Uneven geographic and enterprise AI adoption Anthropic

Introduction

AI differs from prior technologies in its unprecedented adoption speed. Within the US alone, 40% of employees report using AI at work, up from 20% in 2023 two years ago.¹ Such rapid adoption reflects how useful this technology already is for a wide selection of applications, its deployability on existing digital infrastructure, and its ease of use—by just typing or speaking—without specialized training. Rapid improvement of frontier AI likely reinforces fast adoption along each of those dimensions.

Historically, recent technologies took many years to achieve widespread adoption. Electricity took over 30 years to achieve farm households after urban electrification. The primary mass-market laptop computer reached early adopters in 1981, but didn’t reach nearly all of homes within the US for one more 20 years. Even the rapidly-adopted web took around five years to hit adoption rates that AI reached in only two years.²

Why is that this? In brief, it takes time for brand spanking new technologies—even transformative ones—to diffuse throughout the economy, for consumer adoption to change into less geographically concentrated, and for firms to restructure business operations to best unlock recent technical capabilities. Firm adoption, first for a narrow set of tasks, then for more general purpose applications, is a vital way that consequential technologies spread and have transformative economic effects.³

In other words, an indicator of early technological adoption is that it’s concentrated—in each a small variety of geographic regions and a small variety of tasks in firms. As we document on this report, AI adoption appears to be following the same pattern within the twenty first century, albeit on shorter timelines and with greater intensity than the diffusion of technologies within the twentieth century.

To check such patterns of early AI adoption, we extend the Anthropic Economic Index along two essential dimensions, introducing a geographic evaluation of Claude.ai conversations and a first-of-its-kind examination of enterprise API use. We show how Claude usage has evolved over time, how adoption patterns differ across regions, and—for the primary time—how firms are deploying frontier AI to unravel business problems.

Changing patterns of usage on Claude.ai over time

In the primary chapter of this report, we discover notable changes in usage on Claude.ai over the previous eight months, occurring alongside improvements in underlying model capabilities, recent product features, and a broadening of the Claude consumer base.

We discover:

• Education and science usage shares are on the rise: While using Claude for coding continues to dominate our total sample at 36%, educational tasks surged from 9.3% to 12.4%, and scientific tasks from 6.3% to 7.2%.
• Users are entrusting Claude with more autonomy: “Directive” conversations, where users delegate complete tasks to Claude, jumped from 27% to 39%. We see increased program creation in coding (+4.5pp) and a discount in debugging (-2.9pp)—suggesting that users might have the ability to realize more of their goals in a single exchange.

The geography of AI adoption

For the primary time, we release geographic cuts of Claude.ai usage data across 150+ countries and all U.S. states. To check diffusion patterns, we introduce the Anthropic AI Usage Index (AUI) to measure whether Claude.ai use is over- or underrepresented in an economy relative to its working age population.

We discover:

• The AUI strongly correlates with income across countries: As with previous technologies, we see that AI usage is geographically concentrated. Singapore and Canada are amongst the very best countries when it comes to usage per capita at 4.6x and a couple of.9x what can be expected based on their population, respectively. In contrast, emerging economies, including Indonesia at 0.36x, India at 0.27x and Nigeria at 0.2x, use Claude less.
• Within the U.S., local economy aspects shape patterns of use: DC leads per-capita usage (3.82x population share), but Utah is close behind (3.78x). We see evidence that regional usage patterns reflect distinctive features of the local economy: For instance, elevated use for IT in California, for financial services in Florida, and for document editing and profession assistance in DC.
• Leading countries have more diverse usage: Lower-adoption countries are inclined to see more coding usage, while high-adoption regions show diverse applications across education, science, and business. For instance, coding tasks are over half of all usage in India versus roughly a 3rd of all usage globally.
• High-adoption countries show less automated, more augmented use: After controlling for task mix by country, low AUI countries usually tend to delegate complete tasks (automation), while high-adoption areas tend toward greater learning and human-AI iteration (augmentation).

The uneven geography of early AI adoption raises essential questions on economic convergence. Transformative technologies of the late nineteenth century and the early twentieth centuries—widespread electrification, the interior combustion engine, indoor plumbing—not only ushered within the era of contemporary economic growth but accompanied a big divergence in living standards around the globe.⁴

If the productivity gains are larger for high-adoption economies, current usage patterns suggest that the advantages of AI may concentrate in already-rich regions—possibly increasing global economic inequality and reversing growth convergence seen in recent many years.⁵

Systematic enterprise deployment of AI

In the ultimate chapter, we present first-of-its-kind insight on a big fraction of our first-party (1P) API traffic, revealing the tasks corporations and developers are using Claude to perform. Importantly, API users access Claude programmatically, reasonably than through an internet user interface (as with Claude.ai). This shows how early-adopting businesses are deploying frontier AI capabilities.

We discover:

• 1P API usage, while much like Claude.ai use, differs in specialized ways: Each 1P API usage and Claude.ai usage focus heavily on coding tasks. Nonetheless, 1P API usage is higher for coding and office/admin tasks, while Claude.ai usage is higher for educational and writing tasks.
• 1P API usage is automation dominant: 77% of business uses involve automation usage patterns, in comparison with about 50% for Claude.ai users. This reflects the programmatic nature of API usage.
• Capabilities appear to matter greater than cost in shaping business deployment: Essentially the most-used tasks in our API data are inclined to cost greater than the less frequent ones. Overall, we discover evidence of weak price sensitivity. Model capabilities and the economic value of feasibly automating a given task appears to play a bigger role in shaping businesses’ usage patterns.
• Context constrains sophisticated use: Our evaluation suggests that curating the precise context for models will likely be essential for high-impact deployments of AI in complex domains. This means that for some firms costly data modernization and organizational investments to elicit contextual information could also be a bottleneck for AI adoption.

Open source data to catalyze independent research

As with previous reports, we’ve got open-sourced the underlying data to support independent research on the economic effects of AI. This comprehensive dataset includes task-level usage patterns for each Claude.ai and 1P API traffic (mapped to the O*NET taxonomy in addition to bottom-up categories), collaboration mode breakdowns by task, and detailed documentation of our methodology. At present, geographic usage patterns are only available for Claude.ai traffic.

Key questions we hope this data will help others to research include:

• What are the local labor market consequences for employees and firms of AI usage & adoption?
• What determines AI adoption across countries and throughout the US? What may be done to be sure that the advantages of AI don’t only accrue to already-rich economies?
• What role, if any, does cost-per-task play in shaping enterprise deployment patterns?
• Why are firms in a position to automate some tasks and never others? What implications does this have for which varieties of employees will experience higher or worse employment prospects?

1. Gallup 2025, AI Use at Work Has Nearly Doubled in Two Years.

2. Bick, Blandin, Deming, 2024 The Rapid Adoption of Generative AI benchmark AI adoption against adoption of PC and the web; Lewis & Severnini, 2020 Short- and long-run impacts of rural electrification: Evidence from the historical rollout of the U.S. power grid analyze the impact of bringing electricity to rural areas on economic outcomes.

3. Kalyani, Bloom, Carvalho, Hassan, Lerner and Ahmed Tahoun 2025 Diffusion of Recent Technologies.

4. See Gordon, 2012 Is U.S. Economic Growth Over? Faltering Innovation Confronts the Six Headwinds for a comparison of early and late twentieth century innovations and their impact on productivity. Pritchett, 1997. Divergence, Big Time documents economic divergence that accompanied transition to era of contemporary economic growth.

5. Kremer, Willis, You, 2022 Converging to Convergence present evidence of growth convergence in recent many years. See Jones, Jones, and Aghion, 2017 Artificial Intelligence and Economic Growth for discussion of growth implications AI-powered automation of innovation.

Chapter 1: Claude.ai usage over time

Overview

Understanding how AI adoption evolves over time may help predict its economic impacts—from productivity gains to workforce changes. With data spanning from December 2024 and January 2025 (from our first report, ‘V1’) to February and March 2025 (‘V2’) to our newest insights from August 2025 (‘V3’), we are able to track how AI usage has shifted over the past eight months as capabilities and product features have improved, recent sorts of users have adopted the technology, and uses have change into more sophisticated. We view the evidence presented below as suggesting that recent product features have enabled recent types of work reasonably than simply accelerating adoption for existing tasks.

How Claude.ai usage for economic tasks has modified

Educational and scientific tasks proceed their rise in relative importance

While computer and mathematical tasks still dominate overall usage at 36%, we’re seeing sustained growth in knowledge-intensive fields. Educational Instruction and Library tasks rose from 9% in V1 to 12% in V3. Life, Physical, and Social Science tasks increased from 6% to 7%. Meanwhile, the relative share of Business and Financial Operations tasks fell from 6% to three%, and Management dropped from 5% to three%.

This divergence suggests AI usage could also be diffusing especially quickly amongst tasks involving knowledge synthesis and explanation, in comparison with traditional business operations—possibly because these tasks profit more from Claude’s reasoning capabilities.

Recent capabilities are shaping usage patterns

At a more granular level, we document changes in task composition that appear linked to features launched between V2 and V3. For instance, searching electronic sources and databases grew substantially (0.03% → 0.49%), likely reflecting our web search release in March. As well as, we also see an increase in internet-based research tasks (0.003% → 0.27%), which aligns with the Research mode we released in April.¹

We also see different kinds of changes. Tasks regarding developing instructional materials increased by 1.3pp, growing from a base of 0.2% to 1.5%—a greater than 6-fold increase that will reflect growing adoption amongst educators.² Creating multimedia documents rose 0.4pp, nearly tripling from 0.16% to 0.55%, potentially driven by continued use of our Artifacts feature for constructing traditional and AI-powered apps inside Claude.ai.

Interestingly, the share of tasks involving creating recent code greater than doubled, increasing by 4.5 percentage points (from 4.1% to eight.6%), while debugging and error correction tasks fell by 2.8 percentage points (from 16.1% to 13.3%)—a net 7.4pp shift toward creation over fixing code. This will likely suggest that models have change into increasingly reliable, such that users spend less time fixing problems and more time creating things in a single interaction.³

Directive automation is accelerating

As in previous reports, we also track not only what people use Claude for but how they collaborate with or delegate to Claude on Claude.ai.

At a high level, we distinguish between automation and augmentation modes of using Claude:

Automation encompasses interaction patterns focused on task completion:

• Directive: Users give Claude a task and it completes it with minimal back-and-forth
• Feedback Loops: Users automate tasks and supply feedback to Claude as needed

Augmentation focuses on collaborative interaction patterns:

• Learning: Users ask Claude for information or explanations about various topics
• Task Iteration: Users iterate on tasks collaboratively with Claude
• Validation: Users ask Claude for feedback on their work

The share of directive conversations sampled from Claude.ai conversations jumped from 27% in V1 in late 2024 to 39% in V3. This increase got here primarily on the expense of task iteration and learning interactions, implying a large net increase within the share of conversations exhibiting automative patterns of use – a notable increase in only eight months. That is the primary report where automation usage exceeds augmentation usage.

One interpretation is that this can be a result of accelerating model capabilities. As models improve at anticipating user needs and producing high-quality outputs on first attempts, users may have fewer follow-up refinements. The jump in directive usage could also signal growing confidence in delegating complete tasks to AI, a type of learning-by-doing.⁴

Whether the expansion in directive usage is attributable to improving model capabilities or learning-by-doing could signal very different labor market implications. If more advanced models simply expand the set of automated tasks, then the chance increases that employees performing such tasks will likely be displaced. Nonetheless, if as an alternative the rise in directive use reflects learning-by-doing, then employees most in a position to adapt to recent AI-powered workflows are prone to see greater demand and better wages. In other words, AI may profit some employees greater than others: it might result in higher wages for those with the best ability to adapt to technological change, at the same time as those with lower ability to adapt face job disruption.⁵ This will likely be a vital area of inquiry for future research.

Looking ahead

The V3 data reveals that AI capabilities and adoption are continuing to progress. Knowledge-based tasks, including educational and scientific applications, proceed their fast growth rate, and recent product features look like enabling several types of work reasonably than simply accelerating existing tasks.

Most strikingly, the information point toward increased delegation of tasks to AI systems–perhaps attributable to some combination of user trust within the technology in addition to improvement of underlying model capabilities. This is also attributable to changes within the underlying user base. The following chapter of this report for the primary time breaks down usage across geography, allowing us to disentangle temporal vs. geographic changes more clearly going forward. We are going to proceed to trace these trends closely in future reports.

1. “Search electronic sources, similar to databases or repositories, or manual sources for information” increased from 0.03% to 0.49%. “Conduct internet-based and library research” increased from 0.003% to 0.27%.

2. Statistic computed from tasks containing the string “develop instructional materials”.

3. Tasks were collated from the set of tasks whose frequency has modified by a magnitude greater than or equal to 0.2 percentage points. Programming creation tasks include: “write recent programs or modify existing programs” (1.5% → 4.9%), “design, construct, or maintain web pages” (1.2% → 2.0%), “write, analyze, review, and rewrite programs” (1.2% → 0.5%), “develop recent software applications” (0.06% → 0.6%), “develop transactional web applications” (0.1% → 0.3%), and “develop application-specific software” (0.05% → 0.3%). Debugging/error correction tasks include: “modify existing software to correct errors” (two variants: 2.5% → 3.8% and 4.8% → 2.7%), “correct errors by making appropriate changes” (3.0% → 2.1%), “perform initial debugging procedures” (2.0% → 0.9%), “diagnose, troubleshoot, and resolve hardware/software problems” (1.6% → 2.5%), “review and analyze computer printouts to locate code problems” (1.3% → 0.9%), and “determine sources of web page or server problems” (0.9% → 0.4%).

4. We note that V3 uses Claude Sonnet 4 for classification, while V2 used Sonnet 3.7, which complicates direct comparison. To deal with this, we reran V3 data with Sonnet 3.7 and still found directive interactions rising significantly (though to a lower absolute level of 45% automation versus 49% with Sonnet 4). We also verified this trend isn’t driven by changes in task mix—the shift toward directive interactions appears across a wide selection of occupational categories, suggesting it reflects real changes in how people interact with Claude reasonably than compositional effects.

5. Nelson and Phelps, 1966 Investment in Humans, Technological Diffusion, and Economic Growth is a classic reference for the worth of education in equipping employees to adapt to vary. See also Goldin and Katz, 2008 The Race between Education and Technology. We thank Anton Korinek for the statement that AI itself might speed up the diffusion and economic impact of AI to the extent that it plays the role that expert employees played previously in determining effectively wield recent technologies in novel settings.

Chapter 2: Claude usage across the USA and the globe

Overview

Where AI gets adopted first—and the way it’s used—will shape economic outcomes the world over. By analyzing Claude usage patterns across 150+ countries and all US states, we uncover three key dynamics: where early adopters are, what they’re using AI for, and the way usage evolves as adoption matures. These geographic patterns provide real-world evidence about AI’s economic diffusion, helping track whether different regions are converging or diverging of their AI adoption, and revealing how local economic characteristics shape technology deployment.

Our data, counting on a privacy-preserving¹ evaluation of 1 million Claude.ai conversations², confirmed a few of our expectations while difficult others. The US dominates total usage at 21.6%, which is unsurprising given its size and high income. But even when adjusting for the working-age population size, higher-income countries are inclined to have higher usage. For instance, Singapore’s usage rate is 4.5 times what its working-age population would suggest, while large regions of the globe show minimal usage. Interestingly, throughout the US, DC and Utah outpace California in usage per capita.

We also observe changes in AI use cases as adoption per capita deepens. Countries with lower AI adoption per capita concentrate overwhelmingly on coding tasks—over half of all usage in India, in comparison with roughly a 3rd globally. As adoption matures, usage diversifies, with a rising emphasis on education, science, and business operations.

Much more striking: mature markets are inclined to use AI more collaboratively, while emerging markets usually tend to delegate complete tasks to it—perhaps reflecting differences in how AI is deployed by economies at different stages of structural transformation. Our data provides a window into these patterns across geographies, and going forward, will enable us to trace whether these adoption gaps narrow, widen, or change in structure over time.

Claude diffusion across the globe

Total Claude usage is highest within the US

Claude adoption overall is very geographically concentrated. By way of total global usage, the USA accounts for the very best share (21.6%), with the subsequent highest usage countries showing significantly lower shares (India at 7.2%, Brazil at 3.7%, see Figure 2.1). Nonetheless, this concentration is affected by the population size of every country³ – larger countries can have larger usage shares purely due to their population size.

Per capita usage of Claude is concentrated in technologically advanced countries

To account for differences in population size, we analyze usage adjusted for the working-age population, introducing a brand new measure called the Anthropic AI Usage Index (AUI): For every geography, we calculate its share of Claude usage, and its share of the working-age population (ages 15-64). We then calculate the AUI by dividing these shares:

This index reveals whether countries use Claude roughly than expected relative to their working-age population. A region with an AUI > 1 has higher usage than expected after adjusting for population, while a region with an AUI < 1 has lower usage.

The outcomes reveal a striking pattern of concentration amongst small, technologically advanced economies. Israel leads global per capita Claude usage with an Anthropic AI Usage Index of seven — meaning its working-age population uses Claude 7x greater than expected based on its population. Singapore follows at 4.57, while Australia (4.10), Recent Zealand (4.05) and South Korea (3.73) round out the highest five countries when it comes to per capita Claude usage.

Next, we create per capita usage tiers based on the AUI. We have a look at countries with a minimum of 200 conversations in our random sample of 1 million conversations, and set thresholds for various usage tiers-based quartiles, i.e. Leading (top 25%), Upper Middle (50-75%), Lower Middle (25%-75%) and Emerging (bottom 25%). We then assign countries, even in the event that they have fewer than 200 observations, to a tier based on their AUI. We assign countries for which we’ve got population data, but no usage in our sample, to a Minimal tier.⁴ Figure 2.3 illustrates the Anthropic AI Usage Index tiers across the globe, Table 2.1 shows an summary of the tiers and country examples.

Zooming into leading and emerging countries when it comes to per capita usage

This concentration in advanced economies with limited population sizes reflects their established patterns as technology pioneers. For instance, each Israel and Singapore rank highly within the Global Innovation Index—a measure of how modern different economies across the globe are—suggesting that general investments in information technology position economies well for rapid adoption of frontier AI. Overall, these economies can leverage their educated workforces, robust digital infrastructure, and innovation-friendly policies to create fertile conditions for AI.

Notable is the position of major developed economies in Claude usage. The USA (3.62) ranks amongst leading countries when it comes to per capita adoption, with Canada (2.91) and the UK (2.67) having elevated but more moderate rates of adoption as in comparison with their population. Other major economies show lower adoption, including France at 1.94, Japan at 1.86, and Germany at 1.84.

Meanwhile, many lower and middle-income economies show minimal Claude usage, with many countries across Africa, Latin America, and parts of Asia showing Claude adoption below what can be expected based on their working-age population. This includes Bolivia (0.48), Indonesia (0.36), India (0.27), and Nigeria (0.2).

This variation in usage is reflective of income differences across these economies. We see a robust positive correlation between Claude adoption and Gross Domestic Product per working-age capita (see Figure 2.4), with a 1% increase in GDP per capita being related to a 0.7% increase in Claude usage per capita.

The disparities in Claude usage likely reflect a confluence of things, a few of that are correlated with income:

• Digital infrastructure: High-usage countries typically have robust web connectivity and cloud computing access needed to access AI assistants.
• Economic structure: As documented on this and former reports, Claude capabilities are well-suited to varied tasks typical of information employees. Advanced economies are inclined to have a greater share of the workforce in such roles as in comparison with lower-income economies with a bigger employment share in manufacturing.
• Regulatory environment: Governments differ in how actively they encourage using AI across different industries and in how heavily they regulate the technology.
• Awareness and access: Countries with stronger connections to Silicon Valley and AI research communities can have greater awareness of and access to Claude.
• Trust and luxury: Public opinion on trust in AI varies substantially across countries.

Claude diffusion across the USA

Throughout the US, California overwhelmingly leads with 25.3% of usage. Other states with major tech centers like Recent York (9.3%), Texas (6.7%), and Virginia (4.0%) also rank highly. Though not adjusted for population, we suspect that these strong adoption figures partly reflect rapid adoption in technology hubs—in step with how economically consequential technologies have historically tended to diffuse.

This narrative becomes more complex, nevertheless, once we adjust for the population size of every state. Surprisingly, the District of Columbia leads with an Anthropic AI Usage Index of three.82, indicating that Claude usage in DC is 3.82x greater than its share of the country’s working-age population. Closely following is Utah (3.78), notably ahead of California (2.13), Recent York (1.58) and Virginia (1.57).⁷

We document the same, but weaker correlation than at the worldwide level between Claude adoption and income per capita across US states. Income differences explain lower than half the variation in cross-state adoption rates. Despite this weaker correlation, we discover that Claude adoption rises faster with income: Each 1% increase in state GDP per capita is related to a 1.8% increase within the AI Usage Index.

Task usage patterns across countries

We observe notable variation in how Claude is used in several countries. As in past reports, we analyze these trends using two different approaches. First, we classify conversations into tasks in line with O*NET, a US taxonomy that maps specific tasks to occupations and occupation groups (e.g., a task involving software debugging would fall into the Computer and Mathematical occupation group).

Second, we use Claude to construct a bottom-up taxonomy of user requests on Claude.ai, which provides insight into usage patterns that don’t fit neatly into existing taxonomies. For instance, the request cluster “help write and improve cover letters for job applications” (lowest level) feeds into the higher-level cluster “help with job applications, resumes, and profession documents” (middle level), which in turn feeds into the cluster “help with job applications, resumes, and profession advancement” (highest level). These two complementary approaches allow us to each report results aligned with standard labor statistics, and supply flexibility to capture tasks that standard taxonomies miss.

Higher per capita Claude usage is related to more diverse task usage

When analyzing O*NET tasks aggregated at the very best level (when it comes to the Standard Occupation Classification occupation groups they belong to), we notice strong variation across countries. While the general pattern is noisy–especially for countries with fewer observations–Figure 2.7 suggests that as we progress from lower to higher per capita Claude adoption, usage shifts away from tasks within the Computer and Mathematical occupation group (e.g., programming) to more diverse tasks in areas similar to education, office and administrative uses, and humanities. We also see increased usage within the life, physical and social sciences.

Country idiosyncrasies also emerge when our bottom-up request taxonomy.⁸ Take, for instance, the USA, Brazil, Vietnam, and India, which represent the country with the very best total usage inside a given Anthropic AI Usage Index tier. Users in the USA disproportionately use Claude for household management purposes, to look for jobs, and for medical guidance in comparison with the worldwide average. Against this, Claude users in Brazil have comparatively high usage for each translation and legal services. Vietnam’s top disproportionate requests are related to software development and education, and India’s top disproportionate requests focus almost exclusively on software development. This likely reflects local specialization: Brazil has been an early adopter of AI within the judicial system, and India has a big information technology sector.

Across all countries, software development emerges as essentially the most common use of Claude. Why do developer tasks consistently lead in overall Claude usage patterns? Several aspects likely contribute to this effect:

• Model-task fit: Claude is a really strong coding model and readily deployed across code generation, debugging, and technical problem-solving tasks.
• Developer receptivity: Developer communities embrace recent tools rapidly, and this usage diffuses through their social and skilled networks.
• Low organizational barriers: Individual developers can typically adopt Claude without complex approval processes—in contrast to, say, medical use cases.

Task usage patterns across the USA

On this section we explore patterns of Claude usage across states throughout the US, giving us further insight into how local economic conditions shape usage patterns. As we discuss above, cross-state differences within the Anthropic AI Usage Index account for lower than half of the variation in income differences across US states. This means that other regional differences—including the compatibility of Claude capabilities with the occupational composition of the local workforce—play a bigger role in determining why usage is more concentrated in some states than others.

In quite a few states, we see evidence that local patterns of AI use aligns with distinctive features of the local economy. When analyzing the highest states in each usage tier—California for leading, Texas for upper middle, Florida for lower middle, and South Carolina for emerging—we see strong variation when it comes to our bottom-up request taxonomy (see Figure 2.9).

For instance, California shows disproportionate use for IT-related requests, digital marketing and translation, likely reflecting its tech sector and linguistically diverse population. California also has disproportionately frequent requests for help with basic numerical tasks, which can represent tests of model capabilities or abuse. Florida sees disproportionate use for business advice and fitness, potentially tied to its role as a financial hub with relatively low tax rates and a warm climate amenable to outdoor activities.

Throughout the US, D.C. leads when it comes to per capita Claude usage, with a disproportionate deal with document editing, information provision and job applications across each the O*NET task classification and bottom-up categorization (see Figure 2.10). For instance, help with job applications is 1.84x as common in DC as within the US overall. Our interactive dashboard allows everyone to explore the total range of variation and patterns across US states.

Geographic patterns in human-AI collaboration

While previous sections examined what tasks people use Claude for, an equally revealing pattern emerges in how they interact with it. Here, we use the identical augmentation and automation collaboration patterns as defined in Chapter 1.

Countries have different task mixes, meaning that they deal with different economic tasks, which can partly explain differences in automation patterns. On this section, we investigate whether automated use is systematically different amongst high and low per capita adoption economies—even when controlling for differences in task mix.⁹

We discover that even when controlling for the duty mixture of a rustic, users from different countries show notably different preferences for autonomous delegation versus collaborative interaction. As Claude usage per capita increases, countries shift from automation-focused to augmentation-focused usage. That is somewhat counter-intuitive, since we’re controlling for the more diverse task composition across different countries. We speculate that cultural and economic aspects might affect the automation share, or perhaps that early adopters in each country are inclined to use AI in a more automotive way—but more research is required here.

Conclusion

Our evaluation of Claude usage patterns across geographies reveals several key insights. Some of the striking is the geographic concentration of Claude usage. The leadership of the US and California when it comes to Claude usage overall, and the strong correlation of Claude usage and income per capita, suggest parallels to past technologies wherein initial geographic concentration and specialized use were a key feature. Drawing parallels to the diffusion patterns of prior technologies may help us higher understand the diffusion and impact of AI.

Surprisingly, geography shapes not only what AI tools are used for, but how they’re used. Users in economies with relatively low per capita usage have a relative preference for delegating tasks to Claude (automation), whereas users in economies with high per capita usage are somewhat more prone to prefer more collaborative or learning-based interactions with Claude (augmentation), even when controlling for the duty mix. Much like the local specialization in task use, the local specialization in AI collaboration patterns suggests that impact of AI might be very different in several regions.

The geographic patterns of AI adoption—where it’s used, for which tasks, and the way—suggest that with a view to realize the potential of AI to profit people across the globe, policymakers have to listen to local concentration of AI use and adoption, and address the chance of deepening digital divides.

1. For privacy reasons, our automated evaluation system filters out any cells—e.g., countries, and (country, task) intersections—with fewer than 15 conversations and 5 unique user accounts. For bottom-up request clusters, we’ve got a good higher privacy filter of a minimum of 500 conversations and 250 unique accounts.

2. Data on this section covers 1 million Claude.ai Free and Pro conversations from August 4 to 11, 2025, randomly sampled from all conversations in that period that weren’t flagged as potential trust and safety violations. The unit of statement is a conversation with Claude on Claude.ai, not a user, so it is feasible that multiple conversations from the identical user are included, though our past work suggests that sampling conversations at random versus stratified by user doesn’t yield substantively different results. Aggregate geographic statistics on the country and US state level were assessed and tabulated from the IP address of every conversation. For geolocation, we use ISO-3166 codes since our provider for IP geolocation uses this standard. International locations use ISO-3166-1 country codes, US state level data use ISO-3166-2 region codes, which include all 50 US states and Washington DC. We exclude conversations originating from VPN, anycast, or hosting services, as determined by our IP geolocation provider.

3. International locations use ISO-3166-1 country codes, which incorporates countries and a few territories.

4. Tier thresholds (quartiles) are based on countries with a minimum of 200 observations for the worldwide level, and on US states with a minimum of 100 observations for the US level. Countries with no observed usage are assigned to the Minimal tier since we have no idea in the event that they have exactly zero usage or little usage that our random sample didn’t capture. Future work, for instance using stratified sampling, will allow us to explore these patterns with higher accuracy given limited observations for smaller countries and states.

5. The world map relies on Natural Earth’s world map with the ISO standard perspective for disputed territories, which implies that the map may not contain some disputed territories. We note that along with the countries shown in gray (“Claude not available”), we don’t operate within the Ukrainian regions Crimea, Donetsk, Kherson, Luhansk, and Zaporizhzhia. In accordance with international sanctions and our commitment to supporting Ukraine’s territorial integrity, our services aren’t available in areas under Russian occupation.

6. “No data” applies to countries with partially missing data. Some territories (e.g., Western Sahara, French Guiana) have their very own ISO-3611 code. A few of these have some usage, others have none. For the reason that Anthropic AI Usage Index is calculated per working-age capita based on working age population data from the World Bank, and population data isn’t available for all of those territories, we cannot calculate the AUI for these territories.

7. When further investigating Utah’s activity, we discovered a notable fraction of its usage gave the impression to be possibly related to coordinated abuse. This can be reflected in a much higher “directive” automation rating than average. Nonetheless, we ran robustness checks and imagine that this activity isn’t driving the outcomes.

8. Requests were filtered to people who represent a minimum of 1% of requests at the worldwide level and 1% of the local level.

9. To isolate the connection between automation preference and Claude usage accounting for task composition differences, we do the next: First, we calculate each country’s expected automation percentage by taking a weighted average. For every O*NET task (e.g., coding, writing, or evaluation), we multiply that task’s share of the country’s usage by the worldwide automation rate for that task type (the proportion of that task that Claude completes via directive/feedback loop patterns globally). Summing these gives us the expected values for every country’s automation percentage given the country’s specific task mix. We then regress each the actual automation % and AUI on this expected automation %. The residuals from these regressions represent the variation in each variable that can’t be explained by task composition. By examining the connection between these residuals (often called partial regression evaluation), we are able to determine whether countries which have higher AI usage than their task mix would predict tend also to have higher-than-predicted automation.

Chapter 3: API enterprise deployment of Claude

Overview

Whether frontier AI capabilities make us more productive, reshape labor markets, and speed up growth will rely on when and the way firms decide to deploy AI. Even when businesses recognize the potential of AI, profitably adopting it might require costly restructuring of production processes, training recent employees, and other sunk-cost investments to facilitate effective deployment.¹

To know business adoption patterns of AI, we turn to a brand new data source: Anthropic’s first-party (1P) API customers—again counting on privacy-preserving methods.² Our API allows customers to integrate Claude directly into their very own products and applications, and charges by the token used, reasonably than a flat subscription fee. This represents a fundamentally different product experience to Claude.ai, which we focused on within the previous two chapters.

Institutional inertia, alongside fixed costs of adoption, suggests that early examples of enterprise use of AI is prone to be concentrated amongst specialized tasks where deployment is straightforward, capabilities are robust, and the economic advantages from adoption are high.

Indeed, we see evidence along these lines in the information presented on this chapter. Our evaluation uncovers several patterns:

• Businesses use Claude in similar but more specialized ways than individual users: Businesses concentrate use in tasks where AI deployment is well suited to programmatic access, like coding or administrative tasks. In comparison with Claude.ai users, businesses use Claude less for educational or creative tasks and in additional automated ways overall.
• API customers are inclined to prefer higher cost tasks: Despite tasks various dramatically in cost, the costliest tasks are inclined to have higher usage, suggesting that model capability, ease of deployment, and economic value of automation determine adoption far more than the fee of the interaction itself.
• Access to appropriate contextual information is required for classy deployment: We discover evidence of a vital potential bottleneck for the usefulness of AI for businesses. API customers that use Claude for complex tasks are inclined to provide Claude with lengthy inputs. This might represent a barrier to broader enterprise deployment for some essential tasks that depend on dispersed context that isn’t already centralized or digitized. Correcting for this bottleneck may require firms to restructure their organization, spend money on recent data infrastructure, and centralize information for effective model deployment.

Setting the stage: AI adoption patterns in public data

Before diving into our API data, it’s price grounding ourselves within the broader landscape of business AI adoption.In line with the Census Bureau’s Business Trends and Outlook Survey, AI adoption amongst US firms has greater than doubled previously two years, rising from 3.7% in fall 2023 to 9.7% in early August 2025 (Figure 3.1).³ Despite this rapid rate of growth, the overwhelming majority of firms within the US don’t report using AI of their production processes.

But these aggregate numbers mask large variation across sectors. For instance, in early August 2025, one in 4 businesses within the Information sector reported using AI, which is roughly ten times the speed for Accommodation and Food Services.⁴

The image from this public data is obvious: enterprise use of AI is growing rapidly, but we’re still within the early stages of AI adoption. Usage stays unevenly distributed across the economy, with the sectors most in a position to quickly adopt and profit from this technology doing so.

As we’ll see below, our 1P API data yields a complementary conclusion: early enterprise use of Claude is likewise unevenly distributed across the economy and primarily deployed for tasks typical of Information sector occupations.

Specialized use amongst Anthropic API customers

To investigate API traffic, we apply the identical privacy-preserving classification methods from previous chapters—categorizing anonymized API transcripts by O*NET tasks and right into a bottom-up taxonomy. The patterns that emerge show enterprise usage concentrated in specialized tasks particularly suited to automation.

Overall, software development dominates the landscape. Among the many top 15 use clusters—representing about half of all API traffic—the bulk relate to coding and development tasks. Debugging web applications and resolving technical issues each account for roughly 6% of usage, while constructing skilled business software represents one other significant chunk. Of note, around 5% of API traffic focuses specifically on developing and evaluating AI systems themselves (Figure 3.2).

But not all API usage is for coding. API customers also deploy Claude to create marketing materials (4.7%) and to process business & recruitment data (1.9%). These two categories reveal that AI is being deployed not only for direct production of products and services but in addition for talent acquisition and external communications.

The O*NET classification makes these patterns even clearer. Little lower than half of all API traffic maps to computer and mathematical tasks—greater than 8 percentage points higher than Claude.ai usage. Office and administrative tasks come second at roughly 10% of transcripts, reflecting their suitability for automation.

However, several interaction-heavy tasks distinguished on Claude.ai have a much smaller share in API usage: education and library tasks drop from 12.3% to three.6%, while arts and entertainment fall from 8.2% to five.2%.

In lots of cases nevertheless, occupational categories are reasonably close between Claude.ai and API data, suggesting that underlying model capabilities, reasonably than the particular product surface, drives adoption in lots of instances.

Occupational segmentation vs. task specialization

Despite serving different users with different interfaces, API and Claude.ai usage follows remarkably similar power law distributions across tasks. Amongst Claude.ai conversations, the underside 80% of task categories account for less than 12.7% of usage; for API customers it’s somewhat more concentrated at 10.5% (Figure 3.4). These extreme concentrations (Gini coefficients⁵ of 0.84 and 0.86) reveal massive variation in AI-task fit—the best-matched tasks see orders of magnitude more usage than poorly-matched ones.

The similarity across platforms is especially striking given their different user bases and use cases. Each converge on comparable concentration levels, suggesting a standard matching process between AI capabilities and associated economic tasks.

Tasks like code generation dominate because they hit a sweet spot where model capabilities excel, deployment barriers are minimal, and employees can adopt the brand new technology quickly. The long tail of rarely used tasks could reflect several aspects.⁶ For instance, some tasks are simply less common—debugging software happens much more often than negotiating circus contracts. The acute concentration also suggests the potential role of O-Ring⁷ forces: if a task needs a level of reasoning Claude cannot handle, internal data the firm cannot access, or regulatory approval that does not exist, any single barrier could prevent adoption.

Automation vs. augmentation amongst API transcripts

The clearest distinction between API and Claude.ai usage lies in how humans and AI divide the work. When businesses embed Claude into their applications, they largely delegate individual tasks reasonably than collaborate iteratively with models.

In our data, 77% of API transcripts show automation patterns (especially full task delegation) versus just 12% for augmentation (e.g., collaborative refinement and learning). Based on a sample of conversations from Claude.ai, the split between automation and augmentation is sort of even. Looking across economic tasks, the degree of Claude automation through the API is even starker: 97% of tasks show automation-dominant patterns in API usage, in comparison with only 47% on Claude.ai (Figure 3.6).

This makes intuitive sense. Programmatic API access naturally lends itself to automation: businesses provide context, Claude executes the duty, and the output flows on to end users or downstream systems.

This pattern echoes how economically consequential technologies change into transformative: becoming embedded in systems that permit employees access productivity gains while not having specialized skills. While each augmented and automatic approaches enhance human capabilities, system-level automation is prone to yield each larger productivity gains across the economy in addition to more significant changes within the labor market: Fully automating some tasks, changing which tasks are essential for various jobs, and even producing recent types of work altogether.

The more Claude does, the more Claude must know

Why do our API customers use Claude for some tasks greater than others? Beyond fundamental model capabilities, a potentially essential explanation is that it is less complicated to supply Claude with the data needed for successful deployment for some tasks than others.

For instance, if the goal is to have Claude refactor a module in a fancy software development project, Claude may have to read—or a minimum of explore—all the codebase to know which changes to make and where. For software development with centralized code repositories, access to this information is in principle straightforward.

For other tasks, the suitable context won’t be available, or it is perhaps difficult to access. For instance, asking Claude to develop a sales strategy for a key account might require Claude having access not only to information contained inside a Customer Relationship Management system, but in addition to tacit knowledge positioned within the minds of account executives, marketers, and external contacts. All else equal, lacking access to such contextual information will make Claude less capable.

We explore this query by the connection across tasks between the typical API input length (i.e., the context given to Claude) and Claude’s average output length (i.e., what the model produces in response).⁸

For every O*NET task in our sample, we calculate the typical input and output length of associated API transcripts. We then divide these values by the typical lengths across all tasks appearing in our sample. This produces an input token index and an output token index for every task. An index value of 1.5, for instance, implies that the API transcripts related to that task are 50% longer than the typical across tasks.

There may be considerable variation across tasks in how long Claude’s API outputs are. For instance, tasks on the ninetieth percentile of output length are greater than 4x longer than tasks on the tenth percentile. Table 3.1 provides example O*NET tasks, together with a Claude Sonnet 4 summarization of the group of tasks at that a part of the distribution.⁹ Figure 3.7 shows that output length varies systematically across occupational categories as well.

What stands out from Claude’s assessment of tasks is that longer output tasks are inclined to represent increasingly complex uses. In fact, output length doesn’t capture all dimensions of task complexity, but it surely appears to be a wise, easily measured proxy.

Because API customers are priced on the margin for each input tokens and output tokens, they’ve an incentive to optimize model prompting to attenuate each input and output tokens when using Claude. In turn, any systematic relationship between input length and output produced by Claude partly captures the underlying contextual constraints in deploying Claude for classy tasks. Stated in another way, API customers are incentivized to only provide Claude with barely enough context to perform their objective and no more. And so we study contextual requirements for tasks with various output length.

Looking across tasks, we see a really stable relationship between how much context API customers provide to Claude and the way much Claude actually produces. Across economic tasks, each 1% increase in input length is related to a less-than-proportional 0.38% increase in output length (Figure 3.7). This elasticity of 0.38 suggests that there are strong diminishing marginal returns in translating longer contextual inputs into longer outputs for these economically useful tasks.¹⁰

The upshot is that deploying AI for complex tasks is perhaps constrained more by access to information than on underlying model capabilities. Corporations that may’t effectively gather and organize contextual data may struggle with sophisticated AI deployment, creating a possible bottleneck for broader enterprise adoption—particularly for occupations and in industries where tacit, diffuse knowledge is crucial to business operations.

Cost per task and substitution patterns across tasks

API customers pay per token, creating variation in the fee of deploying Claude for various tasks. More sophisticated tasks will are inclined to cost more, given their higher input and output token counts. This variation helps us explore whether cost is a significant component in determining which tasks businesses decide to automate with Claude.

The information suggests it isn’t, a minimum of relatively speaking.¹¹ For instance, tasks typical of computer and mathematical occupations cost greater than 50% greater than sales-related tasks, yet dominate usage.¹² Overall, we discover a positive correlation between cost and usage: higher-cost tasks are inclined to have higher usage rates (Figure 3.9).

The positive correlation between cost and usage suggests that cost plays an immaterial role in shaping patterns of enterprise AI deployment. As an alternative, businesses likely prioritize use in domains where model capabilities are strong and where Claude-powered automation generates enough economic value in excess of the API cost.

While this positive correlation holds overall, we next ask whether demand for Claude capabilities is lower amongst otherwise similar but costlier tasks. With the essential caveat that this needs to be viewed as a preliminary exploration, that is what we discover.

Controlling for task characteristics, we discover that every 1% cost increase is related to a 0.29% reduction in usage frequency in our sample of API transcripts (Figure 3.10).¹³ While consistent with standard economic theory that higher prices result in lower demand, the implied increase in usage to a drop in cost is proscribed. In line with this estimate, a ten% cost reduction for a selected task would only increase usage by around 3%.

Other aspects, beyond the fee of using Claude for particular tasks, appear to matter more for patterns of use.

Conclusion

Our API data captures enterprise AI adoption in its early stages: highly concentrated, automation-focused, and surprisingly price-insensitive (a minimum of among the many tasks our API customers use Claude for).

The 77% automation rate suggests enterprises use Claude to delegate tasks, reasonably than as a collaborative tool. Such systematic deployment is prone to be a vital conduit by which AI delivers broader productivity gains throughout the economy. Given clear automation patterns in business deployment, this will also bring disruption in labor markets, potentially displacing those employees whose roles are almost definitely to face automation.

However the implications for the labor market aren’t entirely clear. As we document above, complex tasks require disproportionately more context. Such information could also be scattered across organizations. In such conditions, employees with tacit knowledge about business operations may stand to profit as complements to classy AI-powered automation.¹⁴ Understanding the uneven labor market implications of AI adoption is a vital area for future research.

Businesses trying to adopt AI effectively may have to restructure how they organize and maintain the data that frontier systems depend on. Whether today’s narrow, automation-heavy adoption evolves toward broader deployment will likely determine AI’s future economic impacts.

1. Within the presence of fixed costs of adjustment, the query businesses face isn’t necessarily if they may adopt AI, but when. See Hall and Kahn 2003, Adoption of Recent Technology: “Crucial thing to look at about this type of decision is that at any cut-off date the alternative being made isn’t a alternative between adopting and never adopting but a alternative between adopting now or deferring the choice until later.”

2. Data on this section covers 1 million transcripts from August 2025, sampled randomly from a pool of 1P API customers constituting roughly half of our 1P API usage. We proceed to administer data in line with our privacy and retention policies, and our evaluation is consistent with our terms, policies, and contractual agreements. Each record is a prompt-response pair from our sample period which in some instances is mid-session for multi-turn interactions.

3. Note that that is a unique measure of adoption than within the introduction to this report. Reported adoption by consumers and employees of AI reached 40% in 2024 whereas when measured on the firm-level, nine out of ten businesses within the US report not using AI.

4. The Business Trends and Outlook Survey (BTOS), published by the Census Bureau, is a good barometer of AI adoption by firms within the US. The survey query we use to measure AI adoption is “Within the last two weeks, did this business use Artificial Intelligence (AI) in producing goods or services? (Examples of AI: machine learning, natural language processing, virtual agents, voice recognition, etc.)”. See Crane, Green, and Soto 2025, Measuring AI Uptake within the Workplace for a comparison of BTOS with other measures of overall AI adoption amongst firms.

5. The Gini coefficient is a measure used to quantify inequality inside a distribution, similar to the distribution of task usage. It ranges from 0 to 1, where 0 represents perfect equality (every task has the exact same usage share) and 1 represents perfect inequality (where one task accounts for all usage, and each other task has none).

6. Power laws in economic settings are an empirical regularity with notable examples of Zipf’s law particularly. Models that generate this kind of consequence feature each underlying heterogeneity and intentional, optimizing decision-making. For more, see Gabaix 2016, Power Laws in Economics: An Introduction.

7. Kremer 1993, The O-Ring Theory of Economic Development.

8. API input length refers back to the text in API messages, system prompts, and any additional content sent to the model, including files and datasets relevant to the duty at hand. Output length refers to Claude’s generated response to an API call.

9. Claude was prompted to discover tasks on the tenth, fiftieth, and ninetieth percentile of the ONET task distribution with the minimal organization of “The columns needs to be ‘Example tasks’, ‘Index Value’, ‘Summary’ where you provide a summary”.

10. One other contributing factor might be the degradation in performance some models experience at longer context lengths. See Liu et al, 2023, Lost within the Middle: How Language Models Use Long Contexts.

11. The query we ask on this section is whether or not, all else equal, cost differences across tasks shapes relative usage patterns. That is different from studying whether overall Claude usage is sensitive to external competitive pricing pressures.

12. To see that that is the case, we first aggregate O*NET tasks that we discover in our API sample by broad occupational category to measure overall usage shares and the typical cost per task in each category. As with the input and output tokens reported by O*NET task, we normalize average cost per task by the typical value across tasks observed in our sample.

13. Controls include fixed effects for broad occupational categories in addition to collaboration mode shares by task from our concurrently sampled Claude.ai conversations. Because some tasks have censored collaboration mode shares, we also include indicators for whether that a selected mode has missing data. We restrict attention to the set of tasks identified in each our API sample and our Claude.ai samples.

14. For instance, see Ide and Talamaś, 2025, Artificial Intelligence within the Knowledge Economy.

Concluding remarks

This third iteration of the Anthropic Economic Index report captures AI adoption at a critical juncture. Existing capabilities of Claude and other frontier AI systems are already poised to remodel economic activity, given how broadly applicable the technology is. Rapidly advancing AI capabilities only reinforce the conclusion that immense change is on the horizon.And yet early AI adoption is strikingly uneven. Usage currently clusters in a small set of tasks, with strong geographic variation that is very correlated with income—particularly across countries. Such concentration reflects where AI capabilities, ease of deployment, and economic value align: coding and data evaluation have high usage, while tasks requiring dispersed context or complex regulatory navigation are further behind.

Early business adoption of Claude is directly each much like consumer use (coding is essentially the most common use for each), and different in several consequential ways. Specifically, with programmatic access to Claude through the API, businesses are inclined to use Claude with greater automation. Such systematic enterprise deployment reflects how AI is poised to reshape economic activity: increasing overall productivity, but with uncertain implications for those employees whose existing responsibilities have been automated.

These patterns risk creating divergence. If AI’s productivity gains concentrate in already-prosperous regions and automation-ready sectors, existing inequalities could widen reasonably than narrow. If AI automation improves the productivity of employees with tacit organizational knowledge—as a few of our evidence suggests—then more experienced employees could see rising demand and better wages at the same time as entry-level employees face worse labor market prospects.¹

Constructing on our previous releases, this iteration of the Index’s reports marks a big expansion in each scope and transparency. We at the moment are open-sourcing comprehensive API usage data alongside our existing Claude.ai consumer data (now including geographic breakdowns at state and country levels), all intersected with detailed task-level classifications.

By making this data public, we hope to enable others to research questions we’ve not considered, test hypotheses about AI’s economic impacts, and develop policy responses grounded in empirical evidence.

Ultimately, the economic effects of transformative AI will likely be shaped as much by technical capabilities as by the policy decisions societies make.

History shows that the patterns of technological adoption aren’t fixed: they shift because the technologies mature, as complementary innovations emerge, and as societies make deliberate decisions about their deployment. The patterns of highly concentrated use that we observe today may yet evolve towards a broader distribution—one which captures more of AI’s productivity-enhancing potential, accelerates innovation in lagging sectors, and enables recent types of economic value creation. We’re still within the early stages of this AI-driven economic transformation. The actions that policymakers, business leaders and the general public take now will shape the years to come back. We’ll proceed tracking these patterns as AI capabilities advance, and supply empirical grounding for navigating probably the most significant economic transitions of our time.

1. Brynjolfsson, Chandar, and Chen 2025, Canaries within the Coal Mine? Six Facts in regards to the RecentEmployment Effects of Artificial Intelligence documents clear evidence that entry-level employees with high AI exposure have had relatively worse employment prospects since late 2022. Setting aside questions of causality, the easy interpretation is that that is attributable to AI substituting for work previously done by early-career employees. Another interpretation is presented by Gans 2025, If AI and employees were strong complements, what would we see?: That relatively faster employment growth for knowledgeable employees reflects AI making such employees more productive and thus in high demand. Whether AI compliments or substitutes work is probably an important query that we hope our data will help answer.

Authors and Acknowledgments

Authors

Ruth Appel*, Peter McCrory*, Alex Tamkin*

Miles McCain, Tyler Neylon, Michael Stern

*Lead authors. Contributed equally to this report

Acknowledgements

Helpful comments, discussions, and other assistance: Alex Sanchez, Andrew Ho, Ankur Rathi, Asa Kittner, Ben Merkel, Bianca Lindner, Biran Shah, Carl De Torres, Cecilia Callas, Daisy McGregor, Dario Amodei, Deep Ganguli, Dexter Callender III, Esin Durmus, Evan Frondorf, Heather Whitney, Jack Clark, Jakob Kerr, Janel Thamkul, Jared Kaplan, Jared Mueller, Jennifer Martinez, Kaileen Kelly, Kamya Jagadish, Katie Streu, Keir Bradwell, Kelsey Nanan, Kevin Troy, Kim O’Rourke, Kunal Handa, Landon Goldberg, Linsey Fields, Lisa Cohen, Lisa Rager, Maria Gonzalez, Mengyi Xu, Michael Sellitto, Mike Schiraldi, Olivia Chen, Paola Renteria, Rebecca Jacobs, Rebecca Lee, Ronan Davy, Ryan Donegan, Saffron Huang, Sarah Heck, Stuart Ritchie, Sylvie Carr, Tim Belonax, Tina Chin, Zoe Richards

Citation

@online{appelmccrorytamkin2025geoapi,

creator = {Ruth Appel and Peter McCrory and Alex Tamkin and Michael Stern and Miles McCain and Tyler Neylon],

title = {Anthropic Economic Index report: Uneven geographic and enterprise AI adoption},

date = {2025-09-15},

yr = {2025},

url = {www.anthropic.com/research/anthropic-economic-index-september-2025-report},

}