Write the AI rule for your training before you use AI on your training

My position is that AI belongs in the drafting stage of training and nowhere near the final say. It is fast, tireless and good at first versions. It does not know your people, your policies or your risk, and it does not stand in the room when a participant pushes back. A team that uses it well has decided in advance what it may do. That decision is the rule.

I have written about what to hand to a tool in what to draft with AI in training, and what not to, and about the limits of the tool in what AI cannot do in a live training session. This piece is about the layer above both: the written rule that lets a team use the tool without drifting.

Because habits form in the first weeks, and few teams stop to write the rule once the habit works. A person who finds a shortcut shares it, others copy it and the team ends up with a practice the team did not choose. If the practice includes pasting employee details into a public tool, you find out the hard way.

A rule also makes the work easier. Staff know what they may do, so they do not wait for permission, and reviewers know what to check. The rule is a small tool for speed with safety, not a barrier.

I keep it to five sections, each a few lines.

• Allowed uses. The drafting tasks you approve: outlines, discussion questions, scenarios, summaries and alternative phrasings.
• Kept out of prompts. Names, complaint details, medical information, investigation notes and performance records. Use placeholders instead.
• Review. A named person reads each draft against the learning objective, checks facts against a primary source and signs off before use.
• Disclosure. A one-line statement to participants that AI helped prepare the material. This site carries one: “Website content is AI assisted.”
• Human decisions. The judgments that stay with a person: how to answer a sensitive question, how to handle a conflict, what to change in the moment and what to say to someone in distress.

Add an owner and a review date. A rule without an owner is a document, and a rule without a review date is out of date within a year.

Read it as if a new colleague wrote it, with a fresh mind and a source in hand. Check each fact, each citation and each rule it states. AI can produce confident text that is wrong, so a draft that reads well is not evidence that it is right.

Then check fit. Does it match the objective? Does it use words your people use? Does it ask for an activity that works on a screen? Does it include any claim you cannot support? Cut anything you cannot defend. AI upskilling: why the session matters more than the course explains why the facilitator, not the draft, decides what the session does.

Anything that depends on judgment about these particular people at this moment. A participant asks a question that touches a real complaint. Two people disagree and the tone shifts. Someone shares something heavy. A tool cannot read the room, and a script written in advance cannot answer for you.

This is where the rule earns its keep. It says plainly that the facilitator owns those moments. How to train your team on AI shows how to teach the rule itself in a session, with practice on real tasks.

The same principle applies to tools that join the meeting. Three bots are in your meeting and nobody asked looks at what recording and summary tools change, and California AB 1883 and your engagement dashboard covers tools that score participants.

Here is a hypothetical sample for a small team. It is an example to edit, not a policy to copy.

That is under a hundred words, and it covers the five sections. A longer rule is not a better rule. The test is whether a new team member could read it in a minute and know what to do on Monday.

Notice how it handles a hard case. A participant asks a question that touches a real complaint, and the facilitator answers from the plan and the organization’s process, not from a generated script. The rule says so in advance, which means the facilitator does not have to decide it in the moment.

Review it on a calendar, not when something breaks. Put a date on the page, assign a person and keep a short list of changes. Tools change faster than rules, so a yearly review is too slow for the use cases and too frequent for the principles.

I suggest a light review each quarter. Ask what staff are actually doing with the tools, what went wrong, what saved time and what needs to change. Update the allowed uses and keep the principles stable. If your organization has employees in more than one jurisdiction, add a line to check local rules before you adopt a new tool.

A plain rule also helps when you buy or hire. A vendor or facilitator who works with your content can be asked to follow it, which is one more reason to write it down.

Write the first draft this week, in an hour, and test it on one real task. Choose a session you are about to build. Draft part of it with AI under the rule, review it as described and note what the rule missed.

The rule will be wrong in places. That is fine, because a rule that exists and gets edited beats a perfect rule that does not. My advice: write the page before you open the tool, and give it to the person who will review the first draft.

Want a second read on your AI rule?

Bring a draft to a free consult. In twenty minutes we test it against one real training task.