How to Train Employees on AI (Safely and So It Sticks)

Almost every small business has run some version of AI training already — a webinar link forwarded in a group chat, a slide deck nobody opened past page three, a "just play around with it" comment in a team meeting. None of that is really training. It's exposure. Training means people leave the room knowing exactly what they can and can't do, having actually tried the tool on real work, and knowing who to ask when they hit something the session didn't cover.

Why most AI training doesn't stick

Thryv's 2026 AI and Small Business Adoption Survey found that 70% of small business owners say they need more, or significantly more, training to use AI productively — despite the fact that most of them are already comfortable with the technology in principle. That's the tell: comfort with the idea of AI isn't the same as being trained to use it well. A one-hour, one-time session creates comfort. It rarely creates a habit.

The other reason training fades is that it's usually generic. A single session that tries to cover a chatbot, an AI note-taker, and an AI feature buried in the accounting software all at once ends up too shallow to help anyone with their actual job. The front desk person and the bookkeeper walked away from the same slide deck, and neither got what they needed.

What to settle before you train anyone

Training before you have rules just teaches people to move faster without guardrails. Get three things in place first, even in rough form:

• A written data rule. One sentence is enough to start: never put customer names, financial details, health information, or passwords into a public AI tool.
• An approved tools list. Name the two or three tools people are actually allowed to use for work. Without this, "ask before trying something new" has nothing to point to.
• A named point of contact. Pick one person to own AI questions and incident reports. Training that ends with "figure it out" doesn't hold up past week one.

If none of this exists yet, an AI acceptable use policy is the fastest way to get all three in one short document, in an afternoon rather than a quarter. It doesn't need to be long or reviewed by three departments before it's usable — a single page that names the data rule, the approved tools, and the point of contact is enough to build a training session on. You can always add sections later as new tools or edge cases come up; the version you have on day one just needs to be true and specific.

A four-part training plan that sticks

The plan below is built to run in a single afternoon for a small team, with reinforcement built in rather than left to chance. Each part is short on purpose — the goal is retention, not thoroughness.

Notice what's missing: a long lecture on how AI models work. Nobody needs that to use the tools safely and well, and it's the part that puts people to sleep and pushes out the parts that actually matter.

A copy-ready 60-minute session outline

Use this as a script for the first training session. Fill in your own tool names and policy owner before you run it.

If you want the role-specific prompts already written, our prompt library guide has 25 starter prompts organized by function — front desk, sales, ops, bookkeeping, and marketing.

Making it stick after week one

The session itself is maybe a third of the work. What happens in the following month decides whether it holds. Three things make the difference:

• A two-week check-in. A short, informal conversation — "what have you tried, what got confusing" — surfaces the questions people were too shy to ask in the room.
• A visible go-to person. If the named contact from training isn't easy to reach two weeks later, people quietly stop asking and start guessing.
• A living tools list. Keep the approved tools list somewhere people already look — the handbook, the onboarding doc, a pinned message — not buried in a folder nobody opens twice.

How to tell if the training actually worked

Most small businesses skip this step entirely, which makes it impossible to tell a good training session from a forgettable one. You don't need a formal test — three simple checks, run a month after the session, tell you almost everything:

• Ask, don't assume. Pick three employees at random and ask them to describe the data rule in their own words. If they can't, the rule didn't land, no matter how clearly it was presented.
• Watch for the tool actually being used. If nobody has opened the approved tool since the session, the training created awareness without changing behavior — usually a sign the use case shown wasn't relevant to their actual job.
• Count the questions, not the silence. A trickle of small questions to the named point of contact in the weeks after training is a good sign — it means people feel safe checking rather than guessing. Total silence usually means people are unsure and staying quiet, not that everything is clear.

If all three checks come back weak, the fix usually isn't more content — it's more relevance. Go back to the role-specific use cases and make sure the prompts and examples map to something people actually do every week, not a generic example that looked good in a slide deck.

Common training mistakes

• Training on tools before rules. This teaches speed without judgment, and it's the fastest way to end up with a data incident three weeks after a "successful" training session.
• One generic session for every role. A single deck for the whole company means nobody gets what's actually useful for their job.
• No hands-on practice. Watching someone else use a tool doesn't build the habit. People need to try it themselves, on real work, in the room.
• No follow-up. A single session with nothing after it fades within a month — reinforcement is what makes training different from a webinar.
• Treating questions as a sign of failure. If asking "wait, can I use this for a customer email?" two weeks later feels embarrassing, people stop asking and start guessing wrong.

This guide is general information, not legal advice. Have a qualified attorney review any policy before you adopt it.