Methods

The primary source is the Anthropic Economic Index release of 26 June 2026 (Anthropic, 2026; Massenkoff et al., 2026), pinned to revision 2ea58ff75e4247d26810c37f10c179edc2466cac. April and May 2026 country consumer observations supply the direct occupation and overall-usage fields. The fixed task catalogue is O*NET 30.2. Source release and documentation.

A conversation is assigned an occupational activity by the publisher's classifier. The assigned activity is not evidence of the user's occupation. Country evidence concerns the documented consumer surface; the first-party API file has no country dimension and is not allocated to countries here.

Historical task observations cover one week in August 2025, November 2025 and February 2026. The April and May 2026 observations cover calendar months. Changes in sampling and classification mean that the five windows are retained as distinct observations rather than a continuous adoption series.

Let $U_{cm}$ be geography $c$'s published percentage share of global consumer usage in month $m$, and let $P_c$ be its population aged 15–64 in the publisher's reference. Conceptually, the relative per-capita index is

The analysis uses the publisher's released index. A value above one denotes more platform use per working-age resident than the reference average. It is not the fraction of residents who use AI.

The primary measure selects the detailed soc_occupation facet, hierarchy level 0 and metric pct. For a published row for occupation $o$,

Its denominator is the geography's consumer usage in that window. Changing geography does not convert an O*NET category into a local occupational classification.

Let $\mathcal{P}_{cm}$ be the set of published occupation rows. Retained occupation mass is

The sum is calculated before mapping and is not rescaled to one. The residual may contain unpublished or unclassified activity and rounding; the public tables do not identify the contribution of each source.

Missingness rule

An unpublished row has unknown actual use. A published rounded zero is a different state. This rule applies to every explorer, comparison and download.

For published-mass bookkeeping, an unpublished row contributes zero published mass. This is not an imputation of zero actual use. The source's two-decimal rounding permits an arithmetic range of up to 0.005 percentage points per numeric row; that range is not a sampling interval.

Let $T_o$ be the fixed set of catalogue tasks associated with occupation $o$, and let $s_{cmt}$ be a task's positive published share in fractional units. Catalogue task coverage is

Let $\mathcal{P}^{\mathrm{task}}_{cm}$ be the set of published task rows. Task-share intensity is

Coverage measures published breadth in a catalogue. Intensity divides released task mass by catalogue size. Because a task may belong to several occupations, neither measure can be added across occupations to recover national usage. Their denominators are neither employment nor working time.

The typed ESCO–O*NET crosswalk contains exact, narrow and broad relations. Alternative routes remain separate. The source library contains 4,253 typed links, including 498 exact links; link counts do not measure the share of employment with an exact correspondence. European Commission crosswalk.

For route $a$, let $D_e^{(a)}$ be the available O*NET donors for ESCO occupation $e$. The ESCO descriptor is

Here $S^{\mathrm{published}}_{cmo}$ equals the source share when a row exists and zero published mass for an absent row within a supported geography-month. The donor set includes all linked O*NET catalogue occupations with a defined published-mass value, including donors with no published row. Donors outside the available catalogue are excluded. This convention measures published mass; actual use for an unpublished donor remains unknown.

Donor-mean rule

A mean of linked source shares describes those donors; it does not allocate their total across destinations. Adding donor means or dividing them by European employment shares does not produce a national usage distribution.

ISCO four-digit descriptors are equal means of represented ESCO occupations. Coarser ISCO groups are direct means over represented four-digit groups. These values follow the donor-mean rule. Alternative routes are sensitivity scenarios, not confidence intervals.

Europe is the detailed mapping case. The common international allocation is also supplied for other countries, where transfer of the same semantic links remains an explicit assumption. Local classifications and expert validation can refine that assumption.

Employment context preserves the source population, classification, year and quality flags. Eurostat and ILOSTAT commonly provide ISCO groups; UK APS uses SOC 2020. The ONS coding-index route uses lexical weights, which are not observed worker transitions.

For the United States, O*NET children aggregate directly to native six-digit SOC codes: occupation shares are summed and task descriptors are averaged. Shares are summed at their original integer rounding precision before conversion to fractions. This exact code route makes the United States a useful benchmark because it avoids the European bridge.

The BLS National Employment Matrix supplies 2025 base-year employment, including unincorporated self-employment. It counts jobs rather than workers. Its official total is 170,280,800 jobs; 772 matched detailed groups cover 92.53% of that total. The 2035 projections are not outcomes in this study. BLS definitions.

For a native group with employment $E_o$ and official national total $E_{\mathrm{national}}$, the usage-to-employment ratio is

A value above one means that the group's share of classified consumer use exceeds its share of national jobs. It does not estimate a worker's probability of using AI. Unmatched employment remains in the national denominator and is reported separately.

The primary regional share construction allocates each published source occupation across destination groups, with an unmatched category:

For the primary equal_all scenario, the pipeline retains all typed links, collapses duplicate links to the same ISCO code, and divides each source equally across its distinct finest known destinations. It then sums into broader groups. Splits are fixed across countries and months; they are never re-estimated from employment or outcomes. Some official links resolve only to three digits. They remain unresolved at four-digit detail and are not expanded into invented children.

The UK continues from ISCO4 to SOC2020 using the reverse ONS coding-index relation: each ISCO source is split equally across its distinct UK destinations. The lexical_all sensitivity uses title counts normalized within the ISCO source. Those lexical proportions are not measured worker transition probabilities. Forward SOC-to-ISCO weights are not reversed by relabelling columns.

exact_narrow and exact_only scenarios restrict the candidate link set before allocating. Excluded or unresolved source categories go to UNMATCHED; they do not disappear and their remaining shares are not rescaled. The allocation table retains all official named destination groups, including groups with no published donors. Their zero published-mass bookkeeping is distinct from unknown actual use; employment ratios are withheld where no published donor exists.

Accounting and mapping bounds

Allocated mass plus unmatched mass equals the original published occupation mass. The separate residual outside the published total is never assigned to occupations. The published source rounding is retained; floating-point conservation error is below $10^{-9}$ percentage points across all country-month, detail and scenario combinations.

Let $C_o^{(a)}$ be the destinations reachable from source $o$ at the reporting level, including unmatched where applicable. Under arbitrary source-level splits across that set, the marginal allocation bounds are

These are sharp marginal bounds conditional on the published values and permitted links. They are not sampling confidence intervals, do not include unpublished activity, and do not validate the semantic links. Marginal extrema need not be jointly attainable. A conservative envelope for total activity would additionally have to consider the unallocated publication residual; that quantity is not supplied as an observed occupation estimate.

Allocated usage relative to employment

For compatible official group employment $E_{cg}$ and the full national total $E_c$, the descriptive concentration ratio is

The numerator uses consumer activity and the denominator official employment, with the actual population and date retained. A ratio above one means a larger allocated activity share than employment share. It is neither an adoption probability nor a causal effect. Unmatched employment remains in the official total. Source flags remain visible; missing or nonpositive employment and absent published donor support suppress the ratio. Country coverage or taxonomy detail alone does not establish comparability.

Explore allocations and employment · Occupation allocation weights · Allocation accounting checks.

Frozen inputs, unique observation keys, source precision and publication states are retained in the outputs. Comparison tables report common occupational samples. Source attribution and redistribution rules accompany downloads. The technical appendices record construction and file-level lineage, while validation distinguishes successful data checks from the unpassed exposure reconstruction.