Roughly one in three working-age Americans now uses AI tools on a regular basis, according to a new population-normalized metric that ranks the United States 21st globally for AI adoption. That is up from 24th in the previous measurement period, a modest climb that still leaves the country nearly 39 percentage points behind the United Arab Emirates, where 70.1% of working-age residents actively use AI-powered applications.
The gap is striking. It suggests that despite the concentration of leading AI companies in Silicon Valley and Seattle, everyday adoption of the technology across the broader American workforce has been slower than in a handful of smaller, digitally aggressive economies.
Where the numbers come from
The rankings originate from a preprint research paper titled “Measuring AI Diffusion: A Population-Normalized Metric for Tracking Global AI Usage,” authored by researchers and hosted on the Cornell University-operated repository arXiv under identifier 2511.02781. Rather than relying on self-reported surveys, the researchers built their metric from telemetry data, capturing actual interactions with large language models, AI coding assistants, and other AI-powered productivity software across dozens of countries.
That methodological choice matters. When workers are asked whether they use AI, answers vary depending on how the question is framed and whether people recognize the tools they already rely on as “AI.” Telemetry sidesteps that ambiguity by measuring behavior directly.
Microsoft’s Q1 2026 AI Diffusion report cited the arXiv paper, linking the tech giant’s own industry analysis to the academic methodology. The preprint carries a stable DOI, making it citable and trackable, though it has not yet undergone formal journal peer review. That means the findings reflect a credible academic submission, not a journal-endorsed consensus figure. Readers should treat the specific percentages as outputs of one well-described model, not settled benchmarks.
What the US figure actually tells us
A 31.3% adoption rate means that roughly two out of every three working-age adults in the United States are not regularly interacting with AI tools, at least as this study’s telemetry captures them. The three-rank improvement from 24th to 21st signals measurable progress, but the absolute number remains below the one-third mark.
For context, other surveys frame adoption differently and produce different numbers. The McKinsey Global Survey on AI, published in mid-2024, found that 72% of organizations worldwide reported adopting AI in at least one business function. Stanford’s HAI AI Index tracks a similar organizational-level metric. The gap between those figures and 31.3% is not a contradiction; it reflects the difference between a company deploying AI somewhere in its operations and an individual worker actually using it day to day. Both lenses are useful, but the arXiv metric zeroes in on the worker, not the firm.
What the study does not provide is a breakdown by industry, occupation, or age group. Readers in healthcare, manufacturing, or education will naturally wonder whether their sector sits above or below the national average. That granularity is not available in the current dataset, and filling it in will likely require follow-up research or complementary data from agencies like the Bureau of Labor Statistics, which has not yet published a parallel measure.
Why the UAE sits so far ahead
The UAE’s 70.1% adoption rate tops the global table, and the reasons are partly structural. The country has a working-age population of roughly 8 to 9 million, a fraction of the US labor force that exceeds 160 million. Smaller populations can achieve high per-capita technology penetration faster, especially when backed by aggressive state investment.
The UAE government has been explicit about its ambitions. Its National AI Strategy 2031 outlines plans to embed artificial intelligence across government services, healthcare, transport, and education, supported by dedicated funding and a cabinet-level Minister of State for Artificial Intelligence, a post created in 2017. That institutional commitment helps explain the high adoption figure, though the arXiv paper itself does not attribute causation to any specific policy.
The comparison between the two countries is informative but not perfectly equivalent. Policy levers available to a compact, wealthy Gulf state differ sharply from those facing a continental economy with wide variation in broadband access, industry mix, and workforce education levels. Policymakers reading the rankings should focus less on replicating a specific national model and more on identifying which underlying conditions, such as digital infrastructure, cloud availability, and skills training, are realistically transferable.
Open questions about the data
Several aspects of the methodology remain unresolved. The preprint describes its telemetry approach in broad terms but does not publish the full list of countries ranked, the exact data sources used, or the precise collection window. No independent research team has yet replicated the results using a competing dataset.
One technical question stands out: how the metric handles users who interact with multiple AI tools. A software developer who uses a coding assistant, a chatbot, and an AI-powered design application could be counted once or multiple times depending on the deduplication method. The abstract does not specify this, and the choice directly affects every country’s reported percentage.
The relationship between the arXiv paper and Microsoft’s own telemetry also deserves scrutiny. The research record confirms that Microsoft cited the paper, but whether Microsoft product usage data forms a significant share of the telemetry inputs is unclear. If it does, the metric could tilt toward economies where Microsoft tools dominate, potentially underrepresenting AI usage that flows through Google, Baidu, or open-source platforms. Without a transparent breakdown of data contributors, the indicator’s evenness across competing ecosystems cannot be fully assessed.
ArXiv itself is supported by member institutions and individual donations, which keeps access free but can limit resources for post-publication data curation. If the authors later revise country-level figures, those corrections may not propagate quickly into corporate reports or policy briefs that already relied on the original version.
Where the US adoption race stands heading into mid-2026
The clearest takeaway is that AI adoption is measurable, uneven, and still accelerating from a relatively low base in most large economies. The United States has moved up the rankings, but the distance to the frontier suggests that early adopters elsewhere are experimenting at a considerably faster clip.
For American employers weighing AI training investments, the 31.3% figure offers a useful baseline: a majority of the workforce has not yet integrated these tools into daily routines, which represents both a gap and an opportunity. For policymakers, the ranking is a prompt to ask harder questions about broadband access in rural areas, AI literacy in public education, and whether federal workforce programs are keeping pace with the technology’s spread.
The arXiv indicator is not the final word on any of this. It is one lens, built on telemetry rather than surveys, normalized for population rather than GDP, and transparent enough to invite the scrutiny it deserves. The real test will come when independent teams attempt to replicate it, and when official statistical agencies begin publishing their own worker-level AI adoption figures. Until then, the numbers point in a clear direction: the US is gaining ground, but the finish line keeps moving.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
