How to train your team to use AI

You can train a team to use AI in 30 days, provided you define “use” narrowly. By day 30 each person should be doing two or three real tasks with an approved tool, checking the output properly and knowing what never goes in. That is a working habit, and it is the right target for a first month. Expertise comes later and cannot be rushed into a calendar.

This article gives you the plan: what to settle before day one, a week-by-week schedule you can run with a team of any size, how to deal with the colleague who does not want to take part, and how to tell at the end whether it worked. It is written for owners and managers who need their people using AI well, not just aware that it exists.

Key takeaways

• Thirty days builds a habit, not mastery. Aim for each person using an approved tool on two or three real tasks, with checking built in.
• Settle five things before day one: a named owner, a one-page rule set, an approved tool, three tasks per role and a baseline measure.
• Teach in this order: rules, clear requests, checking, routine. Features come last and matter least.
• Train on each person’s own work. Shared rules, role-specific tasks.
• Resistance is information. Fear of looking foolish, of losing a job or of a past bad result each has a different fix.
• Measure use, time, rework and confidence before and after, or you will not know whether the month changed anything.

Can you train a team to use AI in 30 days?

Yes, if you ask for the right thing. Thirty days is long enough to change what people do on an ordinary Tuesday. It is not long enough to turn a cautious team into confident power users, and a plan that promises that sets everyone up to feel they failed.

The distinction that matters is between knowing about AI and using it. Many teams have had a talk, a demo or a recorded webinar, and most of those are forgotten within a fortnight because nobody went back to their desk and used what they heard. A 30-day plan flips the order. People use the tool on their own work from the first day, and the sessions exist to make that use safer and better, not to replace it.

So define “done” before you start. A sensible definition for day 30 has four parts:

1. Everyone can say which tools are approved and which information never goes into them.
2. Everyone has done at least three real tasks with the tool and kept the result.
3. Everyone checks output before it leaves their hands, and knows how.
4. The team has a shared set of requests that worked, so the next person does not start from nothing.

If you can tick all four, you have trained the team. If you have run six sessions and cannot tick the second one, you have run six sessions. This is the same lesson that sits behind what makes AI training actually stick: the training that lasts is attached to the work people do, not to a slide deck.

What should you do before the 30 days start?

Most of the difference between a plan that works and one that fades is decided in the week before day one. Five jobs, none of them long.

Name an owner. One person who runs the plan, answers questions and approves new tools. Without one, questions go unanswered, people fall back to guessing and the plan quietly stops in week two. In a small business this is often the owner or an operations lead.

Write the rules on one page. People cannot use a tool safely if nobody has told them what is allowed. A one-page AI use policy covering approved tools, what data stays out, who checks output and who to tell when something goes wrong is enough. It can be tightened after the first month, but it has to exist on day one.

Approve a tool and give everyone access. Pick one approved assistant rather than letting everyone choose. A single tool means one set of examples, one set of rules and one place to ask questions. Make sure every person has a working account before the first session, because a day lost to logins is a day of momentum gone.

Choose three tasks per role. Not three tasks for the team, three for each kind of job. Pick work people already do, that repeats, and where a mistake is cheap to catch. Our guide to which AI tasks a small business should start with first sets out a test for this: low risk, easy to check, done often enough to notice the gain. A recruiter might choose drafting job adverts, summarising interview notes and writing candidate updates. A finance assistant might choose summarising long email chains, drafting payment reminders and tidying meeting notes.

Take a baseline. Time one of the chosen tasks the way it is done today. Ask each person to rate their confidence with AI from one to five. Write both numbers down. Without them, day 30 is an opinion.

Then brief the team before the first session, in person, for fifteen minutes. Say what the 30 days are for, what is expected each week and that nobody is being tested or replaced. We cover how to brief the team in a separate article, and it is worth reading before you send the calendar invitation.

What does a 30-day AI Training plan look like?

Here is the shape at a glance, followed by each week in detail.

Keep the sessions short. The learning happens in the daily use between them, so protect ten to fifteen minutes a day for it rather than stretching the sessions.

Week 1: Rules and first tasks

The first session covers three things and nothing else.

First, the rules. Walk through the one-page policy and take the data question slowly, because it is where most harm comes from. A three-tier rule works well: never paste this, paste only with care, safe to paste. Our guide to what is safe to paste goes further, and it is worth using its examples. Use real examples from your own business, such as a customer list or a draft contract, rather than generic ones.

Second, what the tool is and is not. Tell people plainly that it can write fluent text that is wrong, which is called a hallucination, and that a confident tone is not evidence. This is the single most useful sentence in the whole month, because it explains why the checking in week 3 matters.

Third, a live first task. Everyone does their first chosen task during the session, with you in the room to answer questions. People remember what they did far longer than what they heard.

For the rest of the week, each person does one of their three tasks each day for ten minutes. At the end of the week, ask each person to share one result, good or bad, with the group.

Week 2: Prompts for Real work

By week 2 people have noticed that the answers vary a lot. Some are good, some are useless, and they cannot yet tell why. This week is about the request itself.

Teach one structure and teach it well: say what you want, who it is for, what the tool needs to know, what to avoid and what shape the answer should take. A useful way to put it to a team is to brief an AI tool like a new hire. A new hire given a vague instruction produces vague work, and the same is true here. Give the tool the background you would give a person on their first day, and the results change quickly.

Start the team’s prompt library in this session. Every time someone writes a request that works well, they paste it into a shared document with a line on what it is for. By the end of the week each person should have added at least two. This single habit does more for adoption than any lecture, because it means the second person in the team starts with a working request instead of a blank box.

Pair people up for the daily practice. A pair swaps requests, runs each other’s, and notes what changes. It turns a solitary activity into a conversation, which is where most of the learning happens.

Week 3: Checking output and finding the limits

This is the week most plans skip, and it is the week that protects the business. By now people are fast. The risk is that they are fast and trusting.

Run the session around checking. Give each person a short checklist: are the facts correct, are the names and figures right, does it say anything the business would not say, does it sound like us. Where the output goes outside the business or affects a person, add a second reader. Our article on how to check AI output before it goes to a customer has a checklist you can adapt, and it fits on a card.

Then do something deliberate: ask everyone to find a task where the tool gets it wrong. Wrong figures, invented sources, a confident answer to a question it cannot know. Have each person bring one example to the session and write it up in two lines. People who have seen the tool fail once check far more carefully than people who have only been told it can.

This is also the right moment to talk about where a person must stay in charge. For anything that affects hiring, pay, a complaint or a customer’s money, keep a human in the loop who makes the decision and records why. The tool may inform it. It must not make it.

Week 4: Making it routine

The last week turns practice into habit. Agree, as a team, which two or three tasks are now done with the tool as the standard way, and write them down: the task, the approved request from the prompt library, who checks and where the result goes. That is a small working procedure, and it is the thing that survives after the month ends.

Take the same measures as before. Time the same task again, including the checking. Ask the confidence question again. Count how many people used the tool in the past week. Put the numbers next to the baseline.

Finally, set the next review date and name who owns the shared library. Tools drift, requests go stale and new starters need the same on-ramp. A plan with no owner after day 30 is a plan that ends on day 31.

How do you get a reluctant team to use AI?

Some resistance is certain. The useful move is to treat it as information about what the person is worried about, because the fix depends on the worry.

“Will this replace me?” Say plainly what the training is for, and what it is not for. If your honest answer is that nobody’s job is changing, say so in writing. If you do not know, say that too, and say what you will do to find out. Silence leaves people to imagine the worst.

“I will look foolish.” This is the most common fear and the least spoken. Start with private practice, not performance. Let people try things alone before they share, and make sure the first group session shows a leader’s mistake before anyone else’s. If the manager goes first and gets something wrong, the room relaxes.

“I tried it and it was rubbish.” Often true. A vague request to a general tool gives a vague answer. Pair this person with a colleague whose requests work, and use their own task rather than a generic one. One good result on their own work is worth more than ten demonstrations.

“It is not allowed.” Sometimes they are right, and sometimes it is an old rule nobody has revisited. Check, and fix the rule if it is out of date.

Two things make resistance worse. The first is a ban. A blanket refusal to let people try anything is exactly why banning AI tools at work does not work: use moves to personal accounts and phones, where you cannot see it and cannot protect it. That hidden use is called shadow AI, and a training plan with an approved tool is the quickest way to bring it into the open. The second is pressure without support. Telling people to use AI and then not giving them time, examples or a place to ask produces compliance on the surface and avoidance underneath.

Give people a champion rather than a mandate. Pick one or two colleagues who are curious, not necessarily senior, and give them an hour a week to answer questions. Peers are believed in a way that managers sometimes are not.

How do you know the Training worked?

Four measures, taken before and after. None needs special software.

Use. How many people used the approved tool in the past week, and for how many tasks? A simple shared tally is enough. This is the measure of AI adoption in its plainest form, and it is the first thing to move.

Time. Pick one task, time it the old way and the new way, and include the checking in the new figure. This matters because a task that takes five minutes with the tool and ten minutes of correction has not saved anything. Our article on whether your AI tool is actually saving time gives a one-task test you can run.

Rework. How often did output need to be corrected or redone? If use goes up and rework goes up with it, the training taught speed without checking, and week 3 needs repeating.

Confidence. Ask the one-to-five question again. A rise tells you people feel capable. A rise in confidence without a fall in rework is a warning, not a success, because confident and careless look the same on a survey.

Read the four together. Use up, time down, rework steady or falling and confidence up is a good month. Any other pattern tells you where to look next.

Be honest about the limits of a small sample. With ten people, one enthusiast can move an average a long way. Look at the spread as well as the mean, and at what individuals said, not only the totals.

When is a 30-day plan not enough?

For many small teams, the plan above is the right whole programme for the first quarter. It stops being enough in three situations.

When AI touches decisions about people, as in recruitment, performance or pay. Those roles need training on bias, fairness and legal duties that a general month does not cover.

When the work is regulated or carries real risk, such as finance, legal or health. The checking and the data rules need more depth and, often, an adviser who knows the sector.

When different teams need genuinely different skills. A single shared plan has a ceiling, and past it the better step is training built around each team’s own workflows.

There is also a legal angle. The EU AI Act expects organisations that use AI to make sure their staff have a sufficient level of AI literacy, and a written plan with a record of who was trained on what is a practical way to show you took that seriously. The Act does not set a course or a number of hours, and the wording in this area has been adjusted recently, so check the current text before you quote it.

If you reach the end of the month and want to go further, there are two routes. Set courses give a group a structured foundation. For sessions built from your own tasks, data rules and tools, choose bespoke training. Which one fits depends on how different your teams’ work is.

What follows

Name an owner, put the rules on one page, choose three tasks per role and take your baseline this week. Then run the four weeks in order, in short sessions, with daily practice between them. If you want a structured foundation for the whole team, our set courses cover the same ground with a trainer in the room.

Sources

• Regulation (EU) 2024/1689 (the EU AI Act), Official Journal of the European Union: the legal text of the EU AI Act, including the provision on AI literacy for organisations that deploy AI. Check the consolidated version for the current wording.