Skip to main content
News Plus 7 Jul 2026  - 6 min read

Australia ranks first among 121 countries on Anthropic's global usage index; 6.4x expected rate; less developer-led, more documents, education and knowledge work; humans in the loop

By Andrew Birmingham - Martech | Ecom | CX editor

Australia has become one of the world’s most intensive markets for Anthropic’s Claude, with new company data showing local usage at 6.4 times the level its population would predict and tilted less toward coding than documents, education, content and general knowledge work. The findings suggest generative AI is moving quickly into everyday white-collar routines, but not always as a fully automated substitute for workers. Anthropic’s data shows Australian users are more likely than the global average to keep humans in the loop, while its broader Economic Index points to a deeper challenge for business: Claude.ai, Cowork and API traffic behave like different products, with different cost, governance and risk profiles.

What you need to know

  • Australia is Anthropic’s most intensive Claude market, with May usage running at 6.4 times the level expected based on working-age population and ranking first across 121 countries on usage per capita.
  • Local adoption is not being led by developers. Australia’s software development share is below the global average, while use skews toward documents, education, content, research, business operations and broader knowledge work.
  • Australians are also more hands-on with AI. Anthropic says 54.5 per cent of local Claude.ai and Cowork usage is “augmentation”, above the 51.4 per cent global average, meaning people are using Claude as a collaborator rather than fully delegating tasks.
  • For Australian businesses, the governance challenge is shifting from managing AI as a technical tool to setting controls for everyday knowledge work, including documents, email, research, copywriting and business support.

 

There are many different narratives about AI usage, what stands out to me more than the ranking is how we're using it. 54.5% of Australian usage is "augmentation", people staying in the loop, using Claude as a collaborator rather than handing off the whole task.

— Theo Hourmouzis, Anthropic General Manager/Head of ANZ & ASEAN.

Australia has emerged as one of the world’s most intensive markets for Anthropic’s Claude, with new company data showing local usage far above what population modelling would predict and tilted less toward software development than documents, education, content and broader knowledge work.

The findings, drawn from Anthropic’s latest Economic Index report and its June 26 data release, suggest Australia is adopting Claude at unusual scale while using it in a more hands-on way than the global average. The data also points to a wider challenge for companies trying to govern artificial intelligence: Claude.ai, Cowork and first-party API traffic behave less like different access points to the same tool than distinct systems with different work patterns, levels of automation and oversight needs.

“Australia just topped the latest Anthropic Economic Index, ranked #1 of 121 countries for Claude.ai usage, by no small margin either. We're using Claude at 6.4x the rate our economic modelling predicts would be expected for our population,” said Theo Hourmouzis, General Manager/Head of ANZ & ASEAN for Anthropic.

“There are many different narratives about AI usage, what stands out to me more than the ranking is how we're using it. 54.5% of Australian usage is "augmentation", people staying in the loop, using Claude as a collaborator rather than handing off the whole task. That's above the global average of 51.4%.”

In May, Australia accounted for 2.65% of global Claude.ai and Cowork usage, up from 2.29% in April. It ranked ninth globally by total usage share and first on Anthropic’s usage-per-capita index, which compares a country’s share of Claude usage with its share of the working-age population. On that measure, Australia scored 6.40 in May, meaning usage was 6.4 times the level Anthropic’s modelling would expect based on population.

The Australian pattern is commercially significant because it does not resemble a purely developer-led adoption curve. In May, Australia’s software development share was 8.32%, below the global Claude.ai and Cowork average of 11.51%. Its usage was more heavily weighted toward content creation and copywriting, education and learning, research and intelligence, hobbies and lifestyle, business operations, and data analysis.

Australian Claude.ai and Cowork usage was also more document-heavy than the global average. Documents and reports accounted for 17.94% of Australian artifacts, compared with 14.91% globally. Email and message outputs were also higher in Australia, while code fixes, debugging, apps and websites were lower than the global average.

That gives the Australian data a different complexion from the common view of generative AI adoption as primarily led by programmers and technical teams. It suggests Claude is diffusing through white-collar and knowledge-work routines, including writing, study, planning, research, correspondence and business support.

The pattern also broadly matches Anthropic’s broader Cadences report, which argues that Claude use has shifted from simple back-and-forth chat toward longer-running work that produces concrete outputs such as explanations, documents, reports, code, advice, plans and summaries. The report says 93% of Claude conversations now produce an artifact, meaning a primary output that can range from a written document to a website, spreadsheet or email.

Augmentation

For Australian executives, the more important finding may be how those outputs are produced. Anthropic classifies Claude use as augmentation when the human remains more involved and automation when the system is asked to complete more of the task. In Australia, 54.5% of Claude.ai and Cowork usage was classified as augmentation in May, compared with 51.4% globally. Automation stood at 45.5%, below the global average of 48.6%.

“We’re leading the way with AI adoption and we’re adopting it in a more hands-on way than the rest of the world. My conversations with local business leaders match these very data points,” Hourmouzis said.

The data does not mean Australian users are avoiding AI delegation altogether. Anthropic’s autonomy score for Australia was 2.72 on a five-point scale, almost identical to the global Claude.ai and Cowork average of 2.74. But the augmentation figure suggests Australian users are more likely than the global average to keep humans actively involved in the task.

That distinction matters because Anthropic’s global data shows AI usage cannot be treated as one uniform category. Claude.ai and Cowork, the company’s consumer-facing and collaborative work surfaces, have a broad mix of personal, work and coursework use. In May, global Claude.ai and Cowork usage was 43.36% work-related, 40.20% personal and 16.45% coursework.

First-party API traffic looked very different. It was 84.03% work-related and 94.22% automated, according to the data release, but its mean AI autonomy score was lower than Claude.ai and Cowork. API traffic also involved much less estimated human-with-AI time and was concentrated more heavily in analysis, summaries, data and spreadsheet outputs.

The split shows why businesses trying to control AI cost, quality and risk may need to govern by product surface and task type rather than treating all model usage as equivalent. A chat-based document, an API-driven data process, a coding session and a collaborative research workflow can all count as Claude usage, but they involve different levels of human involvement, review burden and operating risk.

Anthropic’s report also ties those differences to cost. It says token consumption, the amount of text processed and generated by a model, tends to rise with the complexity and value of the work. Conversations mapped to higher-wage occupations consumed more tokens, and more complex artifacts such as apps and websites used far more compute than simple explanations.

That finding gives the Australian data a sharper business edge, suggesting that if local usage is concentrated in documents, education, research, copywriting and business workflows, the near-term governance task is less about managing a niche developer tool and more about setting controls for everyday knowledge work.

“Beneath that headline there’s a lot of substance worth digging into in our latest Economic Index, Cadences. This is the 6th edition of our sustained effort to deliver longitudinal measurements on how AI is diffusing in the economy and the workplace,” Hourmouzis said.

Anthropic’s data has limits. The Economic Index is based on Anthropic’s own systems and classifiers, and its survey of Claude users is not representative of the general workforce.  Even so, the Australian data suggests Claude has already moved beyond experimental use among a narrow technical audience. It is being pulled into the work of writing, learning, researching, planning and communicating, with humans more often staying in the loop than handing off the task entirely.

What do you think?

Search Mi3 Articles