Should CEOs Use AI Themselves? Why Executive Fluency Sets the Adoption Ceiling

By Victor Hoang · Updated Oct 11, 2026

Most CEOs will say AI is a top priority. Ask them when they last used it for real work, and the answer gets vague. Someone on the team has been trying things. A consultant ran a session. The CEO has read the briefings and seen the demos.

That gap matters more than it looks. Our position: a company rarely adopts AI further than its leaders personally use it, so a CEO who uses AI hands-on is not indulging a gimmick. It's the cheapest adoption lever a small or mid-sized company has.

This isn't the same argument as the one in the AI skills gap executives get wrong, which is about what skills the workforce needs. And it isn't about who runs the program day to day, which we cover in who should own AI in a small company. This piece asks a narrower, older question: does the person at the top have to be a practitioner, or can leadership be done from a distance?

Key Facts

• Gallup's Q4 2025 data, published in January 2026, found that 69% of leaders use AI at least a few times a year, compared with 55% of managers and 40% of individual contributors.
• In the same Gallup data, frequent use among leaders rose from 17% to 44% since Q2 2023, while individual contributors rose from 9% to 23%. The gap between the top and the bottom has widened.
• Gallup's February 2026 survey of 23,717 U.S. employees found that 78% of employees who strongly agree their manager actively supports AI use are frequent users, versus 44% of those who don't strongly agree.
• BCG's 2025 survey of more than 10,600 workers found that regular GenAI use among frontline employees has stalled at 51%, while more than three-quarters of leaders and managers use it several times a week.
• In the same BCG report, positive feelings about GenAI rise from 15% to 55% with strong leadership support, yet only about one-quarter of frontline employees say they receive it.

The Ceiling Rule

Here's the mental model we use: the Ceiling Rule. Every organization has a ceiling on how deeply it will adopt a new way of working, and the ceiling is set by the most senior people's direct experience of it.

Why would that be? Three mechanisms, none of them mysterious.

Judgment. Executives allocate budget, approve process changes, and decide which pilots continue. Someone who has never watched a model produce a confident wrong answer will either over-trust it ("just have the team automate it") or under-trust it ("I read these things hallucinate"). Both lead to bad decisions. First-hand use is how you calibrate.

Signal. Employees read their leaders closely. If the CEO says AI matters but never mentions using it, people draw their own conclusion about how seriously to take the instruction. If the CEO shares a prompt that saved her an hour on a board memo, the instruction becomes concrete and permission becomes real.

Friction removal. Most AI adoption stalls on things only senior people can fix: a data-access policy, an approval that takes three weeks, a metric that punishes the time spent learning. A leader who feels these frictions personally removes them faster than one who hears about them in a status update.

The same logic applies to every layer. A manager whose own use is shallow can't coach a team past the first plateau. We make the middle-layer version of this argument in middle management as the AI obstacle and opportunity. The Ceiling Rule just starts one level higher.

What the Data Says, and What It Doesn't

Two patterns show up in recent large surveys, and both fit the rule.

First, leaders are already ahead, and pulling further away. Gallup's trend shows frequent use among leaders going from 17% to 44% in a bit over two years, while individual contributors went from 9% to 23%. BCG found the same shape in a different sample: frontline use flat at 51% while more than three-quarters of leaders and managers use GenAI weekly or more.

Second, the support people feel from above tracks how much they use it. Gallup's 78% versus 44% gap for employees who strongly agree their manager supports AI use is large. BCG's jump from 15% to 55% positive sentiment under strong leadership support is larger still.

Now the honest caveat. These are survey correlations, not experiments. They don't prove that a CEO using ChatGPT causes a sales team to use it. Support from a manager can mean many things: permission, training time, tools, or just encouragement. And leaders may use AI more partly because their desk work (writing, planning, analysis) suits it, which Gallup itself notes.

So we're not claiming a law of nature. We're claiming something narrower and more practical. Leadership support is the variable that moves adoption in these datasets, support is cheap to give, and the form of support employees can least discount is a leader visibly doing the thing. That is a reasonable bet, with a very low downside.

Delegating AI vs. Leading It

Executives often hear this argument as "the CEO should become a prompt engineer." That's not it. The question is where on a ladder your own practice sits.

Most companies sit between Sponsor and Practitioner. The jump from Sponsor to Practitioner is the step with the best return, because it costs hours, not budget.

Notice what the table does not require: technical depth, building agents, or knowing how models work. It requires using the tools on your own job often enough to have opinions grounded in experience.

Where a CEO Should Actually Use It

Don't start with the most sensitive work. Start with tasks you do repeatedly, can judge the output of, and wouldn't mind getting wrong in a draft.

Two things to notice. Every row has a "keep human" column, which is where your judgment gets trained. And none of these require sharing sensitive data with a tool you haven't cleared, which brings up the one genuine limit: use approved tools, and follow the same data rules you set for everyone else. A CEO who breaks the policy to try something new teaches the wrong lesson.

The Objections, Honestly

"My time is the scarcest resource in the company." True, and the commitment is smaller than it sounds. Thirty minutes a few times a week on tasks you already do is enough to build real fluency. If a tool can't save you that much on your own work within a month, that's information too.

"It's a gimmick. A figurehead playing with a chatbot." It is, if the use is performative: a screenshot in a town hall and nothing after. The test is whether your own workflow changed and whether you can tell the team what stopped and what started. People can tell the difference quickly.

"I have people for this." You do, and you should. The point isn't to replace the owner or the champions. It's that your decisions about their work improve when you know what the work feels like. We've written about how peer-led programs spread skills, and peers carry far more weight when the top isn't visibly exempt.

"I'll look unsure or behind in front of my team." Occasionally, yes. That's the actual cost, and it's also part of the benefit. A leader who says "I tried this, it got the second paragraph wrong, here's how I fixed the prompt" models the calibrated skepticism that every employee needs. It also lowers the fear that struggling with AI is a personal failing, which matters given how much anxiety sits underneath adoption. How leaders respond to AI job anxiety goes deeper on that.

"The tools change every month." They do, and that argues for the habit, not against it. Fluency is less about any one product than about knowing what to ask of it and how to check the answer. The same shift in mindset we describe in moving from AI tool to AI teammate comes only from repeated use.

An Illustrative Example

Illustrative scenario, not a real company. A 90-person services firm has an AI owner, a small budget, and a Slack channel nobody posts in. The CEO is supportive but hasn't used any of the tools. Usage hovers around a quarter of staff.

Over a month, the CEO starts drafting her weekly leadership note with AI, uses it to summarize two long client contracts, and asks it to argue against a pricing change she's considering. She shares one failure (it invented a clause that wasn't in the contract) and one win (the pricing counterargument surfaced a customer segment she'd overlooked). In the monthly meeting she asks every department head to bring one task they rebuilt around AI and one thing that went wrong.

Nothing about the tooling changed. What changed is that the question "is this expected here?" got answered. Department heads who had been waiting for permission started bringing examples. The Slack channel finally has traffic.

What to Do Next

This fits into a month.

1. Place yourself on the ladder. Be honest: Spectator, Sponsor, Practitioner, or Redesigner? Ask two direct reports to place you independently and compare.
2. Pick three recurring tasks from the table above and use an approved tool for each, weekly, for four weeks. Keep a short note on what saved time and what needed correcting.
3. Share one failure and one win with the leadership team. Specifics beat enthusiasm.
4. Redesign one of your own recurring processes (the board pack, the weekly review, the hiring screen for your direct reports) and publish the before and after.
5. Remove one friction only you can remove. A policy, an approval step, a metric. Announce that you did it.
6. Set the expectation downward. Ask each of your direct reports to do the same exercise, and revisit the results at the next leadership meeting. For new hires, this is also the baseline they increasingly expect, as we cover in what AI-native new hires expect from employers.

The test is simple. If an employee were asked "does our CEO use this stuff?", would the honest answer be yes, and could they name one thing she changed? If so, your ceiling just moved. If not, the next budget line probably won't move it.

For the underlying leadership principle, see our primer on change management. Visible sponsorship is a long-standing theme there, and AI changes little about it.

Learn More

• The AI Skills Gap Executives Are Getting Wrong
• Who Should Own AI in a 50- to 200-Person Company?
• Middle Management: AI Obstacle or Opportunity
• Peer-Led AI Programs
• How Leaders Respond to AI Job Anxiety
• From AI Tool to AI Teammate
• Change Management