The claim is that 90% of working hours in the United States are spent on tasks that current AI benchmarks do not cover well. Benchmarks are tests used to measure how well AI systems perform on selected tasks. The claim suggests that results on those tests may not reflect the full range of work people do on the job.

The supplied research context does not name the source of the 90% figure, describe how it was calculated or define what counts as “not covered well.” It also does not establish whether the claim refers to all workers, particular industries or a specific set of benchmarks. The number should therefore be treated as a headline claim, not a verified measure of workplace AI coverage.

The topic is appearing in trend lists alongside recent headlines about new AI models and research into which jobs robots may do. That timing offers a clear connection: as companies announce more capable systems, questions about how to measure their usefulness at work become more prominent. The headlines provided do not confirm that any one of them caused the claim to trend.

The underlying issue is the difference between performance on a defined test and usefulness across a job’s varied tasks. A benchmark can assess particular abilities, while a role may involve many activities that the test does not represent. The available context supports that general concern, but it gives no examples of omitted tasks or evidence about how often AI is used in workplaces.

For readers, the figure is best understood as a prompt to examine what AI evaluations measure and what they leave out. The supplied material does not establish that AI can already perform 90% of workplace tasks, or that benchmarks miss exactly 90% of working hours. A source study and details of its method would be needed to assess either conclusion.