Most training fails the same way. Everyone sits through a demo, says "cool," and goes back to doing it the old way on Monday.
AI training works when it is short, practical and followed up. This post gives you a one-day plan you can run yourself, the rules to set on day one, and the follow-up that makes it stick. It is part of our guide to AI implementation for small business, where training is step three.
Your team is probably already using AI
Start with what is already happening. The U.S. Chamber of Commerce Foundation and Ipsos surveyed 1,070 employed adults at small businesses in May 2026. They reported that half of all workers at small businesses already use AI at work. When asked what drives adoption, 19% of those workers said it was mostly employees exploring on their own, and 11% said it was mostly organizational guidance.
The owner side looks different. In a separate Ipsos survey of 750 small business owners, 29% said adoption is mostly driven by ownership or leadership, and 19% said it is mostly employee-driven. Different questions, different groups, but the picture is that staff are often ahead of the guidance they get.
That is the risk and the opportunity. People are already trying things, and nobody set the rules. Training is how you bring that into the open.
The ask is real too. Goldman Sachs reported that 73% of the small business owners it surveyed would benefit from more training and implementation resources (alumni of its small business program, not a random sample). Thryv's 2026 survey, as reported by CPA Practice Advisor, put it at about 70% of owners. Thryv is a vendor, so weigh that one lightly.
Before the day: three decisions
Do these first. The day goes badly without them.
- Pick the jobs. Train on two or three real tasks your team does every week. Not "AI in general." Use our time-sink worksheet if you have not picked yet.
- Pick the tool. One general AI assistant, plus whatever is already connected to your jobs. Not five. Everyone should use the same one so people can help each other.
- Write the data rules. What can and cannot go into AI tools. Details below.
The one-day plan
This assumes a team of 3 to 12 people and about six hours with breaks. Adjust for your team.
Hour 1: What it is, in plain words
Explain only what people need.
- AI tools like ChatGPT, Claude and Gemini answer when you type to them. What you type is called a prompt.
- They are good at drafting, summarizing, sorting and answering. They are not good at knowing your business unless you tell them.
- They sometimes sound sure and are wrong. Always check.
Keep it under 30 minutes. Then show a real example from your own business, live.
Hours 2 to 3: Practice on real work
This is the core. Each person brings a real task from the past week. Pair them up. They run it through the AI, compare it to what they would have written, and fix it.
Teach a simple prompt pattern:
- Who it is helping: "You are helping a front desk person at a dental office."
- What you want: "Draft a reply to this customer."
- What to include: paste the message, plus your facts (hours, prices, policy).
- What good looks like: "Friendly, under 80 words, no promises about prices."
Do not hand out a long list of magic prompts. Teach the pattern, then let people build their own.
Hour 4: What to check, and what to keep human
Run through the common failures: invented facts, wrong numbers, a tone that does not sound like you, and answers that ignore your policies. The Stanford and UC Berkeley researchers who studied how model behavior changed between versions in 2023 recommended continuous monitoring of hosted models. The lesson for a small team is simple: check the output, every time, and tell someone if it starts acting strange.
Agree on what stays human: pricing decisions, hard customer conversations, anything legal or medical, hiring and discipline.
Hour 5: Data rules
This is the part owners skip and regret. Put it in writing, one page.
- Never paste: passwords, card numbers, Social Security numbers, health information, anything covered by a customer contract.
- Be careful with: customer names and contact details, financials, staff records, anything a competitor would love to see.
- Use only the approved tool, logged in with the company account.
- Tell someone if you pasted something you should not have.
Our free AI-use policy generator drafts this page from a few questions. The Pax8 Q3 2026 pulse report, a vendor survey of 1,150 leaders across four countries, reported that 35% of AI-engaged firms discovered shadow AI (staff using AI tools without the owner knowing) in the prior six months. A policy in writing is how you keep that visible.
Hour 6: Write the playbook
Before everyone leaves, write down what you learned: the three jobs, the prompt for each, what to check, and who to ask. One to three pages. Put it where the team can find it.
This is the most important output of the day. People forget a workshop. They reread a playbook.
After the day: the part that makes it stick
A day of training on its own changes little. Plan for what comes next.
Week 1. Everyone uses the AI for their chosen job at least three times. Ask for one example each in a shared thread, the good and the bad.
Day 30. A one-hour check-in. What is working? What did it get wrong? Update the playbook. Drop what nobody uses. We include a 30-day check-in in our own Train step.
Day 60. Look at the numbers. Hours saved on each job, and hours spent checking. Decide whether to add the next job.
Every quarter. Review the playbook and the data rules. Tools change and staff turn over, so new people need the same training.
Name one person as the owner of all this, even if it is a small part of their job. Without an owner, it fades.
In person, on video, or both
- In person works best for hands-on practice, and for teams who work side by side. It is easier to catch confusion.
- Video works when your team is spread out or you need it quickly. Build in breakouts so people practice instead of just watching.
- Both is common: an in-person day to start, then short video sessions for the check-ins.
At Di-Hy, we train in person in Tampa Bay, on video, or both, on your real work, and leave you with a written playbook. That is the Train step of our AI Implementation, and you can read the Tampa-specific version in in-person AI training for Tampa teams.
Do it yourself first
You can run all of this without us. If you want a free outside view first, the SBA says SCORE mentors offer area-specific advice at no cost. The Florida SBDC at USF also lists an AI webinar for small business owners.
If you would rather have it done with your team, on your own jobs, with the playbook written for you, book a strategy call. Or start with the free AI audit so the training is built around the right three jobs.