The finding that complicates the advice
Every AI adoption playbook says the same thing: leaders should be seen using it. Model the behavior, remove the stigma, and the organization follows.
Then there is this. 19% of workers say their trust in leadership drops when leaders lean heavily on AI. Split by level, that is 26% of individual contributors against 9% of directors and above. A quarter of the people the advice is aimed at, moving the wrong way — and near-agreement among the people already sold.
It is not anxiety. It is arithmetic about what can be seen.
The easy read is that individual contributors are more anxious about AI generally, and some of that is real. It does not explain the shape.
Everyone is watching the same event, and only some of them can complete it. What a room can observe is the tool being used and an output coming back. What it cannot observe is the deciding — what got rejected, what got kept, what the person would have written differently. That part never was visible, with or without AI.
Someone who makes decisions like that themselves supplies the missing half automatically. They have stood in that spot, so they fill in the judgment they cannot see with the judgment they know is there. Someone whose job does not include that call has nothing to fill it in with. They are left with what is on the screen.
That is not a difference in attitude. It is a difference in what each person can supply from where they stand.
Which makes the advice much narrower than it sounds
The useful version is not being seen using it. It is saying what you used it for, where you stopped, and what you checked before you believed it.
That version is harder to perform. It is also the only one that hands the room the part it was never going to see: the tool was used, and the responsibility did not move.
What to do with it this week
The instinct after reading this is to use AI less visibly, and that is the wrong correction — invisible use is its own problem, and a worse one. The move is to change what is visible.
Next time you share something AI helped produce, add one sentence naming the judgment you applied: the thing you rejected, the claim you went and checked, the part you rewrote because the output was not good enough. One sentence. It costs nothing, and it supplies exactly the half the room cannot infer for itself.
And if you want to know whether this is live in your organization rather than theoretical, the number is askable. The people furthest from the decision are the ones whose answer matters, and they are the least likely to volunteer it.
Deciding what your team should see, and what stays a human call, is structural work: See how we approach it →
Keep reading: Six hours a week correcting AI · The bar you set is the bar you get
This is the thinking behind a Trusted AI Culture.
Sources: SHRM, "Navigating AI in the Workplace" (2026). All figures are for workers who use or have used AI at work.