AI training for teams: a practical workshop, tools and a policy you can copy

Good AI training for employees is a half-day workshop where every person works on a task from their own job, with a written policy that says what data stays out of the chatbot. Generic demos rarely change how people work. A finished piece of real work, done faster, is what people keep using.

Short answer

Pick one approved tool on a business plan, write a one-page usage policy, and run a half-day workshop where each participant brings a real task. Measure after 30 days by checking whether people still use it weekly and how much time those tasks take now.

What does a practical AI workshop for a team look like?

Before the workshop I ask each participant for one or two tasks they do every week and dislike: a monthly report, replies to the same customer questions, meeting notes, a price comparison, a job ad. Those tasks become the material. Nobody practises on invented examples about a fictional coffee shop.

I taught teenagers to programme for seven years. The main thing it taught me is that people learn by doing their own work with someone next to them, not by watching slides. So most of the session is hands-on, I move between people, and the theory comes up when someone hits a problem that needs it.

A sample half-day agenda for up to about ten people:

1. 0:00 to 0:20. What these tools are good and bad at, shown on one task from the group.
2. 0:20 to 0:40. The usage policy: what data stays out, which plan the company uses, who checks output.
3. 0:40 to 1:40. Each person works on their first real task. I help one by one.
4. 1:40 to 1:55. Break.
5. 1:55 to 2:40. Writing better instructions: context, examples, format, and asking the tool to check its own answer.
6. 2:40 to 3:30. Second real task, harder. Files, tables, long documents.
7. 3:30 to 4:00. Each person saves one reusable prompt for their task and names the task they will try next week.

ChatGPT, Claude or Copilot: which should a team use?

All three are good enough for everyday office work. The choice depends more on where your team works than on which model wins a benchmark this month. Prices below are from the vendors' pages at the time of writing and change often.

Sources: OpenAI business pricing, Claude pricing, Microsoft Copilot for business. Microsoft also listed a first-year promotional price lower than this at the time of writing.

My opinion: do not buy two tools for the same team at the start. Pick one, run the workshop on it, and revisit in three months with real usage data.

What data should never go into a chatbot?

Business plans change the default on training, which matters. According to OpenAI, ChatGPT Business has no training on business data by default. Anthropic's commercial terms say it may not train models on customer content from its commercial services. Microsoft's Copilot privacy documentation says prompts and responses are not used to train foundation models.

Not training is not the same as safe for anything. The data still leaves your systems, and personal data still falls under GDPR. My rule for teams is that these stay out, whatever the plan:

• Passwords, API keys, access tokens and bank details.
• Health data and other special categories of personal data.
• Customer lists or exports with names, emails and phone numbers, unless the company has approved that use.
• Documents under an NDA or a client contract that forbids sharing.
• Unpublished financial results and deal terms.
• Anything typed into a personal or free account rather than the company plan.

A one-page AI usage policy you can copy

Keep it to one page. People follow rules they can remember.

1. Approved tools. Name the tool and plan the company pays for. Work happens there, not in personal accounts.
2. Data that never goes in. Copy the list above and add anything specific to your business.
3. Customer data. Remove names and contact details first, or get approval from the person responsible for data protection.
4. Checking output. A person reads and checks every AI text, number or translation before it reaches a customer, a contract or a public page.
5. Facts and figures. Check every number, date, legal claim and quote against the source.
6. Telling people. Say when a customer is talking to an AI system, not a person.
7. Who to ask. One named person answers questions and updates the policy.
8. Review date. Read the policy again in six months, because the tools change.

How do you measure if AI training worked after 30 days?

Write down the baseline before the workshop: each participant's tasks and how long they take now. After 30 days, check these:

• How many participants used the tool at least once a week. Most business plans show usage in the admin panel.
• Time per task now compared with the baseline, for the tasks people brought.
• How many reusable prompts or templates the team saved and shared.
• How many policy questions came up, and whether anyone broke the data rules.
• One workflow per team that is now written down and repeatable.

If weekly use has dropped, the usual cause is that the tasks were too far from daily work. A one-hour follow-up on the tasks people gave up on fixes more than another general session.

Once a task is repeated the same way every week, it may be worth automating rather than prompting by hand. That is a separate step, covered under AI automation.

What does the training cost?

I run AI training for teams at €45 an hour, and I prepare the session around the tasks your team sends in advance. If you want to talk it through first, get in touch. The first 30-minute call is free.

FAQ

How long should an AI workshop for a team be?

Half a day is enough for a first workshop with up to about ten people, as long as everyone works on a real task. A follow-up session of one or two hours after a month catches the problems that only show up in daily use.

Which is better for a team: ChatGPT, Claude or Microsoft Copilot?

If your team lives in Outlook, Word and Teams, Copilot fits best because it works inside those apps. For writing, analysis and long documents, ChatGPT Business and Claude Team are close, and the best test is to give the same real task to both during a trial.

Do ChatGPT Business and Claude Team train on company data?

According to OpenAI's and Anthropic's pricing pages at the time of writing, neither trains models on business content by default. Microsoft says prompts and responses in Copilot are not used to train foundation models. Personal and free accounts have different settings, so staff should use the company plan for work.

Does the EU AI Act require AI training for employees?

Article 4 of the AI Act has applied since 2 February 2025 and asks companies using AI systems to take measures to support AI literacy of their staff. The European Commission says no certificate is needed, and keeping internal records of training is one way to show you did it.

What should an AI usage policy for employees include?

Which tools are approved, what data must never go in, who checks AI output before it reaches a customer, and who to ask when unsure. One page that people actually read works better than a long document nobody opens.

Evaldas Doda

Founder of DodaCraft, developer. I build websites, online stores and internal systems for businesses in Lithuania and abroad. I taught teenagers to programme for 7 years.