Only 5.5% of knowledge workers are actually good at using AI.
Two-thirds of them use it at least once a week.
I’ve been turning those two numbers over in my head since reading a June 2026 AI proficiency report, because they describe something I see constantly in financial services. Plenty of advisors have opened ChatGPT this week. Far fewer have changed how they work.
That’s what this post is about. Starting with AI has become easy. Getting good at it is a separate skill, and it’s built through repeated practice at specific points in your week, not in one impressive session.
Most people are using AI. Almost nobody has built it in.
The research surveyed 5,026 U.S. knowledge workers. 67% now use AI at least once a week, up from 55% in October 2025. Daily use climbed from 22.5% to 36.7%.
Adoption, more or less, is solved.
The same research also examined how well people actually use it. Only 5.5% clear the bar for proficiency, which the report defines as AI being embedded in repeatable workflows rather than pulled out for the occasional question. Another 73.5% sit in the group the report calls Experimenters: genuine usage, one-off tasks, nothing that compounds.
I see the same shape in financial services. An advisor rewrites an email. Asks a question they’d have Googled a year ago. Both are real uses of AI. Neither one changes the week.
Usage tells you someone opened the app. It tells you nothing about whether the work changed.
A good prompt is where this starts, not where it ends
There’s real progress worth naming here. The share of workers who can write an effective prompt went from 7% to 21% in eight months. It tripled.
But when participants were tested hands-on, only 3.8% could write workable instructions for an AI assistant, the kind of setup behind a Custom GPT or a Gemini Gem.
That gap is the part I find most useful. Asking one good question is a skill you can pick up in an afternoon. Giving a model enough direction to produce work you’d put your name on, over and over, is something else. It asks you to know what a good answer looks like in your own practice, to notice exactly where the output fell short, and then to fix the instruction rather than patch the answer.
That’s judgment, and it’s yours. The model doesn’t supply it.
It’s also why I don’t think AI training works as a single session. A workshop can create the first breakthrough, and those moments matter. Trust arrives later, on the fourth or fifth attempt, when an advisor sees it help again and starts to know when to push back on what they’re given.
A prompt gets you an answer. An instruction gets you a process.
What moving forward actually looks like
When I go back through the Optimal AdvisorAI Learning Center data, the pattern is consistent. Advisors start with the foundations: how models work, how to set them up, how to write a prompt that goes somewhere. Almost nobody stays there long. They move into client communication, meeting preparation, and business development, the parts of the job that quietly eat the week.
Something shifts when they stop opening AI because they remembered it exists and start reaching for it at familiar points instead. Before a client meeting. While shaping a message they’ve been avoiding. When a page of meeting notes needs to become follow-up.
None of those moments sound like much on their own. Put together, they’re most of the distance between the 73.5% and the 5.5%.
In our post-session surveys between 2024 and 2026, 93% of the financial professionals we’ve trained said they applied what they learned. That figure matters more to me than any satisfaction score, because application is the whole game. Nobody’s practice improves because they enjoyed a session.
So what does this mean for you?
First, don’t feel behind. If only 5.5% of knowledge workers have genuinely built AI into their work, then almost everyone reading this is early, whatever the news cycle suggests.
Second, pick one point in your week rather than one new model to try. The advisors making real progress aren’t chasing every release. They’ve found two or three places where AI reliably helps, and they go there on purpose.
Third, keep your hand on it. AI can draft the first version of a client email. Your judgment is what decides whether it goes out. That doesn’t change as the models improve, and I’d argue it becomes more valuable, not less, as more of the first draft gets automated.
A little focused practice each month will do more for you than a week spent reading about what’s coming.
The future isn’t about choosing between human intelligence and artificial intelligence. It’s about blending the two. What I keep seeing in this data is that the blending is a skill, and like any skill, it belongs to the people who practice it.
If you’d like a structured place to build that habit, you can explore the AI Learning Center here: Optimal AdvisorAI Learning Center
