This country brief, published on 14 July 2026 by Peter McCrory of Anthropic's Economics team, applies the Anthropic Economic Index to Canada, in English and in French: how much Canadians use Claude, where, and for what, compared with other countries. It is descriptive; it proposes explanations for its patterns but does not test them. The French text matches the English; the citation block on the page gives the French title and the date "2026-06".
Data and measures
The brief uses the sample of 1 million Claude.ai conversations from February 2026 (Figure 1 caption), the sample of the March 2026 index report, which dates it to 5 to 12 February 2026 and draws it from the Free, Pro and Max plans (Learning curves). In the released data, 25,902 of those conversations are assigned to Canada, and about 18% of the whole sample has no assigned country (this note's count). By July the index had moved to continuous sampling, with April and May 2026 data released on 26 June (Cadences); the brief does not use them. That release, on a changed pipeline, gives Canada 2.8% of usage and an AUI of 4.65 in April and 2.63% and 4.13 in May (this note's reading).
- The Anthropic AI Usage Index (AUI) divides a place's share of conversations by its share of working-age population. For countries, 1.0 is the world average; for provinces, Figures 3 and 4 label 1.0 as the Canadian average.
- The use case (work, coursework or personal) and request clusters are Claude's classifications of each conversation, as in the index reports. The brief does not describe how they were validated; the January 2026 report describes its check of such classifiers as one of directional accuracy (Economic primitives).
- Provincial characteristics come from Statistics Canada, country income from the IMF and World Bank (figure captions).
What it finds
High adoption, above what income predicts
Canada accounts for 2.6% of global Claude.ai traffic, eighth among countries, and has an AUI of 4.4 (section "Canada is at the forefront of Claude adoption"; Figure 1). Among the ten countries with the most usage, which together account for "more than half" of it (51% of all sampled conversations in the released data, or 62% of those with an assigned country; this note's sums), Canada's AUI is second to the United States' 4.6 (Figure 1, printed labels). Among IMF advanced economies with at least 200 conversations, usage per capita rises with GDP per working-age person, and Canada sits above the fitted line (Figure 2). In that figure Canada's AUI is below Singapore, Israel, Luxembourg, Switzerland and the United States (read from the figure), the same ordering as in How Australia Uses Claude: Findings from the Anthropic Economic Index. The brief suggests Canada's "highly educated workforce and proximity to the US technology frontier" as possible reasons.
Concentration in four provinces
Ontario has 43.9% of Canadian conversations, Quebec 20.8%, British Columbia 18.9% and Alberta 10.2%, together about 94%. Adjusted for working-age population, British Columbia (1.36) and Ontario (1.11) are above the Canadian average and the other provinces below it, down to 0.21 for Newfoundland and Labrador (Figure 3, printed labels). The territories are below the reporting threshold.
Figure 4 plots provincial AUI against GDP per working-age person, where the fitted line is nearly flat, and against the employment share of professional, scientific and technical services, where it rises steeply. The brief concludes that workforce composition rather than income is "a key determinant of within-Canada patterns of adoption," consistent with its earlier finding for US states, and that AI adoption in high-income countries "appears" to be shaped "primarily" by how well model capabilities match the local economy. The figure prints no coefficients, fit statistics or intervals, and the comparison rests on ten provinces.
Similar use-case mix across provinces
For the eight provinces in Figure 5, personal use is 44% to 51% of conversations, work about 34% to 40% and coursework the rest (Figure 5; the key findings give 13% to 18% for coursework). The brief describes this as "largely uncorrelated" with adoption rates. In the released data, Prince Edward Island's personal share is 40% and its coursework share and Newfoundland and Labrador's are suppressed, which may be why those two provinces are not shown (this note's reading).
Translation and public administration
Translation and editing requests are the largest share of conversations in New Brunswick (8.8%), Nova Scotia (8.8%) and Quebec (8.5%), the three provinces with the highest public administration employment (9.5%, 7.4% and 6.3%), for the seven provinces above the threshold (Figure 6). The brief links this to official bilingualism in federal services, "perhaps" and "likely" in its own words. The cluster behind Figure 6 matches the released cluster "Assist with translation, writing, editing, and document formatting tasks," which covers more than translation; on it, the US share (7.6%) is above Canada's (6.6%) (this note's reading of the release).
A tilt toward education and job seeking
Against the average of Australia, the UK and the US, for clusters with at least 1% of Canadian conversations, document translation is the most overrepresented use in Canada, followed by mathematics, academic assignments, STEM coursework, résumés and programming help; workplace email is the most underrepresented, followed by marketing, legal assistance, recipes, home maintenance and medical information (Figure 7). Figure 7's bars are differences of under one percentage point; its printed percentages match Canada's own shares in the released data, not the differences (this note's reading). Document translation (the released cluster "Translate and format diverse professional, academic, medical, and religious content between languages") is 1.2% of Canadian conversations against 0.5% in Australia and the UK and 0.4% in the US, and 2.9% in Quebec; workplace email is 3.3% against 3.6% to 4.4% in the peers.
Some of the coursework difference is a difference from Australia. Academic assignments are 5.1% of Canadian conversations, the same as the US (5.1%) and above the UK (4.8%) and Australia (4.1%) (this note's reading of the release). The February sample week fell in the Australian summer break, which this collection's note on How Australia Uses Claude: Findings from the Anthropic Economic Index raises as a likely reason for Australia's low coursework share (the Australia brief itself does not discuss timing). Résumés and cover letters (2.1% in Canada against 1.6% to 1.7% in the peers) are above all three.
Limitations and open questions
- Claude.ai only, one week. API usage is excluded and all figures come from 5 to 12 February 2026. The AUI is a ratio of conversation shares; it does not count users or firms, or other AI products. The brief reports no uncertainty for its comparisons.
- Small units. The provincial comparisons rest on ten provinces, eight or seven of them in some figures; Prince Edward Island and Newfoundland and Labrador contribute 52 and 67 conversations (this note's count).
- Explanations are proposed, not tested. Workforce composition and bilingualism are offered as reasons; no regression with controls is reported.
- Model-made categories. Use cases and request clusters are Claude's classifications, and cluster names define what "translation" covers.
- Interest. The brief is a model developer's report on use of its own product, published in both official languages.
Research questions
- Does professional-services employment predict provincial adoption once other industries are controlled? A regression on census employment shares across provinces, and across Canadian metropolitan areas if samples allow, would test the brief's main explanation.
- Is translation demand driven by public-sector work? Separating translation from writing and editing, and comparing French-English pairs with other language pairs by province, would show whether bilingual public-service requirements account for the pattern.
- How much of Canada's coursework tilt depends on timing? Comparing Canada with Southern and Northern Hemisphere peers in samples taken at matched points of their academic years would answer it.
- Has Canada's AUI changed with the sample? The September 2025 report gave Canada 2.91 on its August 2025 sample (see below), and the June 2026 release 4.65 and 4.13 for April and May 2026; a series on one pipeline, with intervals, would show how much of the movement is sampling, method or use.
- Do the same patterns hold for API and enterprise use? Business usage is outside the brief's sample.
How it connects
- The brief cites Economic primitives (January 2026), which introduced its use-case classifier, for the finding on adoption within the United States. Its sample is that of Learning curves (March 2026), which the brief does not cite in its text (the page lists it only as a related article).
- On the August 2025 sample, Anthropic Economic Index report: Uneven geographic and enterprise AI adoption gave Canada an AUI of 2.91, below the US (3.62) and Australia (4.10) (its Figure 2.2); this note connects the two, which the brief does not.
- This note connects the brief to the two earlier country briefs: How Australia Uses Claude: Findings from the Anthropic Economic Index (March 2026, same sample, reporting a similar lack of an income relationship across Australian states) and India Country Brief: The Anthropic Economic Index (February 2026, November 2025 sample).
- This note also connects the brief to Cadences (June 2026), which describes the pipeline that replaced the weekly sample used here.
References
- McCrory, P. (2026, July 14). How Canada uses Claude: Findings from the Anthropic Economic Index (with French text, Utilisation de Claude au Canada : constatations présentées dans l'Anthropic Economic Index). Anthropic. https://www.anthropic.com/research/how-canada-uses-claude. record
- Massenkoff, M., Lyubich, E., McCrory, P., Appel, R., and Heller, R. (2026, March 24). Anthropic Economic Index report: Learning curves. Anthropic. https://www.anthropic.com/research/economic-index-march-2026-report. record · 7 pages
- Anthropic (2026). EconomicIndex dataset, release of 24 March 2026 (5 to 12 February 2026), Claude.ai file. https://huggingface.co/datasets/Anthropic/EconomicIndex. record