Roll out AI training in three waves: publish a one-page AI policy first, then give everyone the same short hands-on session on safe use and prompting, and only then build role-based tracks for each department. Start the first wave with Roll Out a Team AI Policy in One Afternoon, which takes you from your rules to a policy, an announcement email and a five-question staff quiz.
The order matters because training without rules teaches people to do the wrong thing faster. Most staff are already using AI in some form. Your job in L&D is not to introduce it. It is to make the use safe, consistent and good enough that managers trust the output.
Why does the policy come before the training?
A course that says "use AI to draft client emails" is useless if nobody knows which tool is approved or whether client names can go in. The policy answers four questions that every later module depends on:
- Which tools are approved, and on which account (work, not personal).
- What never goes in: client financials with names, passwords, health information and other personal data.
- Who checks the work: a person reviews every output before it reaches a client, employee or decision.
- Who to ask, and what happens when someone pastes the wrong thing.
Write it with IT, legal or compliance and one business leader, so it is a company rule and not an L&D suggestion. If your company has a no-policy gap today, close that before you schedule a single session.
What should the company-wide session cover?
Keep the shared core short, the same for everyone and hands-on. Two skills do most of the work:
- Safe use. Check the tool's data settings, take names and account numbers out before pasting, and check every fact, number and promise before anything goes out. Use AI at Work Without Leaking Client Data covers all three in about 7 minutes.
- Prompting. Role, Context, Task and Format. People who learn this stop getting generic answers. Prompt Writing 101 takes about 8 minutes.
Before either, a two-minute warm-up like How Stepthrough Works gets nervous learners clicking in a practice copy, which lowers the fear of breaking something.
Practice in the tool your company actually licenses. If you run Microsoft 365 Copilot, train on Copilot. If you have a ChatGPT or Claude business plan, train there. Prompting carries over between tools, but the settings and screens do not.
How do I build role-based tracks?
After the core, train by job, not by tool. A useful track is three to five real tasks a role does every week:
- Finance and accounting: analyzing an export, reconciling, explaining variances.
- Sales and customer teams: research, first drafts of replies, meeting recaps.
- HR: policy comparisons, job postings, handbook answers.
- Operations and admin: meeting notes to action items, vendor quotes, contract dates.
Ask each department head to name the tasks, then pick lessons or build exercises around those tasks. The free plan for small teams and the free hands-on course for office workers list ready-made lessons you can map to roles.
What does this look like in a real lesson?
In Roll Out a Team AI Policy in One Afternoon you are setting rules for Northfield Consulting Group, a made-up 18-person firm where someone pasted client numbers into a free tool last month. You give Claude your rules: which tools, what never goes in, who checks, who to ask, and a line telling it not to invent laws. Claude writes a one-page policy. You then check three sections yourself: the never-paste list, the rule that a person checks every output, and the no-blame rule for reporting a wrong paste the same day. Last, Claude writes a short announcement from the managing partner and a five-question true or false quiz, where question 3 catches the common mistake: an approved tool with no human check is still not allowed.
That quiz is a useful L&D pattern. It tests the rules people get wrong, not the ones they already know.
How do I measure whether training worked?
Skip completion rates as your main measure. Look at behavior instead:
- Are people using the approved tool, or still on personal accounts? IT can usually tell you.
- Do managers report fewer AI drafts that need rewriting?
- Do people report wrong pastes, as the policy asks? A few reports is a good sign, not a bad one.
- Can each department name one task that is now faster, and what the person still checks?
What mistakes should L&D avoid?
- Lecturing instead of practicing. People learn AI by doing a real task, not by watching slides about it.
- One generic course for everyone. The core is shared. Everything after it should be by role.
- Training on a tool you have not licensed. It pushes staff toward personal accounts.
- Treating the policy as legal's job alone. If L&D does not teach it, nobody reads it.
- Never updating. Tools change their screens and settings often. Review the material every quarter.
Key takeaways
- Policy first, then a short shared core on safe use and prompting, then role tracks.
- Train in the tool your company pays for, with work accounts.
- Build each role track from three to five real weekly tasks.
- Measure behavior: approved-tool use, fewer rewrites, honest reporting.
Frequently asked questions
- How long should company-wide AI training take?
The shared core can fit in under an hour of hands-on practice: a short policy walkthrough, safe use and prompting. Role tracks then run as short sessions over the following weeks.
- Should AI training be mandatory?
The policy and the safe-use session should be, because they protect clients and the company. Role tracks work better when managers pick the tasks and staff see the time they get back.
- Who should own AI training, L&D or IT?
Share it. IT owns the tools, accounts and settings, and L&D owns the learning design and the rollout. See how IT teams roll out AI safely for the IT side.