When to open: when you need to answer questions like "Does AI actually make people faster? / What kinds of jobs will AI replace? / Which industries are most impacted? / What is AI penetration like in China, India, Brazil, and Africa?"data-supported questions, rather than relying on intuition or hot takes from tech bloggers. Anthropic uses a dual track of Clio (privacy-preserving analytics) + user surveys, and has been publishing reports in the series since 2025 — this wiki is an index of these data and metrics.
One-sentence core:The Anthropic Economic Index is one of the few AI economic impact measurement projects with "raw data + public methodology + ongoing updates" — not secondhand reports in the style of a16z / McKinsey's "we surveyed X companies and think Y", but metrics derived directly from real Claude usage data (after privacy protection).Therefore conclusions such as "the distribution of what software engineers use Claude for", "the largest gains accrue to both high- and low-wage workers", and "differences in how AI is used in China, India, Brazil, and Africa versus Europe and the United States" all have a replicable data foundation.
Boundary with other wikis: business judgment + investment frameworks (valuation / cycles / model company wars) are in AI Industry Judgment and Company Competition (Cross-Author Topic Synthesis) / Wang Kai RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework; AI safety / alignment research is in Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy). This wiki focuses on Anthropic's own economic impact data + methodology.
0. Overview of 10 pieces / 5 report categories
| Time | Report / Article | Core question | Type |
|---|---|---|---|
| Early 2025 | The Anthropic Economic Index (project announcement) | "Can we continuously measure AI's role in the economy?" | Project announcement news |
| 2025-Q1 | AI's impact on software development | "How is AI changing the programming profession?" | 1st series report (software dev) |
| Mid-2025 | Anthropic Economic Index: Insights from Claude Sonnet 3.7 | Task usage distribution under the impact of Sonnet 3.7 | Update in news form |
| 2025-09 | Uneven geographic and enterprise AI adoption(September Report) | "40% of US workers use AI / but large regional and enterprise differences" | 2nd series report (geography) |
| 2025-09 | Tracking AI's role in the US and global economy (companion piece) | Same as above, with dashboard / data links | Companion data presentation |
| 2026-01 | Economic Primitives(January Report) | "We are creating new metrics: scope / breadth / depth" | 3rd series report (primitives) |
| 2026-01 | New building blocks for understanding AI use (companion piece) | Introduction to Economic primitives | Companion news article |
| 2026-04 | What 81,000 people told us about AI economics | "User survey validates / supplements Clio data" | 4th series report (81K survey) |
| 2026-04 | Anthropic Economic Index Survey launched (monthly) | "Continuous monthly survey data stream" | Project update |
| 2026-Q1 | Anthropic Economic Futures Programme UK + Europe launched | "Turning economic impact research into a cross-national policy project" | policy news |
5 major series reports + companion news / project updates totaling 10 pieces; Anthropic has made this an ongoing project rather than a one-off paper.
1. Methodology: How does Anthropic measure?
1.1 Three data sources
| Track | Source | Advantages | Limitations |
|---|---|---|---|
| Clio (primary) | Actual Claude.ai conversations (fully anonymized + aggregated, see Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy) §6) | Large-scale + real usage + continuous | Can only observe "what tasks / how prompts are phrased", cannot see how good the outcomes are |
| User survey | Anthropic Interviewer tool, monthly questionnaire | Subjective feedback + observation of real work scenarios | Biased toward self-reporting + selection bias |
| Enterprise API data | 1P API usage patterns (enterprise anonymized) | Observes differences between enterprise vs consumer usage | Enterprise customers only |
1.2 Key metrics (introduced in the Primitives report, 2026-01)
Anthropic's 3 dimensions of "economic primitives":
| Metric | Meaning | How it is measured |
|---|---|---|
| Scope scope | How much of a task AI completes | For example "AI wrote 3% of the code vs 80% of the code" |
| Breadth breadth | Across how many types of tasks | For example "only doing translation vs translation + summarization + analysis + coding" |
| Depth depth | Task complexity / depth of multi-turn interaction | For example "quickly translating one sentence vs multi-turn iterative design of a solution" |
Significance: previously the industry used metrics such as "usage rate / time spent / satisfaction", these generic SaaS metrics,cannot accurately describe how AI changes work. Anthropic primitives Specifically measuring "what kinds of things AI is doing"is more suitable for economic research.
1.3 Privacy protection strategy
All data goes through Clio:
- Local aggregation + abstraction(PII removed / no identification of individuals / no identification of companies)
- Only outputs statistics + patterns, does not output specific conversations
- Privacy-preserving algorithms ensure that no single conversation can be reconstructed
This is the methodological foundation of the entire Economic Index series; without Clio there would be no such reports.
2. Core findings of the 5 major reports
2.1 Report 1:AI's impact on software development(2025-Q1)
Core data(extracted by me from the original text):
- Programmers are the most disproportionate user group of Claude — US data
- Programming tasks account for a large share of total Claude usage
- Not only "writing code", but also review / debug / system design / learning new frameworks, etc.
Significance: this was the first report in the series,establishing a baseline "starting with software engineers". Subsequent reports repeatedly compare against this one.
2.2 Report 2:Uneven geographic and enterprise AI adoption(2025-09)
Core data(extracted by me from the original text):
- 40% of US workers use AI(2025), double the 20% figure from 2 years prior
- Extremely uneven geographic distribution: "travel planning" in Hawaii, "scientific research" in Massachusetts, and "web application development" in India are each the most overrepresented
- Enterprise vs consumer: enterprise usage is narrower (concentrated in coding / data analysis / process automation), while consumer usage is broader
Significance:The "global AI revolution" is a different story in different regions. What China / India "use AI for" differs from the United States — the assumption of uniform global penetration does not hold.
2.3 Report 3:Economic Primitives(2026-01)
Core findings(extracted by me from the original text):
- Introduced the 3 dimensions scope / breadth / depth (see §1.2 of this wiki)
- Task distribution: Claude is increasingly used for "complex, multi-turn, cross-domain" tasks, not "quick one-question-one-answer" interactions
- Data snapshot on the eve of the Opus 4.5 release (2025-11) — later compared against data after the release of Opus 4.5 / 4.6
2.4 Report 4:81,000 people told us about AI economics(2026-04)
Core data: surveyed 81,000 Claude users (using Anthropic Interviewer), with 3 key findings:
| # | Finding | Meaning |
|---|---|---|
| 1 | Occupations with high AI exposure = high concern | The more replaceable by AI a job is, the more workers worry about being replaced — not surprising, but supported by data |
| 2 | Early-career workers are more concerned | Senior practitioners have relatively lower concern — possibly because the expertise accumulated by senior workers is harder to replace |
| 3 | Highest-paid + lowest-paid occupations see the largest productivity gains | "U-shaped distribution" — the middle tier (mid-wage / standardized tasks) sees smaller gains instead |
Comparison with the "Seniority-Biased AI" research mentioned in Wang Kai's investment framework (see Wang Kai Collection / 2024–2026 AI Frontier Papers and Reports Close Reading) — junior hiring -7.7% / U-shaped heterogeneity, which mutually corroborates findings #2 and #3 of this report.
2.5 Report 5: Anthropic Economic Index Survey (2026-04, ongoing monthly)
Projectization: Anthropic turned the "economic survey" from a one-off report → an ongoing monthly project, automated through Anthropic Interviewer.
Significance:
- The scope upgraded from a one-off paper to long-term panel data(panel data) → the gold standard for economic research
- Cross-validation with data from the Federal Reserve / BLS / national statistics bureaus of various countries becomes possible
3. Key data points (pulled out separately, grep-friendly)
The following figures are used repeatedly across multiple reports,listed separately:
| Number | Meaning | Source |
|---|---|---|
| 40% | 2025 US workers reporting "using AI at work" | September Report |
| 20% | 2023 same metric (doubled in 2 years) | September Report |
| 6% | claude.ai conversations are of a "personal guidance" nature (1 million conversation sample) | claude-personal-guidance article (categorized under Anthropic Interpretability + Claude Behavior Research (Natural Language Autoencoders + User Usage Insights)) |
| 81,000 | Survey sample size (2026-04) | 81k-economics |
| U-shaped distribution | Largest productivity gains at both the high-wage and low-wage ends | 81k-economics |
| scope / breadth / depth | 3 dimensions of Economic primitives | January 2026 Report |
4. Anthropic Economic Futures Programme (policy project, UK+Europe)
From news/economic-futures-uk-europe.
Project: The Anthropic Economic Futures Programme was launched in UK + Europe.
What it does(extracted by me from the original text):
- Funds academia / think tanks in the UK + Europe to conduct research on AI's economic impact
- Shares Anthropic data + tools (in part, under compliance frameworks)
- Engages with policymakers (UK / EU) to bring data into policy decision-making
Significance:Upgrades research from "Anthropic publishing papers on its own" to "conducting it in collaboration with international policy institutions". This is consistent with Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy) §7 "2028 AI Leadership policy paper" — Anthropic is developing both the economic and geopolitical narratives together.
5. Comparison with external research / reports
| Source | Perspective | |---|---| | Anthropic Economic Index (this wiki) | Empirically measured data, continuously updated, 3-dimensional primitives | | a16z / Sequoia / McKinsey | Surveys + investment narrative, often including market inferences | | Federal / official statistics of various countries | Labor force + income data, but measurement of AI impact lags | | OpenAI / Google internal data | Exists but is not public | | Academia(MIT Sloan / Stanford) | Paper-based, but with small samples and discontinuity | | Wang Kai / Lao Huang and other third-party practitioners | Individual / small-team firsthand experience + post-level judgments | See Wang Kai's Observations on AI Industry Judgments and the Competitive Landscape of Large Companies / newtype Observations on AI Business, Industry, and Investment |
Anthropic's unique advantage:The only entity with the full package of "real large-scale usage data + public methodology + continuous updates + policy engagement".
Industry criticism:
- "The data comes from Claude users, which may not represent the entire AI economy" (Claude users skew toward developers / the technology industry)
- "Some economic conclusions overlap with Anthropic's commercial interests" (the claim that working with Claude delivers +X% productivity = marketing talk)
- Conclusions such as the "U-shaped distribution" require independent external replication to be robust
This wiki does not pass judgment; it explicitly lists disagreements.
6. Key timeline
| Time | Milestone |
|---|---|
| Early 2025 | The Anthropic Economic Index project is announced |
| 2025-Q1 | Report 1:AI's impact on software development |
| Mid-2025 | Insights from Claude Sonnet 3.7(news update) |
| 2025-09 | Report 2:Uneven geographic + enterprise AI adoption |
| 2026-01 | Report 3: New Economic Primitives metrics |
| 2026-Q1 | Anthropic Economic Futures Programme UK + Europe launched |
| 2026-04 | Report 4:81,000 people survey |
| 2026-04 | Monthly survey made ongoing (automated via Anthropic Interviewer) |
One major report every 3–6 months + monthly small data releases. It is the most serious economic impact measurement project among AI companies.
7. Quick reference for PM / policy / investment perspectives
| Your role | Which thread should you dig into? |
|---|---|
| PM building AI products | The 3 dimensions of scope / breadth / depth (measuring how much "share" your product occupies in users' work) |
| HR / recruiting | 81K survey #1+#2 (occupational exposure + early-career concerns) to adjust your recruiting narrative |
| Corporate economic consulting | September Report geographic + firm-level differences (for building corporate AI deployment strategies) |
| Investment / VC | High-wage + low-wage U-shape (risk of "middle-layer squeeze" for standardized mid-wage roles) |
| Policymaking | Economic Futures Programme (aligned with UK+Europe data) |
| Academic research | Monthly survey panel data for use in economics research |
| Personal career planning | Look at your occupation's "exposure" + early-career concern level to make long-term decisions |
8. Related wikis / cross-reference map
9. Meta-information
Rewriting strategy: Reorganize 10 articles / 5 major reports according to "methodology + core findings of the 5 reports + key data points + timeline + policy programs + industry comparisons".The focus is not on close reading report by report, but on compressing the entire project into a map of "understand at a glance + key numbers + what to dig into for your role".
Bilingual strategy(translation_status: bilingual):
- Section titles use mixed Chinese-English formatting (Economic Index / Primitives / scope / breadth / depth are not translated)
- Core English terms are given Chinese explanations on first appearance
- Data points retain original numbers + Chinese explanations
Industry criticisms are explicitly listed(not smoothed over): 3 industry criticisms are listed in §5 — data skews toward Claude users / overlap with commercial interests / U-shaped distribution requires independent replication.
Phase C progress: 3 / 3 completed. Phase C is fully complete, totaling 3 wikis:
- Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy)
- Anthropic Interpretability + Claude Behavior Research (Natural Language Autoencoders + User Usage Insights)
- Anthropic Economic Index Complete Collection (AI Economic Impact Measurement, 2025–2026 Series)(this article)
Next step: Phase C finalize (backfill frontmatter ingested_to + add 3 articles to INDEX + commit), then Phase D.
Sources and related materials
- https://www.anthropic.com/research/anthropic-economic-index-january-2026-report
- https://www.anthropic.com/research/anthropic-economic-index-september-2025-report
- https://www.anthropic.com/research/economic-index-geography
- https://www.anthropic.com/research/economic-index-primitives
- https://www.anthropic.com/research/economic-index-survey-announcement
- https://www.anthropic.com/research/81k-economics
- https://www.anthropic.com/research/impact-software-development
- https://www.anthropic.com/news/the-anthropic-economic-index
- https://www.anthropic.com/news/anthropic-economic-index-insights-from-claude-sonnet-3-7
- https://www.anthropic.com/news/economic-futures-uk-europe
- Anthropic AI Safety / Alignment / Policy Research Complete Collection (Constitutional AI + RSP + Welfare + Policy)
- Anthropic Interpretability + Claude Behavior Research (Natural Language Autoencoders + User Usage Insights)
- First-Party Perspective from the Anthropic Engineering Team: Agent Engineering / Skills / MCP / Claude Code (Complete 2025–2026 Anthology)
- AI Industry Judgments and Corporate Competition (Cross-Author Thematic Synthesis)
- Wang Kai's Observations on AI Industry Judgments and the Competitive Landscape of Large Companies