This short brief, published on 31 March 2026 by Peter McCrory of Anthropic's Economic Research team, applies the Anthropic Economic Index to one country. Its stated occasion is Anthropic's expansion to Australia: a new office in Sydney and a Memorandum of Understanding with the Australian government on AI safety research and the goals of Australia's National AI Plan (introduction). The brief uses the February 2026 sample of 1 million Claude.ai conversations (Figure 1 caption) and compares Australia with other countries on the index's standard measures. It is a descriptive report with no causal analysis.
Measures used
The brief relies on measures defined in earlier index reports:
- The Anthropic AI Usage Index (AUI) compares a geography's Claude usage with what its working-age population would predict, indexed so that 1.0 represents parity (Figure 3 caption); an AUI of 4.1 means usage over four times the expected level.
- The economic primitives, introduced in the January 2026 report, describe conversations by use case (work, coursework or personal), the autonomy given to Claude on a 1 to 5 scale, task success, and task complexity, measured both by the years of schooling needed to understand a prompt and by the estimated time a skilled professional would need without AI (section "Australia's use of Claude resembles its Anglosphere peers").
- Tasks are classified using O*NET, a US taxonomy of job tasks, and grouped into SOC major groups, the broad occupational categories of the US Bureau of Labor Statistics.
What it finds
High adoption, concentrated in two states
Australia accounts for 1.6% of global Claude.ai traffic, eleventh among countries, and has an AUI of 4.1, seventh in per-capita adoption behind Singapore, Israel, Luxembourg, Switzerland, the United States and Canada (section "High adoption overall, unevenly distributed across states and territories"; Figure 1).
Within Australia, New South Wales accounts for 37.2% of conversations, Victoria 30.8%, Queensland 17.7%, and the remaining states and territories 14% combined (Figure 2 and caption). Adjusted for working-age population, only New South Wales (AUI 1.20) and Victoria (1.19) are above 1; adoption is lowest in Western Australia (0.68), Tasmania (0.32) and the Northern Territory (0.12).
Unlike the strong relationship between income and per-capita usage across countries, income does not predict adoption across Australian states and territories (Figure 3). The brief notes that with only eight units this is "suggestive rather than conclusive." It attributes the pattern to workforce composition: Western Australia, mining-heavy and with the highest gross state product per capita, and the Northern Territory both have low usage, plausibly because fewer workers are in occupations where Claude is used; the Australian Capital Territory has above-average income but lower-than-expected usage, which the brief suggests may reflect barriers in its large public-sector workforce; and New South Wales and Victoria, with slightly below-average income, have more workers in finance, professional services and technology.
A usage profile like other English-speaking countries
On the primitives, Australia resembles the other Anglosphere countries in the comparison (the US, UK, Canada, New Zealand and Ireland; Figure 4 caption). Of Australian conversations, 46% are work-related, near the median; 7% are coursework, toward the bottom of the distribution and below the US, UK and Canada, while countries with lower per-capita adoption have coursework shares two to three times as high; and 47% are personal, near the top. The brief links this to the cross-country pattern from the January report, in which richer countries have lower coursework and higher personal shares.
Australia's AI autonomy score is 3.38 on the 1 to 5 scale, toward the low end and close to other Anglosphere countries, which the brief reads as more collaborative, less delegated use, as in other high-adoption economies. On complexity the two measures point different ways: the average Australian task would take a skilled professional an estimated 2.7 hours without AI, against a cross-country average of 3.3 hours (the key findings call this "roughly 20% less"), but Australian prompts need an estimated 11.9 years of schooling to understand, comparable to other Anglosphere countries and above the global median. Both patterns, the brief says, match cross-country relationships documented in the January report.
A broader task mix with less coding
Australian usage is more diverse than that of its peers and the world as a whole (section "In Australia, a broader task mix with a smaller coding share"). Its 100 most common O*NET tasks account for 47.3% of its usage, compared with 47.7% for the US, 48.3% for the UK, 50.2% for Canada and 52.3% globally (Figure 5 and text).
The difference comes mostly from coding. Computer and Mathematical tasks are 8.0 percentage points lower in Australia's top 100 tasks than in the global top 100, and education-related tasks nearly 3 points lower; the offsetting gains are spread across groups, led by Management (+2.3 points), Office and Administrative Support (+1.3) and Life, Physical and Social Science (+1.3) (Figure 6 and text). By request cluster, general coding assistance is 13.5% of Australian use against 16.8% globally, and document translation is also underrepresented, which the brief attributes to Australia being mainly English-speaking. The most overrepresented clusters include personal life management (+1.9 points), health and well-being support (+1.8), workplace correspondence (+1.7), business documents (+1.6) and financial guidance (+1.3) (Figure 7 and text).
The brief notes that a lower coding share is expected in higher-adoption countries: Australia's 8.0-point shortfall is close to the Anglosphere average of 8.9 points. What differs is what fills the gap. Australia leans further into Management (+2.3 points against +1.2 for its peers) and Office and Administrative Support (+1.3 against +0.1), and away from Educational Instruction (−2.7 against +1.6).
Limitations and open questions
- Claude.ai only. The analysis uses consumer Claude.ai conversations; API usage is not included.
- Small subnational samples. The state comparison rests on eight units, as the brief notes, and Tasmania and the Northern Territory together contribute under 1% of Australian conversations.
- Explanations are proposed, not tested. Workforce composition and public-sector barriers are offered as likely reasons for the state pattern; the brief does not measure them.
- One sample period. All figures are from the February 2026 sample.
Research questions
- Does state-level adoption track the share of workers in high-usage occupations? The brief proposes workforce composition as the explanation. Regressing state AUI on employment shares in finance, professional services and technology from census data would test it.
- Is public-sector work a barrier to adoption? The Australian Capital Territory is the brief's example. Comparing usage across regions with different public-sector shares, in Australia and elsewhere, would test the suggestion.
- Why does Australia shift toward management and office work rather than education? Comparing occupational structure and AI policy in schools and universities across Anglosphere countries would help explain the different composition of the non-coding share.
- Is the low coursework share stable through the year? The brief uses a single February 2026 sample. Repeating the comparison with samples from different points in the Australian academic year would show how much of the 7% figure depends on timing.
- Do the same patterns hold in API usage? Applying the analysis to Australian API traffic, if sample sizes allow, would show whether businesses use Claude differently from consumers there.
How it connects
- The brief's framework is from the January 2026 index report, which it calls the fourth Economic Index report and cites for the four economic primitives and the cross-country relationships (Anthropic Economic Index report: Economic primitives).
- The cross-country relationship between income and usage is attributed to the September 2025 report (Anthropic Economic Index: Tracking AI’s role in the US and global economy).
References
- McCrory, P. (2026, March 31). How Australia Uses Claude: Findings from the Anthropic Economic Index. Anthropic. https://www.anthropic.com/research/how-australia-uses-claude. The citation block on the page gives the URL https://www.anthropic.com/research/australia-brief-economic-index-march-2026.
- Data sources named in the Figure 3 caption: Anthropic Economic Index, February 2026; Australian Bureau of Statistics, Estimated Resident Population, June quarter 2025; ABS, Australian National Accounts: State Accounts, 2024–25.