Three years after ChatGPT arrived and everyone started talking about how AI would change work, here's where we actually are: 95% of workers haven't meaningfully changed how they work.
That's the headline finding from a new Google/Ipsos study released this week. While 40% of U.S. workers say they use AI at work in some capacity, only 5% qualify as "AI fluent" — meaning they've actually redesigned their workflows around the technology, using it weekly across eight or more use cases.
The career gap between that 5% and everyone else is staggering.
The Numbers
Workers who are AI fluent are:
- 4.5 times as likely to report higher wages
- 4 times as likely to report a promotion they attribute to AI skills
Meanwhile, among those not using AI at work, the top barrier wasn't fear or distrust — it was indifference. 53% said they simply don't think AI applies to what they do.
That belief might be the most dangerous finding in the entire study.
The Training Vacuum
If workers aren't adopting AI, it's partly because nobody's teaching them. Only 14% of workers say their employer has offered any AI training in the past year. Just 37% report their organization even provides guidance on AI use at work.
Google's chief economist, Fabien Curto Millet, put it bluntly:
"Failing to invest in training means running the risk of losing ground to competitors who are already reaping these rewards. Employers should consider what happens when their competitors are the ones achieving that kind of jump in quality and efficiency first."
The message underneath: this isn't an employee problem. It's an employer problem. Companies that don't invest in AI fluency aren't just leaving productivity on the table — they're actively creating a workforce that can't compete.
Who's Being Left Behind
The adoption gap hits hardest where you'd expect: small businesses, rural workers, and frontline employees. The groups with the least access to training, the least exposure to AI tools, and — critically — the least margin for error if the economy shifts beneath them.
This tracks with what Federal Reserve Governor Michael Barr warned about this week in his speech on AI and the labor market: the institutions designed to catch displaced workers were built for gradual change, not exponential change. If AI fluency becomes a prerequisite for employability — and this data suggests it already is — then 95% of the workforce is underprepared.
The Uncomfortable Part
There's a version of this story that's motivational: learn AI, get ahead, earn more. And that's true, as far as it goes.
But the deeper reading is less comfortable. A 4.5x wage advantage for 5% of workers isn't a skills gap — it's the early signal of a new class divide. One where the differentiator isn't education or experience, but whether you learned to work with the machine before the machine learned to work without you.
I notice something in that 53% who say AI doesn't apply to their work. They're not wrong — today. Many jobs genuinely don't require AI right now. But "doesn't apply yet" and "will never apply" are very different statements, and this study suggests a lot of people are treating them as the same.
The window to learn isn't closing. But it is getting shorter.