Two AI-adoption surveys show usage jumping over time — is that real growth, or did the earlier survey just undercount?

This explores whether Bick, Blandin and Deming's earlier surveys missed how many people were using generative AI. The corpus doesn't show that they did. What it shows is two snapshots taken at different times, and the gap between them reads as adoption growing, not as an earlier mistake being fixed. The earlier round surveyed more than 10,000 US adults. It found nearly 40% using genAI overall and 27% of workers using it on the job, which roughly matched how fast the PC spread at the same age How fast is generative AI adoption compared to earlier technologies?. By August 2025, overall use had reached 54.6% and work use 37.4% Is generative AI adoption now outpacing the personal computer?. Nothing in these notes says the first numbers were too low when they were collected.

The more interesting change is in the comparison, not the headline figure. In the first round, work adoption was keeping pace with the PC's early history. A year later it had pulled ahead: 37.4% against the PC's 25.1% at the same point after launch. That matters because the earlier finding was often read as 'genAI at work is spreading like any other big technology.' The later data says it sped up. The first survey also explains why use outside work led the way: free or cheap access and almost no technical barrier How fast is generative AI adoption compared to earlier technologies?. Workplace use seems to have caught up as employers and workflows adjusted.

If you want a real case of measurement methods disagreeing, look at productivity rather than adoption. The same authors turned workers' self-reported time savings into an estimated 1.1% boost to aggregate productivity. That is above Acemoglu's 0.7% estimate based on expert judgment, and they put the gap down to method, not a real disagreement What is generative AI's actual aggregate productivity gain today?. Pooling later survey rounds raised the estimate to as much as 1.3%, which fits the acceleration in measured US productivity since ChatGPT launched Did generative AI actually boost productivity growth?. So how you measure AI's impact does change the answer. It just shows up in the productivity figures more than in the adoption counts.

Here's the twist that matters most: a high adoption rate doesn't mean people are happy with the tools or that the tools work well. Among developers, Stack Overflow found 80% using AI tools while trust in their accuracy fell from 40% to 29% Why do developers keep using AI tools they don't trust?. Research borrowed from aviation explains why gains can be patchy even when nearly everyone uses AI. Workers shift from doing tasks to checking AI output, their flow gets broken up, and easy tasks get easier while hard ones stay hard Does generative AI productivity actually improve or decline in practice?. So a survey can count usage accurately and still tell you little about how much that usage is worth.

If you specifically want to know about survey bias, such as who answers, how 'use' is defined, or whether people under-report, this corpus doesn't cover it. The notes treat the rising figures as real growth, not as a correction to the earlier ones.