Your Next Big Market Gap

Image created by the author using AI

Anthropic publishes something most AI companies never share: a running record of how millions of people actually use their model. It’s called the Anthropic Economic Index, and its most recent editions paint a detailed picture of how Claude fits into work, school, and everyday life across the globe.

The numbers are worth paying attention to if you’re curious about where AI adoption is heading, not just in Silicon Valley, but in households and offices worldwide. Below is a clear breakdown of what the data actually shows, drawn directly from Anthropic’s published reports.

Work Still Leads, But Personal Use Is Catching Up

As of November 2025, Claude.ai conversations broke down as 46 percent work, 35 percent personal, and 19 percent coursework, according to Anthropic’s Economic Index. Work remains the single biggest category, but the split shifts depending on where you live.

Higher-income countries show more personal use of Claude, treating it more like an everyday assistant for planning, writing, and casual questions. Lower-income countries lean harder into coursework, using it to support schooling and skill-building. Anthropic frames this as a natural adoption curve: early users in less developed markets tend to be technical specialists or students, while usage broadens into casual, personal territory as a market matures.

Coding Dominates, But the Gap Is Narrowing

Computer and mathematical tasks, mostly software development, are still the single largest category of Claude usage. But their share has actually fallen on Claude.ai, from a peak of 40 percent in March 2025 down to 34 percent by November 2025.

Meanwhile, education-related use has grown steadily, climbing from 9 percent of Claude.ai conversations in January 2025 to 15 percent in November 2025. This covers coursework help, tutoring, and building instructional materials. Writing and design tasks, like copyediting and fiction, have picked up too. The overall pattern is a platform that started coding-heavy and is steadily diversifying into education, writing, and everyday problem-solving.

On the other side, Anthropic’s first-party API traffic (the developer and business-facing side of Claude) looks different. It skews far more toward computer and mathematical tasks, at 46 percent, and office and administrative work like email drafting and scheduling has grown to 13 percent. That reflects businesses automating routine back-office workflows rather than individuals exploring the tool.

Where in the World Claude Gets Used Most

The United States, India, Japan, the UK, and South Korea lead in total Claude.ai usage, but raw volume only tells part of the story. Anthropic also tracks something called the Anthropic AI Usage Index (AUI), which measures whether a country uses Claude more or less than its working-age population would predict.

Denmark, for example, has an AUI of 2.1, meaning its residents use Claude at roughly twice the rate you’d expect from population size alone. Globally, usage is strongly tied to income: a 1 percent increase in a country’s GDP per capita is associated with a 0.7 percent increase in Claude usage per capita. Wealthier nations also tend to use Claude more collaboratively, treating it as an assistant to work alongside, while usage in lower-income countries skews more toward direct, single-shot requests.

That gap has held steady rather than closing. Anthropic’s data shows little evidence that lower-usage countries are catching up to higher-usage ones globally, even as usage within the US has become more evenly spread across states.

How People Actually Work With Claude

Anthropic separates Claude usage into two broad styles: augmentation (collaborating with Claude, iterating, asking follow-ups) and automation (handing Claude a task and letting it run with minimal back and forth).

This balance has shifted over time. In January 2025, augmentation was clearly dominant at 56 percent versus 41 percent automation. By August 2025, automation had briefly overtaken it. Then, by November 2025, augmentation bounced back to 52 percent, edging out automation’s 45 percent, coinciding with the rollout of features like file creation, persistent memory, and Skills.

In short, people aren’t just delegating more and more to Claude over time. Product changes that make collaboration easier appear to pull usage back toward a more hands-on style, at least for now.

What People Actually Walk Away With

Anthropic’s most recent report introduced a new way of looking at usage: what each conversation actually produces. Their classifier found that 93 percent of Claude conversations result in a concrete output, called an artifact, and the most common ones are explanations (17 percent), documents and reports (15 percent), and guidance (11 percent).

Some outputs are almost entirely personal. More than 80 percent of conversations that produce creative writing, guidance, or recipes are personal rather than work-related. Others lean hard toward work: creating marketing content and writing database queries are both work-related more than 80 percent of the time.

Usage also has a rhythm. Personal conversations spike from about 35 percent of all traffic on weekdays to just under 50 percent on weekends. Recipe requests jump 2.3 times above average around 6 p.m. local time. And in the US specifically, tax-related conversations spiked to eight times their normal rate right before the April 15 filing deadline.

How People Feel About It

Beyond raw usage, Anthropic surveyed roughly 9,700 linked Claude users about their experience. The majority reported real productivity gains: 86 percent noted faster work, 82 percent said it expanded the scope of what they could take on, and 69 percent said it improved quality. Around 27 percent said it saved them money they’d otherwise spend on outside services.

Most respondents also felt the tool was making them more capable rather than replacing them. 68 percent said they were learning more because of AI, and 57 percent felt it was making their existing skills more valuable. That said, the survey also flagged real anxiety, particularly among early-career workers and those in the most AI-exposed roles, around what widespread AI adoption might mean for job security over the next year.

One more pattern worth noting: usage differs by gender. Women, who made up just 12 percent of the linked survey sample, tend to use Claude more iteratively and spend more active time per session, while showing a lower share of fully automated, hand-it-off usage compared to men, even after accounting for differences in occupation.

A Few Honest Caveats

It’s worth being direct about what this data can and can’t tell you. Anthropic’s figures come from privacy-preserving classifiers, meaning another instance of Claude reads and categorizes anonymized conversations rather than a human analyst. That approach is scalable, but it means individual classifications carry some noise, even if the overall patterns hold up.

The survey data also isn’t representative of the general population. It reaches a sample of existing Claude users, skews toward computer, mathematical, and management occupations, and likely over-represents people already comfortable experimenting with AI. Treat the country-level and demographic patterns as directional signals about where AI adoption is heading, not precise, final numbers.

Key Takeaway

Put together, the picture is less about AI replacing entire jobs overnight and more about a tool that’s diversifying fast: still coding-heavy, but increasingly used for teaching, writing, planning, and daily life, with real gaps between how wealthy and developing economies put it to work.

If you’re building a business, a side project, or just your own workflow around AI tools, understanding these usage patterns is a genuinely useful starting point. It shows you where the crowd already is, and where there’s still room to do something different.

If you’re exploring which AI tools are worth building into your own workflow, it’s worth spending some time testing a few options directly rather than going off usage stats alone. What works best often comes down to the specific tasks you’re trying to solve.