Claude’s Heavy Users in Australasia

Nikita Melashchenko · (26 July 2026) · opinion

Anthropic has done something useful with its Economic Index, the research programme that tracks how Claude is actually used. The dataset behind the reports is now queryable directly from inside Claude, through the Anthropic Economic Index connector, one of the MCP connectors that can be switched on in Claude’s settings. Enable it and the dataset answers questions in plain language. Every number in this post comes from interrogating it that way, from the release covering the period ending 1 May 2026 across 121 countries.

First, what the index actually measures. A country’s usage index is its share of global Claude usage divided by its share of the world’s working-age population. A score of 1.0 means usage in line with population, and 2.0 means twice that. The data behind it are anonymised conversations matched to occupational tasks, so the index can say Claude is used for editing tasks, not that editors are the ones using it. And it is a snapshot of a single period, so it says nothing about trends or jobs. What it offers is a profile, and the profile down here is quite interesting.

Australia sits first of 121 countries at 6.4, New Zealand fifth at 4.84, with Singapore, Switzerland and Luxembourg in between. Note that the index mechanically flatters small working-age populations. Luxembourg, Iceland and Malta in the top ten make the point for me. In absolute terms Australia accounts for 2.65 per cent of global usage and New Zealand for 0.39 per cent. Still, these differences are there. Another interesting question is what Australia and New Zealand are actually doing with Claude.

The two profiles are nearly identical, which is itself a finding, since the index treats them as entirely separate markets. Work sits at the global average in both. The deviation is coursework, at 20.7 per cent in Australia and 19.6 per cent in New Zealand against 16.5 globally. Both countries also lean slightly more towards augmentation, where the human stays in the loop, at 54.5 and 53.7 per cent respectively against 51.4 globally. Universities are where a disproportionate share of Australasian usage is being formed, which is exactly where the questions I raised in the answerability debate will have to be somehow decided first.

The topic mix reveals the regional character. Content creation leads everywhere, but education and research run above the global baseline in both countries while software development runs well below it, at 8.3 per cent in Australia and 9.3 in New Zealand against 11.5 globally.

The task-level data provide more food for thought. The tasks most over-represented against their baselines are editing and rewriting other people’s material, running a third to a half above the global norm. Matched against occupations, the leading profiles are librarians, editors, writers and support specialists. Taiwan, for comparison, sends 27.9 per cent of its usage to computer and mathematical tasks. Australasia uses Claude as a reference desk and an editing suite. For two services-heavy, English-speaking economies whose exports lean on documents rather than software deployments, that is an expected profile.

Important to note, this public statistics on AI use in New Zealand comes from a US lab’s telemetry of its own product. No official agency publishes anything at this grain, and while the government drafts its AI strategy, the sharpest evidence of what New Zealanders actually do with these systems sits in a vendor’s dataset, generously but voluntarily disclosed. None of it is a user count either. Anthropic publishes no headcounts.

The asymmetry has a second edge. Fiji, Papua New Guinea, Samoa, Tonga and the rest of the Pacific states appear nowhere in the 121 countries. Apparently, this is not because usage there is zero but because it falls below the publication threshold.

I wrote earlier this year that trade policy here is made on the edge of the map. This index redraws the map with Australasia at the very top and our Pacific neighbours off the edge entirely. Both results are a reflection of who is doing the measuring and at what threshold. Perhaps, it is worth considering to introduce a reporting requirement and standards for AI usage disclosures by AI vendors. This should ease the asymmetry and provide a more complete picture of AI adoption across the region.