Measurement glossary

The observations for which both quantities in a comparison are available. Correlations based on different common-support samples are not directly comparable without inspecting those samples.

The published share of a geography's consumer usage assigned to a detailed soc_occupation row. It retains the source denominator and classification.

An equal mean of values from source occupations linked to a destination occupation. It describes represented source categories. See the donor-mean rule.

A dimension in the source release, such as task or occupation, through which the same underlying activity is classified and published.

A four-digit unit group in the International Standard Classification of Occupations. Coarser ISCO1 and ISCO2 groups contain broader occupational families.

A comparison that uses the source classification without an international crosswalk. The US SOC route is the native benchmark in this study.

No source row is present for the geography, period and category. See the missingness rule.

The sum of source shares that appear in a released facet, using its original country denominator. Its complement can include unpublished or unclassified activity and rounding. Also called retained mass or publication mass. Occupation mass and task mass refer to different facets.

A source row exists but its displayed share is zero at the source precision. It remains distinct from an absent row.

A reporting summary of observable publication breadth, published occupation mass, repeat observation and employment support. It is not a validated measure of representativeness or statistical quality.

The fraction of a fixed occupation task catalogue with a positive published row in a geography and period. It measures publication breadth, not the fraction of workers adopting AI.

The sum of published task shares associated with an occupation, divided by that occupation's catalogue task count.

A crosswalk relation labelled by semantic type, such as exact, narrow or broad. The relation type records mapping evidence; it does not supply an allocation probability.

For a native US occupation, its share of classified consumer usage divided by its share of national jobs. It is a relative concentration statistic, not a worker-level probability. Also called the concentration ratio or usage relative to employment. The European donor means do not support this calculation.

The 11 grade-A geographies emphasized in the paper’s European mapping analysis: Belgium, France, Germany, Italy, the Netherlands, Poland, Portugal, Spain, Switzerland, Türkiye and the United Kingdom. This is a reporting subset defined after inspecting the data, not the full observed sample or a representative sample of Europe. Selection criteria.

Anthropic’s March 2026 measure combining task capability with qualifying observed work-related use, automation and task-time aggregation. It differs from a simple published conversation share. Comparisons with the score test alignment between constructs; the independent public-input reconstruction did not pass its acceptance gate. Formula and reconstruction result.

The information available to sustain an estimate: published categories and their mass, observation windows, source precision and, where relevant, compatible employment evidence. More support does not by itself establish representativeness. A reporting grade summarizes selected components; the underlying components remain available separately.

A classification bridge that uses available source-to-destination links without restricting them by semantic relation type. It is a sensitivity comparison with typed routes, not evidence that all links have the same meaning or reliability.

The complete lookup directory of 250 statistical countries and areas: 248 UNSD M49 entries plus Kosovo and Taiwan. Inclusion records an identifier, not an observed AI-use estimate. Only 121 entries have May 2026 occupation observations.

A rule that divides each published source occupation's share across destination occupations, with weights summing to one, including unmatched activity. Summing destination shares therefore recovers the original published source total. The division is an assumption about semantic links, not observation of individual users' jobs.

The smallest and largest allocation a destination can receive when ambiguous published source shares may be split arbitrarily among their permitted destinations. These bounds exclude unpublished activity and are not confidence intervals. Different destinations cannot generally attain all their upper bounds at once.

An occupation's allocated share of national consumer activity divided by its share of the full official national employment total. It describes activity concentration across two populations, with source dates and flags retained. It is not an adoption rate or an estimated economic effect.