The open Atlas compares work-related AI use across countries and occupations. A linked science study reports time savings, but also finds that checking results and physical work still slow research.
New to this? Read it in simple words
- Google updated its AI and Economy Atlas with a public, interactive tool. It compares work-related AI use across countries and jobs.
- Google reports creative jobs make up 19 percent of work-related AI use in India. That is 1.6 times the global average.
- In a linked survey, scientists said they saved just under seven hours each week. But checking results and waiting for experiments still slowed their work.
- The data comes from Google’s own AI products, not the whole economy. It does not show how many jobs were improved, changed or removed.
- De-identified
- Personal details are removed, so the data does not show who a person is.
- Bottleneck
- The slowest step in a process, which slows down all the work.
The Atlas makes differences easier to explore
Google’s Atlas groups large numbers of AI interactions by task, occupation, country, and language. The September update adds an open interactive explorer. Readers can compare patterns instead of seeing only one headline number.
Google reports that creative occupations form 19 percent of work-related AI use in India. That is 1.6 times the global average. In the United States, computer and maths occupations form 30 percent, about twice the share elsewhere.
These figures show the mix of activity inside the dataset. They do not mean that 19 percent of Indian creative work uses AI. They also do not show how many jobs were improved, changed, or removed.
The Atlas covers Google product use. Its science survey reports saved time and new bottlenecks that still need study.
The science study finds speed and new bottlenecks
A linked study combines more than 2,600 specialised AI models with a survey of over 600 scientists in the United States and United Kingdom. Nearly half of those surveyed said they use AI every day.
Scientists reported saving just under seven hours each week. Yet saved time did not remove every delay. Researchers still had to check results, choose among more ideas, and wait for physical experiments or clinical work.
This is a useful reminder that faster digital work may move the bottleneck. A model can help find a pattern, but a laboratory must still test it. A larger list of possible ideas can also create more review work.
The method should travel with every chart
Atlas is based on de-identified use of Google AI products. People who use other services, avoid AI, or lack internet access are not fully represented. Country and job comparisons should keep that limit visible.
The explorer is still valuable because readers can inspect more than one result. It can help researchers ask why adoption differs and which tasks people choose. It should not be used alone to claim that AI caused an economic change.
Future updates should show how the sample changes over time and how categories are checked. Good economic evidence needs clear definitions, stable measures, and links to independent labour, wage, and productivity data.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3Research · September 2026AI in Science: Adoption, impact and the changing research process Google AI
We used Google’s update for the new comparisons, opened the public Atlas, and read the linked science study. We describe the figures as Google research and survey results. We do not treat use of Google products as a measure of all AI use or as proof of job impact.