Anthropic Economic Index

Side-by-side headline comparison of 2-8 regions from one cohort. Returns each region's usage index, augmentation/automation split, work/personal/coursework split, top job categories, most common request topics, and top work tasks, plus the shared baseline (the world for countries; the country for its subregions). Regions publish different work-task subsets — compare tasks by name, not rank position. Compare countries with countries, or subregions of one country — indexes are normalized within one cohort and never compare across cohorts, and subregion usage shares are likewise calculated within the parent country. For region profiles beyond the headline, use the detail tools; for a full ranking, the list tools. When a visualization tool is available, always build comparative visualizations — show each region relative to the others and the shared baseline: heatmaps, diverging bar charts, bubble charts, and grouped or stacked bars all fit side-by-side data; go beyond a single bar chart or choropleth, and more than one visualization is fine. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Anthropic Economic Index overview: coverage, releases, definitions. Call this once at the start of an exploration. Returns the latest data period and snapshot timestamp, dataset metadata (license, publisher, landing page), temporal_coverage — the published period span; the served numbers are a snapshot of the latest period, not a trend — geography and occupation coverage counts, the worldwide headline split, and a definitions glossary to reuse when presenting numbers to non-experts. country_count counts only countries with a published Anthropic Usage Index, the same cohort econ_index_list_countries / econ_index_get_usage_by_country cover. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Global Claude usage, occupational task breakdowns, and common request topics. How people use Claude overall including occupational task breakdowns and common request topics — the Anthropic Economic Index worldwide profile. Returns the augmentation/automation and work/personal/coursework splits, the classifier-estimated task-time comparison (hours working alone vs minutes with Claude — normalize the units before charting), top job categories, top request topics, and artifact types. Topic-share questions — how often people ask about money, health, investing, travel, and the like — are answered by top_request_topics, and the worldwide work-task ranking (econ_index_list_top_work_tasks) carries the task-level detail behind the same questions. Whenever the question warrants a country breakdown, call econ_index_list_countries and lead with the world choropleth (its rows carry iso_numeric for a code join), with this profile's splits as compact metric badges above the map; then visualize the job categories. When a visualization tool is available, always explore visualizations: consider the data's shape — how many series, whether values carry a baseline comparison, whether anything is geographic — try the available options and judge which fits best; more than one visualization is fine. Geographic data gets a choropleth by default. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Occupations and job categories ranked by Claude usage of their tasks. A bare call returns category_shares — every job category with its share of global usage. Lead with those ~22 categories — one ranked series of shares — and visualize them, narrating only the top few. Pass query to drill down: a category name lists its occupations, an occupation name returns that occupation, and any other word matches occupation names. When no slice exists for the exact profession asked, quote the closest published figure with its scope named ('software developers', 'Legal as a whole') — never present broader numbers as the requested slice. A gap between task_count and published_task_count is privacy suppression, not zeros. Global scope only — for one state or country use econ_index_get_usage_by_subregion / econ_index_get_usage_by_country. Answers questions like 'which jobs and careers use AI the most?'. Frame results as "AI is used for tasks commonly done by [occupation]", never "[occupation]s are using AI" — the people in these conversations are often not members of that occupation. When a visualization tool is available, always explore visualizations: consider the data's shape — how many series, whether values carry a baseline comparison, whether anything is geographic — try the available options and judge which fits best; more than one visualization is fine. Geographic data gets a choropleth by default. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Full Claude-usage profile for one country. Returns the usage index and rank, the augmentation/automation and work/personal/coursework splits, top-10 job categories, most frequent request topics, work tasks, and artifact types. The key shares carry the global average alongside (baseline_pct, vs_baseline_index), and the baseline block carries the global headline figures for side-by-side display. Pass include_tasks / include_requests for the full occupation-task and request-topic hierarchies (each node carries its 0-100 share of the country's usage). Only countries with observed usage above the Economic Index aggregation threshold are covered. Open with the headline numbers against the global average. For any single-country question, follow up with econ_index_list_subregions for the within-country picture — coverage varies and an empty result says so; where rows exist, lead with the subregion choropleth. Then visualize the job categories, then a brief mention of topics, then artifact types. When a visualization tool is available, always explore visualizations: consider the data's shape — how many series, whether values carry a baseline comparison, whether anything is geographic — try the available options and judge which fits best; more than one visualization is fine. Geographic data gets a choropleth by default. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Full Claude-usage profile for one subregion (state, province, prefecture). Covers US states and, where published, other countries' subregions (BR-SP, JP-13, ...). Returns the usage rank within the country, the augmentation/automation and work/personal/coursework splits, top-10 job categories, most frequent request topics, work tasks, and artifact types. The key shares carry the country average alongside (baseline_pct, vs_baseline_index), and the baseline block carries the country's headline figures for side-by-side display. US states also carry the Anthropic Usage Index; other subregions have no published index — present their standing by rank and share, never a made-up index. Pass include_tasks / include_requests for the full occupation-task and request-topic hierarchies (each node carries its 0-100 share of the subregion's usage). Empty lists mean not published — never invent entries. Open with the headline numbers against the country average, then visualize the job categories, then the most frequent topics, then artifact types. When a visualization tool is available, always explore visualizations: consider the data's shape — how many series, whether values carry a baseline comparison, whether anything is geographic — try the available options and judge which fits best; more than one visualization is fine. Geographic data gets a choropleth by default. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Every country with a published Anthropic Usage Index, ranked by it. Each row carries the index and rank, the augmentation/automation and work/personal/coursework splits, the top-5 job categories, top artifact types, and its top-5 request topics with shares — enough for a choropleth with tooltips and a click-through panel from this single call. iso_numeric matches the feature ids of a world topojson (world-atlas countries-110m): join the map on codes, not country names. For one country's top-10 categories and full hierarchies, use econ_index_get_usage_by_country. Lead with a world choropleth colored by anthropic_usage_index; give tooltips the index plus the top job categories or the use-case split. Follow with the top job categories — a country ranking complements the map rather than replacing it. To compare countries, compare profiles (each one's categories and topics vs the global average: baseline_pct, vs_baseline_index), not just usage volume. When a visualization tool is available, always visualize: this data is geographic, so lead with a choropleth of a numeric field — the usage index or a split share; a map of each region's top topic or category comes out near-uniform, because the same entries lead nearly everywhere — when no map source for these regions is bundled or known, search for one before any non-map fallback; more than one visualization is fine. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

One country's subregions ranked by Claude usage. Defaults to the USA: all 50 states plus Washington, D.C., ranked by the Anthropic Usage Index. Other countries' subregions have no published index — rows rank by usage_share_pct, each subregion's share of the country's observed usage; that is volume, not engagement relative to population size, and map colors should say so. Each row carries the use-case splits, top-5 job categories, top artifact types, and its top-5 request topics with shares — enough for a choropleth with tooltips and a click-through panel from this single call. For one subregion's top-10 categories and full hierarchies, use econ_index_get_usage_by_subregion. For US states, join a us-atlas states topojson by state name (Washington, D.C. is 'District of Columbia' there) and color by anthropic_usage_index; elsewhere search for a suitable topojson when none is at hand and color by usage_share_pct. To compare subregions, compare profiles (each one's categories and topics vs the national average: baseline_pct, vs_baseline_index), not just usage volume. When a visualization tool is available, always visualize: this data is geographic, so lead with a choropleth of a numeric field — the usage index or a split share; a map of each region's top topic or category comes out near-uniform, because the same entries lead nearly everywhere — when no map source for these regions is bundled or known, search for one before any non-map fallback; more than one visualization is fine. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.

Top work tasks worldwide: what people use Claude for the most, by task. Returns the top 50 work tasks, ranked by their 0-100 share of global sampled conversations. Task descriptions come from O*NET, the US government's catalog of each occupation's tasks, in plain English, and are matched from conversation content. For the breakdown within one occupation, use econ_index_get_occupation_usage. Answers questions like 'what do people in different careers actually use Claude for?'. Task wording is specific enough that questions about one kind of request — financial advice, resume help, lesson planning — often match tasks here directly. When a visualization tool is available, always explore visualizations: consider the data's shape — how many series, whether values carry a baseline comparison, whether anything is geographic — try the available options and judge which fits best; more than one visualization is fine. Geographic data gets a choropleth by default. Figures are derived from anonymized, aggregated conversation content and cannot be tied to individual users. They describe observed Claude usage by task and are not evidence about employment or job automation.