You can tell in about two minutes per person, with no survey software. Ask each person whether they used AI on real work in the last five working days, on how many of those days, and which step of the job it sits in. High on the first count alone means your AI use is widespread. High on all three means it is deep.
Across the business teams I train, an owner's answer to "is your team using AI?" is usually a feeling. The feeling comes from the two or three people who talk about it most, and it is wrong in one direction or the other. This article swaps the feeling for three counts, gives you a one-page sheet to copy, and shows you how to read what comes back.
Why does the headline AI figure tell you so little about your own team?
Because the big surveys count businesses rather than people, and most ask one leader to answer for everyone.
The UK government's AI Adoption Research (Department for Science, Innovation and Technology, published 28 January 2026) interviewed 3,500 UK businesses with five or more employees by telephone between February and May 2025. The respondent was the person responsible for technology. Among the businesses that used AI, the average share of staff using it was 30%, and 77% said fewer than half their staff were using it. In the same group, 53% said they used AI constantly. The report is candid about its own blind spot: "the survey will not provide insights into shadow AI adoption."
Put those numbers side by side. A business can truthfully say it uses AI constantly while most of its people never touch it, and the person answering may not know which people do.
The picture repeats elsewhere. The Small Firms Association's survey in Ireland, published 10 August 2026, covered 404 small businesses (fewer than 50 employees) and carried a warning about "shallow" adoption in its headline. It found that 92% of small firms use generic off-the-shelf tools such as ChatGPT and Microsoft Copilot, while 30% use AI models tailored to their business. In the United States, Intuit's 2026 AI Impact Report (12 May 2026; more than 34,000 small and midsize business owners across the US, Canada, the UK and Australia) found that 77% of US businesses used AI regularly by January 2026, a count that includes free tools and built-in features.
None of these numbers is wrong. None can answer your question, which is about the particular 24 people on your payroll. Having a tool is a fact about a business. Depth is a fact about one person's week, so you have to ask the person.
What do reach, rhythm and role measure?
They measure how many people use AI, how often, and whether it has a settled place in their work.
The Reach, Rhythm and Role Check is my suggested way to separate deep use from widespread use. Here is its definition: "The Reach, Rhythm and Role Check is three counts I suggest a business takes before it calls its AI use deep: reach, how many people used AI on their own work in the last five working days; rhythm, on how many of those days each of them used it; and role, whether each can name the step in their job where it sits and who checks the result. Deep use is where all three are high. Widespread use is where only reach is."
Reach is the share of the team who used AI on a real piece of work. Real means it went to a colleague, a customer or a file. A test prompt does not count.
Rhythm is how many of the last five working days a person used it. A habit shows up in a week long before it shows up in a quarterly review.
Role is whether the person can name the step in their job where AI sits and who looks at the result before it goes anywhere. Without a role, use depends on mood and memory. With one, it survives a holiday, a busy month and a change of staff.
Each survey above captures one of the three, usually from the leader's chair. The government's share of staff is a reach figure. Its "constantly" is a rhythm figure. The Irish split between generic and tailored tools is the nearest thing to role. That is my reading, not the authors' claim.
How do you run the check?
Ask every person the same four questions, one to one, and write down what they say. A conversation takes about two minutes, so a team of 25 fits into a morning.
Copy this sheet:
| # | Ask | Write down |
|---|---|---|
| 1 | In the last five working days, did you use AI on a real piece of your own work? | Yes or No |
| 2 | On how many of those five days? | 0 to 5 |
| 3 | Which step in your job is it part of? | Their own words. Leave blank if they cannot name one |
| 4 | Who looks at the result before it goes anywhere, and what do they check for? | A name and a check. Leave blank if either is missing |
Then tally three numbers:
- Reach = the number of Yes answers divided by headcount.
- Rhythm = the middle value of question 2 among the people who said Yes.
- Role = the number of people with both question 3 and question 4 filled in, divided by the number who said Yes.
Four habits keep the answers honest. Write down your own guess of the three numbers before you ask anyone, because the gap between your guess and the count is the finding. Ask one to one, from someone people trust, never through a team leader answering for the team. Ask for last week's example rather than an opinion. And say at the start that results are reported as team numbers, that nobody is graded, and that using a tool the company has not approved will not be punished in this conversation. If you skip that last promise, you will learn nothing about shadow AI.
What do the scores tell you to do next?
Read the lowest of the three, not the best, and match the pattern to one move.
These bands are my rule of thumb, not research findings:
| Count | Low | Middle | High |
|---|---|---|---|
| Reach | under a third of the team | a third to two thirds | two thirds or more |
| Rhythm | 0 or 1 days | 2 or 3 days | 4 or 5 days |
| Role | under a third of users | a third to two thirds | two thirds or more |
| Pattern | What it usually means | Your next move |
|---|---|---|
| High reach, low rhythm | Everyone has tried it and few kept it | Pick one recurring job for the whole team and use AI on it together for two weeks |
| Low reach, high rhythm | A few deep users and an island around them | Ask each of them to write their step down in a page, then pair them with two colleagues |
| High reach and rhythm, low role | Busy but unanchored; the checking is nobody's job | Write the step and the checker into the job itself, as in the margin of a job description |
| High on all three | Deep use | Now ask whether it works, using the Before Number Rule |
| Low on all three | Not started, or hidden | Take the readiness scorecard below and run one short first session |
If your team is reluctant rather than merely unpracticed, start with the Chore Rule for skeptical employees before you push for rhythm.
What does a shallow result look like in practice?
It looks flattering on reach and thin on role. Take a hypothetical 24-person insurance brokerage; the figures are invented for illustration.
The owner's guess was "nearly everyone, most days." The count came back like this. Seventeen of 24 said Yes, so reach was 71%, which is high. Among those 17, the middle value for days used was 2, which is the middle band. Only 5 of the 17 could name both a step and a checker, so role was 29%, which is low.
The owner was roughly right about reach and wrong about the other two. The lowest count is role, so that is where to work. She picks one job, client renewal letters, and sits with the two staff who already use AI on it. They write down the step and name the person who reads each letter before it goes out and what that person checks. Six weeks later she runs the same four questions and compares the numbers. If role climbs and rhythm follows, the work is paying off. If neither moves, the job was the wrong one, and she picks another.
What I See When I Train Business Teams
The same shape turns up again and again: two or three people using AI nearly every day, a larger group who tried it once or twice, and a leader who describes the whole team by the first group. Nobody is lying. The enthusiasts are simply the people the owner hears from.
My view comes from where I started. I was a teacher, an assistant headteacher and a Director of Digital Strategy in further education, and a teacher learns early that asking a class "does everyone understand?" tells you nothing. You ask someone to show you. The sheet above is that habit applied to a team. It is also why I say what I say about business: "The hard part of AI at work isn't the technology. It's helping people use it well, safely and every day. That's what an educator does."
What should you do this week?
- Today: write down your guess of reach, rhythm and role for your team. Put it in a drawer.
- Tomorrow: decide who asks the questions. It should be you or someone people talk to freely, not a manager marking their own team.
- By Thursday: hold the two-minute conversations. For a team that will not be candid face to face, put the four questions in a shared sheet instead and keep the names out of your tally.
- By Friday: tally the three numbers, compare them with your guess, and find the lowest. Pick one move from the table.
- Before you close the laptop: put a date six weeks away in the calendar to ask the same four questions again. A single reading tells you where you are. Two tell you which way you are heading.
Where to go next
If you want a wider picture of where your business stands, the Business AI Readiness Scorecard takes about five minutes and covers six areas in 12 questions. If the sheet shows people using AI but nobody anchoring it in the job, that is the gap my AI training for teams is built to close.
Sources and further reading
- Department for Science, Innovation and Technology, AI Adoption Research, GOV.UK, published 28 January 2026 (fieldwork February to May 2025; 3,500 UK businesses with five or more employees; the full report PDF gives the staff-share figures on pp. 6 and 23, constant use on p. 7 and the shadow AI limitation on p. 5).
- Small Firms Association (Ibec), Small Firms Association Sound Alarm Over 'Shallow' AI Adoption, 10 August 2026 (Ireland; 404 small businesses surveyed by Amárach).
- The Irish Times, Fewer than one in five small firms have built AI into their business models, survey finds, 10 August 2026 (definition of a small business as fewer than 50 employees).
- Intuit, 2026 AI Impact Report, 12 May 2026 (more than 34,000 small and midsize business owners in the US, Canada, the UK and Australia).
- On this site: Shadow AI: What to Do When Your Staff Are Already Using ChatGPT; How to Get Skeptical Employees to Use AI; How to Measure Whether AI Is Actually Working; Do You Need to Rewrite Job Descriptions for AI?.
Dan Fitzpatrick is The AI Educator, a Forbes contributor, international keynote speaker and bestselling author who advises governments and system leaders and has trained more than 150,000 people across 30+ countries. About Dan.
