Microsoft's Global AI Adoption 2025 report introduces “AI diffusion” as a metric for examining how generative AI spreads across societies. The goal is straightforward: turn global AI usage into something measurable, comparable, and mappable.

As one of the first large-scale efforts to establish a global benchmark for AI usage built primarily on telemetry data, the report achieves several important things.

It offers clear charts, consistent rankings, and a structured way to compare countries at a glance. As a high-level snapshot, it provides a helpful starting point for thinking about global AI uptake.

At the same time, the report has sparked debate.

In China, official statistics for 2025 show that roughly 36.5% of the population uses generative AI, more than double the figure reported in the study (16.3%).

In Russia, Deputy Anton Gorelkin publicly questioned the findings, noting that Microsoft’s withdrawal from the Russian market left telemetry extremely limited, making it challenging to derive meaningful insights.

These responses highlight a broader methodological issue: the report focuses more on AI visibility than on AI integration.

More specifically, it tracks how AI usage appears across Microsoft’s observable channels and scales that visibility into a global diffusion narrative.

Visibility, however, is not the same as adoption, and this gap helps explain why national statistics, market reality, and lived experience can diverge so sharply from platform-based rankings.

Let’s take a structured look at the main methodological limits and how they shape the report’s conclusions.

The study relies primarily on telemetry from Microsoft products and extrapolates from this data to infer global AI usage. Data from Android, iOS, and macOS is incorporated indirectly through assumptions. 

In many regions, Microsoft's ecosystem no longer dominates, making its telemetry a weak proxy for real adoption patterns.

In places where most users opt out of sharing data, the technical paper adjusts AI usage estimates “by blending the observed country-level usage with the global average” — a method that flattens regional differences and undercounts actual adoption.

Large AI platforms operating outside Western ecosystems aren’t fully captured. For instance, while the study asserts that DeepSeek is the dominant consumer AI app in China, the actual market leader is reportedly ByteDance's Doubao, which exceeds 100 million daily active users.

Platforms like Doubao fall outside Microsoft's observable channels, not because of a small user base, but because the domestic ecosystem is integrated into areas the report doesn’t track.

The methodology does not capture corporate, industrial, or public-sector AI deployments—internal tools, on-prem systems, or custom applications —despite these often being the main drivers of AI integration and economic impact in many countries.

As a result, countries with strong domestic AI ecosystems can appear to have “low AI adoption” because their AI use is off-platform. That doesn’t make it absent.

In much of the world, AI is not consumed via standalone chatbots or websites, but embedded into super-apps, browsers, and platform ecosystems. 

That usage pattern doesn’t map cleanly onto tracked AI websites and apps, so it largely disappears from the dataset. Examples include Yandex Browser in Russia and Baidu’s Ernie Assistant in China, integrated across search, maps, health, and e‑commerce apps, all reaching hundreds of millions of users outside the sites the report monitors.

When AI is infrastructure, not interface, telemetry stops being a reliable lens.

In many markets, AI access happens through messaging and bots, not branded AI products. By excluding these integrations, the report privileges a particular usage pattern: consumer-facing, standalone, platform-branded AI interfaces.

For instance, WeChat’s Yuanbao assistant operates within the messenger, providing AI capabilities to millions without ever appearing on the report’s tracked websites or apps.

Countries with domestic AI platforms and active import substitution may appear to have “low diffusion” because usage occurs outside Western services.

This is especially evident in China and Russia, where large-scale local ecosystems dominate actual AI use, spanning consumer apps, enterprise solutions, and state platforms.

In regions where access to global AI services requires technical workarounds, user activity is often misattributed geographically.

This is particularly relevant for Iran, China, and Russia, where VPN usage is widespread and frequently necessary to access international AI platforms.

 As a result, domestic activity is recorded abroad, further depressing reported local adoption and distorting cross-country comparisons.

The authors admit that their AI user estimates rely exclusively on Microsoft telemetry: “AI User Share estimates are based on Microsoft telemetry data. In some countries — like Russia, Iran, and partly China, telemetry data is limited, and usage estimates may be incomplete.”

They acknowledge that this reliance “introduces potential biases related to this user base,” even as they present it as an “important new lens into how AI is spreading globally.” Despite these caveats, the report applies the data without meaningful correction to support broad global conclusions.

Only users who spend at least 90 minutes per month on the tracked services are counted, sidelining irregular, exploratory, educational, and early-stage use. Precisely the forms of interaction that often signal diffusion in emerging, fast-evolving markets. 

In addition, the analysis excludes users under 15 and over 64, omitting important segments such as highly active youth and older adults, further narrowing what counts for “adoption.”

By defining AI use as sustained engagement with a narrow set of services, the report favors a one usage model, long sessions, consumer interfaces, and branded platforms, while sidelining the more fragmented, embedded, or practical ways AI is often used.

The report captures AI activity as seen through Microsoft’s own channels — not the complete picture of how AI is adopted, integrated, or leveraged across industries, governments, and daily life. 

It reflects platform visibility rather than the real diffusion of AI across societies. That doesn’t make the dataset wrong, but it defines its scope.

The report is valuable as a first step toward building standardized global AI benchmarks.

Future analyses can expand on this foundation by incorporating domestic platforms, local ecosystems, enterprise deployments, public-sector AI, and embedded systems, providing a more complete and nuanced picture of how these technologies are truly reshaping economies and daily life.

Together, these data points would offer a more complete picture of how AI is diffusing across regions, not just where it is easiest to observe.

Hope you enjoyed this post. We’re resuming the AIport weekly newsletter after a long pause. Thank you for your patience and understanding.

There’s a lot of great content to follow. Stay tuned!