H1 2025
Global AI Atlas / Historical evidence
Three years.
The evidence over time.
Explore country indicators, AI adoption and social participation across the three-year window in this edition. Every value keeps the period and definition its source actually reported.
Historical backfill2023–20262026-08.r2
Checked 2026-09-07
Checked 2026-09-07
Historical window: September 2023–August 2026. Complete annual waves overlapping that window keep their full original year. Source coverage varies: no values are filled in, and no historical atlas scores are reconstructed. This edition records evidence retrieved now.
Virgin Islands (U.S.)
AI diffusion · provider estimate
% · Working-age population, including people offline
H2 2025
Virgin Islands (U.S.)
Not availableQ1 2026
Virgin Islands (U.S.)
Not available0 reported periods for Virgin Islands (U.S.). A change is shown only for matching definitions and period lengths, without a reported source break.
| Source period | Virgin Islands (U.S.) | Evidence |
|---|---|---|
| H1 2025 | Not available | |
| H2 2025 | Not available | |
| Q1 2026 | Not available |
Definition and source.
Estimated generative-AI use as a share of the working-age population. Microsoft source series.
Provider estimate with device/population scaling and limited-sample adjustments; underlying telemetry is not publicly reproducible.
Retain the source report period; no conversion to 30-day survey adoption.
How far the evidence goes.
| Evidence | Native periods in this edition | Limit |
|---|---|---|
| Employment in agriculture (% of total employment) (modeled ILO estimate) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Access to electricity (% of population) | 2023, 2024 | Country coverage varies; annual published values. |
| GDP per capita, PPP (current international $) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Government Effectiveness - Governance score (0-100) | 2023, 2024 | Country coverage varies; annual published values. |
| Inflation, consumer prices (annual %) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Individuals using the Internet (% of population) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Life expectancy at birth, total (years) | 2023, 2024 | Country coverage varies; annual published values. |
| Population ages 65 and above (% of total population) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Population, total | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Lower secondary completion rate, total (% of relevant age group) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Unemployment, total (% of total labor force) (modeled ILO estimate) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Urban population (% of total population) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Vulnerable employment, total (% of total employment) (modeled ILO estimate) | 2023, 2024, 2025 | Country coverage varies; annual published values. |
| Level of water stress: freshwater withdrawal as a proportion of available freshwater resources | No years in the window | Country coverage varies; annual published values. |
| AI diffusion · provider estimate | H1 2025 · H2 2025 · Q1 2026 | Available country series starts in 2025; no 2023–2024 values. |
| Generative AI · survey | 2025 survey | Available survey starts in 2025. |
| Social networks · survey | 2023 survey · 2024 survey · 2025 survey | People participating in social networks in the three months before interview, as a share of all people aged 16–74. |
| Generative AI for work · survey | 2025 survey | Available survey starts in 2025. |
| Enterprise AI · survey | 2025 survey | Enterprises with 10+ persons employed using at least one covered AI technology. Adoption is not measured productivity. |
| Enterprise AI · 2023–2024 basket | 2023 survey · 2024 survey | Historical enterprise AI survey covering seven technology categories. Separate from the broader 2025 basket; no cross-basket growth rate is calculated. |
| Foreground attention per online person | Not available | Measured human foreground hours in a matched online population. App active-day minutes and elapsed web visits cannot be combined. |
| Verified net economic benefit | Not available | Realized incremental benefit after all attributable costs, under a documented counterfactual. |
Economic evidence / native study periods
What happens when people use AI?
Different tasks, different results. These studies are outside the country ranking.
Consulting tasks and the limits of assistance
Over 25% faster on tasks within GPT-4’s capability758 BCG consultants completing realistic knowledge-work tasks. The reported speed gain applies to the tasks within the tested model’s capabilities.
Working paper reported 21 September 2023 · journal publication March 2026
Performance depended on the task; assistance could worsen results beyond the model’s capabilities. This is historical, task-specific evidence, not a national productivity estimate or a claim about today’s models.
Read the study ↗ · 2026 publication ↗Experienced developers, familiar repositories
AI-allowed tasks took 19% longer16 experienced open-source developers; 246 issues in repositories they knew.
Early 2025 experiment · published July 2025
This setting does not represent all developers or current tools. METR’s February 2026 follow-up identifies selection effects and redesigns the study; do not extrapolate this effect to countries.
Read the study ↗ · 2026 publication ↗Customer support in one organization
15% more issues resolved per hour5,172 customer-support agents in the authors’ updated paper.
Study published 2025 · earlier company rollout
Effects vary by experience and skill. A company-specific productivity result is not measured national output or realized net savings.
Read the study ↗Countries and economies follow source reporting conventions; boundaries and listings imply no position on sovereignty. Missing polygons remain accessible through country search and the table. AI estimates, surveys and studies are separate evidence. No universal impact score is calculated.
Eurostat data selected and reformatted by humAIne with attribution. Microsoft data © 2026 Microsoft, reused under the MIT License. Interpretation is humAIne’s responsibility.