Documentation Index
Fetch the complete documentation index at: https://docs.delphina.ai/llms.txt
Use this file to discover all available pages before exploring further.
The Anthropic Economic Index (AEI) is a snapshot of how people actually use AI — which occupational tasks it handles, how often it succeeds, how autonomous it is, and where in the world it’s being used. The data links observed Claude usage to real occupations via O*NET task classifications.
At a glance
| |
|---|
| Schema | ANTHROPIC.ECONOMIC_INDEX |
| Platforms | 1P API (developer integrations) and Claude.ai Free/Pro (consumer) |
| Data window | November 13-20, 2025 (one-week static snapshot) |
| Refresh | None — this is a fixed release |
| Geography | Global, country, and state level (state-level for Claude.ai only) |
| Source | Anthropic Economic Index research paper |
This measures AI task-usage patterns, not wages or employment. The API and Claude.ai channels serve different populations (developers vs consumers) — treat them separately unless you’re explicitly comparing.
Tables
| Table | Description | Rows |
|---|
AEI_1P_API_METRICS | API channel (developer integrations) | 187,772 |
AEI_CLAUDE_AI_METRICS | Claude.ai channel (Free and Pro users) | 458,778 |
How the data is structured
Each row is one metric observation. The key columns:
| Column | What it is |
|---|
FACET | The dimension being measured (e.g. onet_task::use_case, request::collaboration) |
LEVEL | Hierarchy depth (0, 1, or 2). Don’t aggregate across levels. |
VARIABLE | What VALUE represents — a count, percentage, mean, index, etc. |
CLUSTER_NAME | Category bucket (empty for scalar metrics; populated for categorical breakdowns) |
VALUE | The metric value |
GEO_ID / GEOGRAPHY | Geographic grain: global, country, or country-state |
Types of metrics
The data includes counts (volume), percentages (shares), averages, index values (ratio vs global average, where 1.0 = average), confidence intervals, and distribution histograms.
What’s measured
| Category | Metrics |
|---|
| Activity | Use-case distribution (work / personal / coursework), request counts by cluster |
| Outcomes | Task success rate, time savings vs human-only, O*NET task success by task type |
| Collaboration | Mode share (directive / feedback loop / validation), human-only ability share |
| Complexity | Education-year proxy for difficulty, AI autonomy levels, O*NET task counts |
| Geography | Usage by country/state, coverage counts |
Try asking
Use-case mix
- “What share of Claude.ai usage is work vs personal vs coursework?”
- “Top use-case clusters by volume on the API?”
- “How does the use-case mix differ between API and Claude.ai?”
Task outcomes
- “Overall task success rate on each platform?”
- “Which O*NET tasks does AI complete most successfully?”
- “How much time does AI save vs human-only completion?”
Complexity and autonomy
- “What education level are the tasks AI is handling?”
- “How does AI autonomy differ between API and Claude.ai?”
- “Which tasks require the most education years?”
Collaboration
- “What share of interactions are directive vs feedback loop vs validation?”
- “How many tasks are flagged as requiring uniquely human ability?”
Geography
- “Which countries use Claude.ai the most?”
- “Top US states by usage volume?”
- “How many countries are represented?”
Good to know
| Note | Detail |
|---|
| One-week snapshot | All data is from November 13-20, 2025. There are no trends over time — just a single week. |
| Don’t mix hierarchy levels | The data has three hierarchy levels (0, 1, 2) that represent different cuts of the same data. Comparing across levels can be misleading. |
| Two different audiences | API users are developers building integrations. Claude.ai users are individuals. Comparing them directly can be misleading. |
| API is global-only | Country and state breakdowns only exist in the Claude.ai table. |
| Not about jobs or wages | This tracks what tasks AI is doing, not whether it’s replacing jobs or affecting pay. |
