Tie the tool to one task each person already does every week, give them a tested starting prompt inside the tool, make clear what they're allowed to use it for, and set aside time to learn. Before any of that, find out why they aren't using it: check the admin usage report and ask each person.
More reminders rarely help. Low use almost always has a specific cause: no obvious task, uncertainty about what's allowed, a bad first attempt, no time, the tool sitting somewhere nobody looks, or a worry about what using it says about them. Each has its own fix. Below: a quick diagnosis routine, the six blockers and their fixes, how to put prompts where people already work, a worked example with numbers, and the moves that make things worse.
Find out who isn't using it, and why
Start with the numbers, then talk to people. Most business plans show usage somewhere:
- Microsoft 365 Copilot: in the Microsoft 365 admin centre, under Reports, then Usage, there are Copilot reports showing active users, prompts and each person's last activity date. Names are concealed by default until an admin changes the report settings.
- Gemini in Google Workspace: in the Admin console, under Generative AI, the Gemini reports show active users, use per app and a per-person usage level.
- ChatGPT Business, Claude Team and others: look in the workspace or admin settings. What's reported varies by plan, so if there's nothing useful, ask people directly.
Then have a ten-minute conversation with each person who isn't using it, or barely is. Make it clear you're trying to fix what's in the way, not checking up on them. Four questions do most of the work:
- "What's the most repetitive writing or admin task in your week?"
- "Have you tried the AI tool on anything? What happened?"
- "Is there anything you're unsure you're allowed to use it for?"
- "What would have to change for you to try it on that task next week?"
Write the answers down. After five or six conversations, the same two or three reasons usually account for almost everyone. In an illustrative eight-person independent opticians with six Copilot seats, the owner's notes after a week of chats looked like this:
| Person | What they said | Reason |
|---|---|---|
| Optometrist 1 | "Can I put a patient's eye-test results in it? I assumed not." | Unsure what's allowed |
| Optometrist 2 | "Same. I didn't want to be the one who got it wrong." | Unsure what's allowed |
| Dispensing optician | "Tried it for a lens-care leaflet. It gave advice for the wrong type of lens." | Bad first attempt |
| Optometrist 3 | "Referral letters have medical details in. Not sure it's OK." | Unsure what's allowed |
| Receptionist | "It's in Outlook? I didn't know." | Wrong place |
| Practice manager | Uses it daily for rota emails | (Regular user) |
Three of five non-users gave the same reason, so the first fix was a rule sheet about patient information, not more training. A week of reminders would have changed nothing, because nobody was forgetting.
Six reasons AI tools sit unopened, and the fix for each
| Reason | How it sounds | Fix |
|---|---|---|
| No obvious task | "I don't really know what I'd use it for." | Give each role one named weekly task (next section) |
| Unsure what's allowed | "Can I put a client's details in it?" | One page of rules with concrete yes and no examples |
| A bad first attempt | "I tried it once. It was useless." | A tested starter prompt for their task, and a second try with you or a colleague |
| No time to learn | "When things calm down I'll have a look." | 30 minutes of protected, paid time, booked in the diary |
| It's in the wrong place | "I forget it's there." | Use the version inside the tools they already have open, such as Copilot in Outlook or Gemini in Gmail, or pin the app |
| Worry about how it looks | "People will think I'm cutting corners." Or: "Is this the first step to replacing us?" | You use it openly first, and talk honestly about what it means for jobs |
The second and sixth reasons are the ones owners underestimate. Careful staff won't touch a tool if they suspect it might get them in trouble, so a clear page of rules actually increases use. The tutorial on writing an AI usage policy has wording for it. On job worries, don't wave them away; the honest framing in will AI replace my employees? is a good place to start your own thinking.
"Concrete yes and no examples" means sentences a nervous person can match their task against. An excerpt from what a five-person recruitment agency's page might say (illustrative):
YES, go ahead:
- Rewrite a job advert to be clearer and shorter
- Draft an interview-confirmation email from the date and time
- Turn your call notes into a client update (first names only)
- Summarise a client's brief into five bullets
NO, never:
- Paste a whole CV with contact details into a personal account
- Ask it to rank, score or shortlist candidates
- Include health, disability or right-to-work details
ASK FIRST (message the owner):
- Anything going to a client under your name for the first time
- Any new use you're not sure about. Asking is never a problem.The "ask first" list does as much work as the other two. It tells people there's a safe route for grey areas, which is what stops the careful ones freezing. The ranking ban is there for a reason beyond caution: AI used to screen or rank job candidates is listed as high-risk under the EU AI Act if you recruit for roles in the EU, and it's where hidden bias does the most damage.
Give every role one recurring task
"Use AI more" is not an instruction anyone can act on. "Use it to draft your Friday parent updates" is. For each role, pick one task that meets all five of these tests:
- It happens at least weekly, so the habit forms.
- It takes at least ten minutes the current way, so the saving is noticeable.
- The output is mostly text, which current AI tools handle best.
- It's easy to check: the person can tell at a glance whether the draft is right.
- A weak first draft does no harm, because a person reviews it before it goes anywhere.
Here's what that looks like for the tutoring agency in the worked example below:
| Role | Weekly task | What AI does | What the person still does |
|---|---|---|---|
| Tutor | Progress note to parents after each session | Turns five rough bullet points into a clear, friendly note | Writes the bullets, checks accuracy, adds one personal line |
| Tutor | Extra practice questions | Drafts ten questions on the topic at a stated level, with answers | Checks every answer and removes anything off-syllabus |
| Administrator | Rescheduling emails and timetable changes | Drafts replies from a short note of what changed | Confirms dates and times against the diary before sending |
| Owner | Enquiry replies and tutor matching | Summarises the enquiry and suggests two suitable tutors from their profiles | Makes the match and personalises the reply |
One task per role is enough to start. Once people use AI for one thing without thinking about it, they find the second task themselves.
Put starter prompts where people already work
A blank chat box is intimidating. A saved assistant with the instructions already written is not. Most business plans now let you share set-up prompts with colleagues:
- ChatGPT Business: shared projects can be opened to workspace members by email or link. People joining by link get chat access by default, and you can give edit access to whoever maintains it.
- Claude Team: projects can be shared with colleagues as "Can use" or "Can edit", so most people use the instructions while one person keeps them up to date.
- Gemini: Gems (which become skills from November 2026) can be shared much like a Drive file, with viewer or editor access, if your admin allows sharing.
- Microsoft 365 Copilot: prompts can be saved in the Copilot Prompt Gallery and shared to a Microsoft Teams team.
An example of the instructions behind a shared "session notes" assistant:
You help our tutors write short progress notes to parents after
a tutoring session.
The tutor will give you 3-6 rough bullet points. Turn them into a
note of 80-120 words that:
- opens with what the pupil worked on
- names one thing that went well and one thing to practise
- suggests one 10-minute activity for before the next session
- is warm and plain; no jargon, no exclamation marks
Use only what's in the bullets. Don't add scores, dates or
comments the tutor didn't give you. Use the pupil's first name
only. If a bullet is unclear, ask the tutor instead of guessing.What testing catches is easiest to see in one run. A tutor's bullets:
- fractions: adding with different denominators, got it by the end
- still rushes, forgets to simplify
- homework sheet 4 not done
- try the fraction pizza gameAn illustrative draft the assistant might return:
Today [Name] worked on adding fractions with different denominators, and by the end of the session she was getting them right on her own, which was great to see. One thing to practise is slowing down and simplifying her final answer. Homework sheet 4 wasn't completed, so please make sure it's done before next Thursday's session. A fun 10-minute activity at home is the fraction pizza game.
Mostly good, and two things to fix. "Next Thursday" was invented, despite the instruction not to add dates, and the tutor sees this pupil on Mondays. And the homework line reads like a telling-off to a parent, which the tutor softened to "Sheet 4 is still to do, and it's good practice for simplifying." That's why the rule is that every note is checked, and why the test run matters: it showed the team exactly which line to read carefully.
Test each starter prompt on five real examples before sharing it. Once you have more than a handful, organise them properly; building a shared prompt library covers naming, ownership and keeping them current.
Make it visible, and a little bit social
People copy what they see colleagues doing. Three small habits make AI use visible without turning it into a performance:
- A five-minute slot at the weekly meeting where one person shows one prompt and what it produced, including when it went wrong. You go first, and you show a failure as well as a success.
- A shared channel or thread for wins and warnings: "This prompt saved me 20 minutes" and "Watch out, it invented a date in my email" are equally useful.
- Pairing: sit a reluctant colleague with a confident one for 20 minutes on the reluctant person's own task.
Useful posts in that channel are short and specific. A typical week in an illustrative seven-person architecture practice:
WIN: Planning-conditions summary prompt on the 14-page decision
notice for the barn job. 25 min down to 8. It's in the library
as "Planning conditions: summary".
WARNING: Asked it for a fee-stage breakdown and it used
percentages from nowhere, not our fee schedule. Paste the
schedule in every time.
TRIED, DIDN'T WORK: Drafting the design-and-access statement.
Too generic to save time. Going back to our own template.The third kind of post is the one to encourage most. When people see a colleague report a failure without any fuss, trying something and dropping it stops feeling risky.
If one person naturally becomes the one others ask, recognise it and give them time for it. The tutorial on choosing and supporting an AI champion explains how to make that role work without burning them out.
Worked example: a tutoring agency with 12 seats and 3 regular users
Say, for illustration, a tutoring agency runs with its owner, a pair of administrators and nine part-time tutors. The owner bought 12 ChatGPT Business Standard seats at $25 per user per month on monthly billing: $300 a month at list price in USD. Six weeks later, three people used it regularly: the owner, one administrator and one tutor.
What the conversations found. Tutors didn't know whether they could mention pupils at all, so they didn't risk it. Most couldn't see what it was for, since nobody had suggested a task. Two had tried it once for lesson plans and found the result generic. And the tool was a website they never had open during a session.
What the owner changed over two weeks:
- A one-page rule sheet: first names only; no information about a pupil's health, learning needs or family circumstances; every note checked by the tutor before sending.
- A shared "session notes" project with the instructions above, tested on ten real notes first.
- A paid 30-minute online session where each tutor wrote one real progress note with it.
- The ChatGPT app installed on tutors' phones, since most wrote notes straight after sessions.
- At the next team call, the owner showed her own enquiry-reply prompt, including one reply she'd had to rewrite.
Six weeks after the changes: 9 of the 12 people used it at least weekly: the owner, both administrators and six tutors. Tutors timed a few notes: about 6 minutes each, down from about 15. Six tutors averaging ten sessions a week, saving roughly 9 minutes a session, comes to about 9 hours a week across the team.
The three tutors still not using it each taught only one or two sessions a week and preferred writing their few notes by hand. That's a reasonable choice, so the owner removed their seats: 9 seats at $25 is $225 a month.
Mandates, leaderboards and other moves that backfire
- Usage quotas. "Everyone must use AI five times a week" produces five pointless prompts a week.
- Leaderboards. Ranking people by usage shames the careful and rewards the careless.
- Reading everyone's prompts. Checking outputs that reach customers is fair. Monitoring what people type as they learn makes them stop experimenting, or move to personal accounts you can't see.
- Forcing it onto tasks where it's worse. If someone's work is genuinely better without AI, respect that and move their seat.
- Buying another tool. If the first one isn't used, the second one won't be either. The blockers are rarely about the tool.
Heavy restrictions have a quieter cost too: people who can't use the approved tool the way they need often switch to free personal accounts, which is worse for your data. The tutorial on shadow AI in small businesses explains how to spot that.
Count tasks done, not logins
Once people start using the tool, measure what matters: how many of the target tasks now go through the new way each week, how long they take, and whether a spot check of five outputs finds problems. A log-in count tells you someone opened the tool; a task count tells you the work changed.
Review seat by seat after a couple of months. Anyone who still doesn't use it after the blockers are removed may simply not need a paid seat. Working out who needs a paid AI licence helps you right-size the plan, and it's usually cheaper to give a few people the right seat than everyone the wrong one.
Other questions about staff AI adoption
Should I make AI use part of staff performance reviews?
Not as a usage target. Reviewing results is fair: if someone's reports now take half the time and read better, recognise it. Scoring people on how often they open an AI tool rewards pointless use and makes the careful people look worst. If you mention AI in reviews at all, frame it as a skill you'll support, with time and training, not a quota.
What if a senior colleague refuses to use it?
Find out whether the refusal is about the tool, the task or something else, such as worry about quality or their standing. Ask them to judge the output rather than produce it: reviewing AI drafts for accuracy uses their experience. If their own work genuinely goes better without AI, that's a legitimate answer, and their seat may be better used by someone else.
Is it better to give everyone one tool or let people choose their own?
One approved tool on a business plan is easier to support, share prompts in and keep data safe with. Letting everyone choose usually means personal free accounts and no shared learning. Allow a second tool only for a specific need the main one can't meet, approved by whoever owns AI in the business.
Further reads
- How to Train Staff to Use AI in a Small Business — A session-by-session training plan once the blockers are cleared.
- AI Change Management for Small Teams: A Practical Plan — The wider eight-week plan this fits into.
- How to Run a 60-Minute AI Workshop for Your Team — A ready-made format for a team session.
- How to Handle Staff Who Over-Rely on AI — The opposite problem, once use takes off.
- How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — Decide whether seats nobody uses should go.
- Standard vs Premium AI Seats: Who Needs the Bigger Plan? — Match plan size to how each person uses AI.
- How to Run Your First AI Pilot Project in a Small Business — Six stages for a first AI pilot, from a one-page charter to the keep, fix or stop meeting, followed through an optician's email pilot with real-looking numbers.
- How to Scale AI From One Workflow to the Whole Business — How to copy your first working AI workflow across the business: a readiness gate, a playbook template, adjacency rules and a pet shop's nine months.
- How to Survey Your Staff Before an AI Rollout (With Questions) — Fifteen ready-to-use questions, an invitation email and anonymity settings for surveying a small team before you choose any AI tools.
- How Much Time and Money Does AI Staff Training Take? — Training hours by role, free and paid course options with current prices, a costed plan for an eleven-person decorating firm, and how to check it paid off.
- How to Use AI on Your Own Admin First, Then Roll It Out — A six-week owner-first plan: pick three of your own admin jobs, log what AI saves and gets wrong, write a playbook, then hand one task to one person.
- Why Your AI Pilot Stalled, and How to Get It Live — A one-hour diagnosis for a stuck AI pilot, the fix for each of five causes, a 30-day restart plan, and when stopping is the better call.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Learn (Microsoft 365 Copilot usage reports, Copilot Prompt Gallery); Google Workspace Admin Help (Gemini usage reports, Gem sharing); OpenAI help documentation (Projects in ChatGPT); Claude help documentation (project visibility and sharing); ChatGPT Business list pricing (checked September 2026).