How do you train a team on AI without a big program?

The instinct is to book a workshop. Somebody comes in for a day, shows forty features, everyone nods, and by Friday the tool is closed again. Information without implementation has no value. What sticks is smaller and slower: a person, their own task, and someone sitting beside them the first time it works.

Do most businesses train their staff on AI?

No. In the Census Bureau’s AI supplement, fielded November 2025 to January 2026, training staff was one of the two most common changes firms made to use AI, and it still applied to only about 15% of AI-using firms. 64% made no organizational adjustments at all. Most teams were handed a login and left alone.

Intentions run ahead of practice. In the Bureau’s earlier survey, covering late 2023 and early 2024, 20.8% of AI-using firms had trained existing staff, while 41.9% of firms expecting to use AI said they planned to. Planning to train is common. Doing it is rare. That’s your opening, because even a light routine puts your team ahead of most.

What should people learn to do with AI first?

The four things workers already use it for most. Among U.S. workers who have used AI chatbots for work, Pew found 57% use them to research or find information, 52% to edit their writing, 47% to draft documents and 40% to summarize information. Those are the jobs where a beginner gets a useful result on day one.

Notice what isn’t on that list: anything that goes to a client unread, anything with patient data, anything that commits the business to a price or a promise. Those come later, if at all, and only with a person approving. The first month is for low-risk work where a bad output costs a few minutes.

How do you train a small team on AI in four weeks?

One task per person, practised on real work, with a short weekly check. Week one sets the rules and the tasks. Week two is thirty minutes side by side with each person. Week three adds a second task. Week four writes down what worked so it survives staff changes. It costs about an hour per person.

1. Week one. Write the one-page rules: which tool and account, what never goes in, who checks output. Name one task per person, taken from their own week.
2. Week two. Sit with each person for thirty minutes on their task. Let them type. When the first answer is weak, show them how to say what’s wrong and ask again. That one habit is most of the skill.
3. Week three. Fifteen minutes as a group. Each person shows one output they kept and one they threw away. Everyone picks a second task.
4. Week four. Write down the prompts and steps that worked, in plain words, in a shared document. That document is your training program from now on.

Who runs the sessions matters too. Pick the person on your team who already uses AI most, and sit in on the first few yourself. They know the work, they speak the team’s language, and they’ve already hit the mistakes a beginner will hit. An outside trainer can come later, once you know which tasks deserve deeper help. For the first month, the expertise you need is already on your payroll, and asking that person to teach tells everyone else the skill is valued here.

The side-by-side session matters more than anything else in the plan. People give up on AI after one bad answer because nobody told them the first answer is a draft. Watching someone say "shorter, and use the client’s name" and get a better result does more than any slide.

What if some people aren’t interested?

Start them on the task they dislike most, not the one you think is most valuable. In Pew’s survey, 22% of workers who don’t use AI chatbots said a major reason was that they weren’t interested, and 36% said there was no use for them in their job. Interest follows the first hour AI gives someone back.

Age matters less than people assume, and it’s worth not making assumptions about who will take to it. Pew’s 2026 summary shows younger workers more likely to have used ChatGPT at work, 38% of employed adults under 30 against 18% of those 50 and older in early 2025. That gap is real and it’s also a reason to pair people across it rather than leave the older half of the team to figure it out alone.

How do you know the training worked?

Check the named tasks after a month. Did the follow-up emails go out faster? Did the proposal drafts need fewer rounds? Ask each person whether they’d keep using it for that task if you took the tool away tomorrow. Pew found 40% of workers who use chatbots for work call them extremely or very helpful for doing things more quickly, and 29% say the same for quality. Your own numbers, on your own tasks, are the only ones that should decide what you do next.

When a task is running reliably and a person is signing off on it every time, it’s a candidate to write up properly, so an agent can eventually take it on with that same person approving.