Adopting Innovative Technologies in Israel: The Claude Case Study in a Global Perspective

Roy Zilber, Prof. Eviatar Matania, and Prof. Udi Sommer [1]

Background

A country’s adoption of new technology, and the way broad audiences use it across the governmental, security, and private-business sectors, is an important component of a state’s technological power. The quality of healthcare, the efficiency of transportation, smart energy, the efficiency of municipal services, and the technological edge of militaries are all examples of what broad adoption of innovative technologies can advance. Sectors that fail to do so (or states that lag behind) begin to fall behind in the global race for technological power, with all the attendant consequences for GDP, GDP per capita, the range and quality of services, and for state power itself (both in the narrow military sense and in the broader sense of national security and the economy).

Various obstacles stand in the way of technology adoption: regulatory (such as outdated service licensing), structural-institutional (such as market structure and its incumbents), technological (legacy systems that do not allow for innovative adoption), and the interests of pressure groups. Yet one obstacle precedes all of these and can be considered the most fundamental: the degree of willingness and ability of populations to adopt innovative technologies, which depend primarily on education and culture. This component is hard to measure because it sits behind all the other obstacles, but there are cases in which it stands out and can be measured free of other noise. A striking example is the rapid adoption of drones in the IDF after October 7. Despite the military's initial lag in battlefield deployment, reserve soldiers quickly drove their widespread operational use, forcing the military establishment to formally integrate the practice after the fact. This is evidence of a population that overcame the obstacles out of necessity and carried out a high-quality diffusion and absorption of a technological product, independent of the slower establishment.

A clear global case in which one can directly measure a population’s diffusion and absorption capacity is the LLM engines developed worldwide, whose use depends on neither a regulator nor a lobby but almost exclusively on the will and ability of the individual, or of relatively advanced organizations. This paper presents the issue in an international comparison based on an analysis of raw datasets and of reports published by Anthropic over the past year. The reports include an examination of the penetration of the Claude engine in various countries and the quality of that penetration. These reports were analyzed for 14 countries of various types and sizes, as detailed below. Claude’s penetration rate and the quality of its use serve here as a partial but illuminating measure - both of the penetration of LLMs into broad use and, more generally, of technological adoption within a country.

Data and Measures

The analysis is based on three raw datasets (August 2025, November 2025, and February 2026; approximately 2.95 million cumulative conversations) and on interpretive reports published by Anthropic over the past year. Two measures are used throughout the paper. The first, the penetration index (AUI - Anthropic AI Usage Index), is computed as the ratio between a country’s share of conversations on Claude and its share of the working-age population (15–64): a value of 1 means use proportional to the country’s size, and a value greater than 1 means use higher than expected. It is therefore a per-capita scope measure. Second, the augmentation rate, measures the quality of use - the share of conversations in which the user and the model operate in a collaborative-iterative loop (iteration, learning, and validation), as distinct from automated use in which an entire task is delegated. A high augmentation rate indicates a mature, skilled user population.

Three methodological caveats are required up front. First, the data reflect use of Claude alone, not of generative AI models as a whole. Thus competitors (ChatGPT, Gemini, and Chinese models) are not represented, and countries whose access to Claude is restricted (China, Russia, Iran, and North Korea) are absent. That said, inferring broad adoption patterns from a single engine is reasonable in this case, given Claude’s leading standing for professional use worldwide. Second, the conversation samples were collected within a one-week window at each point in time, and should be interpreted as an indication of the period rather than as a fixed figure. Third, and particularly relevant to comparisons across time segments, the addition of Max users in the February 2026 dataset skews the sample composition toward intensive professional users and may explain part of the changes between the time points.

International Comparison: Penetration and Quality

Graph 1 presents Claude’s penetration dimensions across 14 countries at three points in time over the past year; Graph 2 presents the quality of use.

Graph 1: Claude’s penetration index (AUI) in 14 countries at three points in time (August 2025, November 2025, February 2026).

Graph 2: Quality of penetration (augmentation rate) in 14 countries over time (August 2025, November 2025, February 2026). A high, stable rate is an indicator of a mature user population.

As can be seen, technological diffusion is not a one-dimensional phenomenon. Two countries may exhibit identical adoption rates yet use technology at different levels of quality. Anthropic’s data make it possible to examine the two dimensions separately - penetration (AUI) and quality of penetration (augmentation rate). Israel and Singapore, for example, are high on both dimensions, whereas Germany and Japan are strong on quality of penetration, but their penetration rate is rising at a more moderate pace. At the other end are countries such as Brazil or India, where both the pace of penetration and its quality are low.

Graph 3 shows the position of the various countries on the two axes at the February 2026 time point, as well as the “volume” of absolute use (bubble size):

Graph 3: Three-dimensional diffusion profile - penetration rate and its quality (AUI × augmentation) and share of global use (February 2026).

Several clear profile groups emerge from the graphs, distinguished by the combination of penetration scope and quality of use.

High and high-quality penetration - small countries with human capital: Israel, Singapore, Switzerland, the Netherlands. Small bubbles in the graph but positioned in the upper-right corner (high penetration, high quality). Their AUI nearly doubled (Switzerland 2.81 → 4.85; Netherlands 2.56 → 4.02), and their global share grew by 29%–40% from the August 2025 level.

Volume powers - large in absolute terms, mid-tier per capita: the United States, India. Giant bubbles (22.2% and 6.16% of global use), but with an entirely different quality profile. The United States, with a penetration of 4.58 and fairly high quality - a “volume power with good quality.” India, with a very low penetration of just 0.28 and a quality of 50% (the lowest in the sample), where the Indian volume stems from population size rather than depth of adoption. India is a model of “initial adoption” confined to limited centers only.

Medium penetration at high quality - Western countries and developed Asia: the United Kingdom, France, Germany, Japan, South Korea. Mid-sized bubbles, high quality (57–63%), and medium penetration (1.81–3.71). The Japanese case presents a unique picture: relatively low penetration (1.81) but one of the highest augmentation rates in the world (63.1%) - that is, not many Japanese currently use Claude, but those who do, do so at high quality.

The division above can also be seen in the following graph when examining a sample of approximately 116 countries:

Graph 4: Two-dimensional diffusion of Claude use in 116 countries (February 2026). Per-capita penetration (AUI, logarithmic axis) versus quality of use (augmentation rate), divided into four quadrants by AUI=1 and the augmentation median (54.4%).

Dividing the countries into the four quadrants reveals that penetration of use and its quality sort the world into distinct country types. The upper-right quadrant (high penetration–high quality) gathers developed, human-capital-rich economies - from Singapore, Israel, and Switzerland to the United States, the United Kingdom, and Japan. The upper-left quadrant (low penetration–high quality) includes developing countries with a small but educated user base, such as Kuwait and Kazakhstan. The lower-left quadrant (low penetration–low quality) concentrates most of the developing world, including India and Brazil despite their absolute volume. The lower-right quadrant (high penetration–low quality) is almost empty - only four countries.

Israel

Israel leads globally in the AUI[2] throughout the entire period. First in August and November 2025 (7.00 and 4.90) and second in February 2026 (5.20), with Singapore narrowly surpassing it (5.53). The downward trend is not a retreat but rather “staying in place,” as noted.

Table 1: The Israeli profile at three points in time. Usage-by-purpose data were not available in August 2025.

The three measures in the table move in different directions, and distinguishing among them is essential. The AUI is a relative, per-capita measure: its decline from 7.00 to 5.20 is not an Israeli retreat but a result of other countries growing faster and closing the gap, while Israel “stays in place” around a high saturation it reached early. At the same time, the augmentation rate actually rose (54% → 58.7%) and stabilized - that is, the quality of use deepened even as the relative scope of use moderated. This opposing movement of the two measures is itself a miniature illustration of the central claim: scope and quality are separate dimensions.

Israel’s global share also declined (1.13% → 0.69%), and it should be interpreted with caution. It reflects first and foremost a relative effect - the rest of the world increasing its share at the expense of a country with early saturation. One should therefore not necessarily infer from it a decline in Israel’s absolute volume of use: the samples are weekly cross-sections of fixed size, so a declining share within a growing global pie is not the same as a decline in actual use. Alongside this, part of the change in February 2026 may also stem from a change in sample composition (the addition of Max users).

Human capital is the central mediating factor in a state’s capacity for rapid, high-quality uptake and absorption of AI: the correlation between the education level of the human input and the output level of the model exceeds 0.92.[3] The combination of conditions in Israel illustrates its leadership: an educated population, a technology sector large relative to the economy, a multilingual population (an advantage in working with models trained primarily in English) and a high cultural willingness to adopt new tools. Israel demonstrates that absorption intensity does not require inventing the products or the knowledge underlying the technology: combining imported foreign technology with a skilled local workforce can produce absorption superiority even beyond that of the country of origin.

Internal Diffusion within Israel

Beyond Israel’s national profile in international comparison, one can examine the diffusion of AI across different parts of Israel: is AI use in Israel concentrated in Gush Dan (Greater Tel Aviv), or does it disperse to other cities? (Graph 5).

There are three prominent patterns in the graph. First, Tel Aviv dropped dramatically from 50% to 37.2% (a drop of 12.8 percentage points in half a year); Jerusalem and Haifa grew significantly (Jerusalem 8.4% → 15.2%, Haifa 7.5% → 11.4%); the Central District grew moderately, and the South and North are stable at around 5% and 3.5%, respectively.

This may be a genuine lateral diffusion of AI use from Gush Dan to the major cities (a natural process following initial saturation in Tel Aviv), or it may be a sample-composition effect.[4] Raising the level of certainty requires an additional time point. Nonetheless, if the trend is real, it is significant. Tech hubs around the world tend to remain stable (Paris = 49% of France, Seoul = 56% of Korea, Tokyo = 39% of Japan). Israel is an exception to this trend, and may signal a broadening of AI use across Israel beyond the high-tech center in Gush Dan.

Graph 5: Intra-Israel diffusion: distribution of use by district (November 2025 → February 2026).

Summary

The adoption of advanced AI models around the world is undergoing rapid change. Nonetheless, initial trends point to several interesting observations. First, model penetration and the quality of use do not necessarily go hand in hand, and there is wide variation among countries even when their economy or GDP per capita is similar. Second, the variation among countries remains stable and high despite the relatively low barriers to use: the per-capita AI penetration in Israel and Singapore is roughly 20 times that of India and Nigeria. Third, the correlation between population education and the quality of AI use is high, and the quality of use is developing in all countries (even if at different paces) as users learn the models’ range of uses, and in line with the general population’s education. The gap between Israel and the leading AI countries is narrowing because Switzerland and the Netherlands, for example, are growing rapidly, while Israel “stays in place” around a high saturation it reached early.

These conclusions carry a direct practical implication for decision-makers in Israel. Israel’s AI advantage rests on human capital (not necessarily “technological” human capital in its traditional definition), on a cultural willingness to adopt technology, and on a broad user base - not solely on a developed R&D infrastructure. At the same time, the intra-Israel diffusion to Jerusalem and Haifa, if confirmed at an additional time point, points to a window of opportunity for broadening the user base beyond the high-tech center in Gush Dan. In broader terms, the Israeli case demonstrates that in the AI era, technological power is no longer measured solely by the ability to develop the frontier of science and technology, but also by the speed and depth with which an educated society can absorb it into routine - a development that grants small and medium-sized countries, such as Israel, a relative source of power in the international system.

Continued monitoring of these reports will make it possible to understand global trends comparatively across countries, alongside an advanced analysis of the dispersion of use at the local level, thereby revealing how different audiences and populations participate in the revolution.

[1] To quote this insight: Zilber, R., Matania, E., & Sommer, U. (June 2026). Adopting Innovative Technologies in Israel: The Claude Case Study in a Global Perspective (Elrom Aerial Insight 4/2026). Elrom Center for Air and Space Studies, Tel Aviv University.

[2]AUI values were computed from the Anthropic Economic Index [Dataset], within the Hugging Face repository. An AUI of 7.00 means that the share of Israeli conversations in the sample is seven times the share of Israel’s working-age population within the global population.

[3]The figure is based on Appel et al. (2026), Anthropic Economic Index report: Economic primitives, in which a cross-sectional analysis of the level of input and reasoning in conversations by country was conducted, cross-referenced with international education measures (Barro-Lee, OECD).

[4]The addition of Max users in the February 2026 data, which may have skewed the distribution of users.