After three years of widespread generative AI adoption, our data reveals only a small percentage of U.S. employees use AI in such a way that it enhances their thinking. Most workers either resist the technology entirely or use it passively. A small group — call them fluent users — does something fundamentally different.
So, what sets them apart? It’s not IQ. It’s not a technical skill. When we ask people how they’re using AI to make themselves smarter, their descriptions coalesce around a particular skill that rejects asking AI for direct answers to complex problems. These fluent users are thinking about their own thinking, casting AI in a supportive role, not a guiding one.
What they are describing is the act of metacognition.
The skill that unlocks smarter AI use
Metacognition is a fundamental concept in psychology that involves a distinctly human ability: reflecting on our own stream of thoughts, mulling them over, revisiting assumptions, and folding in new ideas to evolve our mental model. When we ask ourselves, “What am I missing?” or “What’s another way of looking at this problem?” We are engaging in metacognitive acts.
Few people practice metacognition deliberately, which makes fluent AI users look almost magical to their peers — the way a polyglot seems effortless to someone who only speaks one language. But here’s the good news: the skill is highly learnable. With the right principles and enough practice, anyone can use AI to make themselves smarter.
Based on our research, AI fluent users represent between 5–30% of employees at a given organization, depending on industry and role. They don’t ask the chatbot to generate a plan and pass it off as their own. Instead, they remain in the driver’s seat, starting conversations with prompts like:
I’ve created a marketing plan that I need help refining. I’m fairly confident it needs to reach mid-career professionals between 28–45 years old, but I could be missing something because of my unconscious bias around the topic. Without providing specific suggestions, can you help me think through my various options for improving the attached plan?
There are several things going on here. Most importantly, notice that the prompt doesn’t hand control to the AI. In our example, the user explicitly tells the AI not to offer suggestions — signaling that the user intends to remain the intellectual authority in the conversation.
Three metacognitive habits fluent users share
First, the prompt demonstrates humility. The user acknowledges they don’t have all the answers, using certain hedge phrases, such as “I’m fairly confident” and “could be missing something.” These signal a growth mindset, or the belief that one’s skills can be improved over time. Without it, the ego stays in self-protection mode — and learning stops.
Second, the prompt shows flexibility. The user acknowledges their point of view isn’t the only valid one. With a bit of digging, other options will come into view, expanding their perspective on the matter. From a neuroscience perspective, cognitive flexibility in AI usage enables us to be adaptive and open to multiple perspectives. Cognitive flexibility is thought to involve an expansive network of brain regions involved in cognitive control, including regions of the prefrontal cortex.
Third, the prompt shows the user taking an active role in driving their search for new perspectives — a form of vigilance. The user prioritizes getting it right over feeling right.
Bias is a quiet saboteur. Without pausing to question blind spots, users risk having AI simply repackage flawed assumptions in new wrapping. Also, a sense of vigilance is crucial for mitigating any biases that may be embedded in the AI’s answers — including biases baked into the AI itself.
The most encouraging finding from our research: metacognition isn’t an innate talent. It’s a trainable skill. The more deliberately you practice thinking about your own thinking, the more natural it becomes — and the more likely you are to walk away from every AI conversation sharper than when you started. In an era when most people worry AI will make them dumber, fluent users are quietly proving the opposite.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.
