Train employees on AI in five steps: write a short policy that says which tools are allowed and what must stay out of them; teach a few core habits in a sensible order; replace one-off workshops with short, regular practice on situations from their own work; measure what people can now do rather than who attended; and follow up where teams get stuck. The goal isn’t to make everyone an expert. It is that people use the AI they already have well, and safely.

Step 1: Set the rules first

Before anyone is trained, they need to know what they are allowed to do. A useful AI policy fits on one page and answers four questions:

Without this, careful people avoid AI and careless people paste in whatever is in front of them. Both are worse than a clear rule.

If your company operates in the European Union, there is also a legal reason. The European Commission’s AI literacy questions and answers explain that Article 4 of the AI Act requires providers and deployers of AI systems to take measures to support the AI literacy of their staff. The same page says no certificate is needed and that organizations can keep an internal record of their trainings. Ask your own counsel how it applies to you.

Step 2: Teach in an order that builds

Most AI training starts with prompts. That is the middle, not the start. A better order, where each step rests on the one before:

  1. The words. What AI is, and the handful of terms that explain its surprises: a hallucination (AI stating something false as if it were true) and the context window (everything the AI can see while it answers). See AI terms explained.
  2. Judgment. Which jobs to hand to AI, which to check first, and which need a different tool. This is the habit that prevents the most harm.
  3. Briefing. Saying what you want back, the limits and what counts as done, so the first answer is close. See how to write a good prompt.
  4. Context. What AI needs to see, what to keep out, and what it forgets between chats. This is where your policy turns into daily habit.
  5. Checking. Reading AI’s work against the facts and catching the confident mistake before it reaches a client. See when to trust what AI tells you.
  6. Reuse. Turning a job someone repeats into saved instructions their assistant can follow each time. See what a Claude skill is.

Step 3: Practice, not just workshops

A workshop is good for launching a policy and answering questions. It is a poor way to build a habit. People hear it once, go back to a full inbox, and do what they did before.

Habits come from practice: short, frequent, on situations that look like the person’s own week. A project manager should practice deciding which parts of a schedule to hand to AI. An accountant should practice it on the month-end close. The skill is the same, but the example has to be theirs, or it feels like someone else’s job.

Good practice has three features:

Avoid streaks, leaderboards and countdowns. They reward showing up, not getting better, and they put off the people you most need to reach.

Step 4: Measure what people can do

Attendance and completion rates are easy to report and tell you little. What you want to know is whether people now make better decisions with AI. To see that:

Our guide to what AI fluency is and how to measure it goes into this in more depth.

Step 5: Follow up where teams get stuck

The useful output of measurement is a short list: where is each team stuck? If several people on a team miss the same kind of decision, bring it to the next team meeting. Spend a few minutes on what the skill is, why it matters for your work, and one question to discuss. That conversation does more than another round of training.

Recognize progress, too. A short note from a manager when someone finishes a stage of the training costs little and tells people it matters.

A simple rollout

  1. Publish the one-page policy.
  2. Pick one team with a clear use for AI and a manager who wants it to work.
  3. Give them regular practice set in their own work, and check in each week.
  4. After a month, look at what they can now do and where they are stuck. Fix the training, then widen it to the next team.

If you use a training product, ask the vendor plainly what it keeps, what your admins will see, and whether it connects to your systems. Our own answers are on the security and privacy page, and seat prices are on pricing.