Anthropic has released its latest Economic Index report, offering a deeper look at how its AI model Claude is being integrated into daily work and life across the globe. Drawing from anonymized transcripts of interactions in November 2025, right before the launch of the advanced Opus 4.5 version, the study introduces fresh ways to measure AI engagement. These basic indicators, or primitives, track aspects like the skills involved for users and the AI, the intricacy of assigned jobs, how much independence Claude gets, its performance outcomes, and whether chats focus on professional, learning, or everyday needs.
The analysis uncovers notable differences in how people wield Claude around the world, alongside practical insights into the length of tasks AI can manage and updates to projections on Claude’s role in boosting overall economic output. Accompanying the report is the most detailed dataset yet, breaking down consumer and business patterns by country and region for the Claude.ai platform.
Comparing to the September 2025 edition, the new data shows Claude interactions still cluster heavily around a handful of activities, especially those tied to programming. Out of thousands of distinct professional duties spotted on Claude.ai, the leading ten make up about a quarter of sampled exchanges, a modest uptick from before. Interactions where users refine ideas, seek guidance, or build on Claude’s input now cover just over half of Claude.ai sessions, reversing a prior trend toward pure automation. For business API calls, however, hands-off operations remain the norm, fitting their scripted setup.
Adoption patterns worldwide stay lopsided, closely tied to a nation’s wealth, with top spots going to the United States, India, Japan, the United Kingdom, and South Korea. Inside the US, areas rich in tech and math professionals drive higher engagement, yet lower-adoption states are catching up quicker than expected. Continued at this rate, usage could even out nationwide in two to five years, a spread far swifter than many past major innovations.
Delving into the new primitives, the report highlights how Claude’s applications broaden as uptake and affluence grow. While professional tasks dominate, lower-wealth countries lean more on it for schoolwork, and wealthier ones for casual pursuits, suggesting early users in emerging markets stick to targeted or educational roles, while established markets branch into varied uses.
Claude performs well on most assignments, aligning its explanations to the user’s level, but falters on tougher ones. As estimated human effort for a job rises, success dips, echoing evaluations that test AI limits on extended work. Factoring in these rates reshapes views of job vulnerability: professions like data input specialists or database designers see Claude handling big chunks effectively, based on coverage, reliability, and task weight.
Overall, Claude handles duties needing more schooling than average economic activities. Stripping out AI-assisted elements could lower skill demands in many roles, though effects vary. For instance, travel planners might shift from intricate itinerary building to basic bookings, simplifying their work, while property overseers could focus on deals and relations after routine accounting fades, elevating their expertise.
These insights sharpen the picture of AI’s economic footprint. Success metrics clarify automatable chores, job risks, and output shifts, while skill matching points to unequal gains. Nations with stronger schooling systems seem primed to capitalize on AI beyond mere access.
Updating from prior findings, task focus has tightened slightly on both consumer and enterprise sides, with coding fixes leading at six percent of Claude.ai use and ten percent of API. Computer-related duties claim a third of chats and nearly half of API activity, though consumer coding has eased a bit as education and creative writing rise. Business API sees gains in admin support, like email handling and scheduling, hinting at backend automation.
Collaboration styles on Claude.ai have swung back to human-led, with iterative refinements overtaking direct commands. Product tweaks, such as file handling, lasting recall, and custom workflows, likely encourage this teamwork. API stays automation-heavy.
Regionally, the US shows workforce makeup steering uptake, with tech-heavy spots like Virginia leading, explaining much of the spread. Globally, no major shifts in disparities, but US internal gaps narrow fast, potentially matching historical tech rolls but accelerated.
The primitives stem from efforts to gauge economic relevance, like time estimates validated against real benchmarks and use classifications checked for broad accuracy. They reveal contrasts: tech requests demand 13-plus years of learning versus nine for daily help, with work skewing professional and success favoring simpler jobs.
Claude.ai differs from API: longer, iterative talks with higher wins and autonomy, covering education and arts more, while API zeros in on code and office routines for automation.
Geographically, wealth drives how Claude serves: richer spots favor collaborative work and personal, poorer lean educational. Adoption ties to income and prompt sophistication, with complex prompts sparking matched replies.
On tasks, tougher ones speed up more but succeed less, hitting 50 percent around three hours for API, 19 for chats. Effective job reach, blending success and time shares, spotlights roles like data entry for full swaps, while coverage alone misses bottlenecks.
AI hits higher-education tasks, so removal deskills most jobs, like writers losing analysis for sketches, but upskills some, like managers ditching records for negotiations.
Productivity projections adjust down with reliability to one percentage point yearly gain, or less if tasks complement tightly, but still notable amid past lows. Future model leaps could expand this.
This data empowers scrutiny of AI’s shifts, urging skill-building for broad benefits. For full details, check the Anthropic report.
