AI Training for Employees: What a One-Day Workshop Covers

AI training for employees should teach five things in one working day: what AI tools do well and where they fail, how to prompt with context, how to check every output, which data may go into which tool, and two or three workflows built on the team's own documents. A slide talk about the future of AI changes nothing on Monday. A day spent on your own emails, reports and client files does.

I run AI training for teams with ChatGPT, Claude and other tools, and I build my own companies with the same tools. In Lithuania the training runs through my company Retos galimybės (opens in a new tab). Below is a one-day agenda built on that format, with the reason behind each block. The rules part cites OpenAI's help centre and the EU sources, checked on 6 October 2026.

When AI training for employees does not stick

A course that only teaches the tool follows a familiar arc. People nod through a demo, try it once on Tuesday, get a mediocre answer and go back to the old way. The gap sits between the demo and their own work.

My team workshops follow from that. Each one takes one team, its own documents and its own clients. People finish with prompts and workflows they use the next day. Everything else in the agenda serves that end.

The one-day agenda

A half-day version keeps the first four blocks and one lab. It works for teams that already use AI daily and need the rules and the habits more than the basics.

1. What AI does well and where it fails

Start with a live task from the team, not a toy example. Then show the failure modes. OpenAI's own help centre describes hallucination as the model producing "responses that are not factually accurate", including "fabricated quotes, studies, citations or references to non-existent sources". It also warns that "the model may express high confidence even in incorrect answers".

Ask people to make the tool invent something on purpose: a court case, a statistic, a quote from their CEO. Once they have seen it lie with a straight face, they check outputs without being told.

2. Prompting with context

OpenAI's guidance comes down to two habits: be clear and specific, and refine in rounds. A four-part prompt puts both into practice:

• Role and context: who you are, who the output is for, what happened before
• Task: one verb, one deliverable
• Material: the document, the notes, the data, pasted or attached
• Format: length, structure, tone, language

Then the second round: "Shorter. Remove the second paragraph. Use the client's wording from the email." Stopping after the first answer wastes the tool. The second and third rounds are where the work gets good.

3. Checking outputs

Every output that leaves the team gets a human check. Four questions cover it:

1. Are all names, numbers and dates in the source material?
2. Did the tool add a fact you did not give it?
3. Would you sign this with your own name?
4. Does it follow your company's rules for that client?

4. Data rules

This block is short and saves the most trouble. Two facts from OpenAI's data controls page shape it. In a personal ChatGPT account, conversations can be used to train OpenAI models until the user switches off "Improve the model for everyone". In ChatGPT Business and Enterprise workspaces, OpenAI does not use content for training by default.

So the team needs one page that answers: which account do we use, what may never go in (client personal data, contracts under NDA, passwords), and who to ask when unsure. Write it before the workshop. The workshop is where people learn it.

5. Workflow labs

This is the reason for the day. Each person brings one recurring task. Typical ones:

• Sales: a first reply to an inbound request, a follow-up after a call, a proposal outline
• Finance: a monthly summary from an export, an explanation of a variance for a manager
• HR: a job ad from a role description, interview questions for a specific role
• Operations: meeting notes into tasks with owners and dates, a process written as a checklist

They build the workflow, run it on a real case, check it with the checklist, and save the prompt. In the second lab they work in pairs and swap: you test my workflow, I test yours.

6. Team library and next steps

The day ends with a shared folder of tested prompts, one owner for that folder, and a date two to four weeks out to review what stuck. In my company programmes I come back until the new habits hold. Without a return date, old habits win.

Who needs which track

One agenda does not fit everyone. A short shared morning, then tracks:

For founders and executives I run 1:1 sessions of 90 minutes on their real tasks: the inbox, the reports, the offers. A group session cannot get that specific.

AI training for employees and the EU AI Act

Article 4 of the EU AI Act requires companies that use AI systems to take measures to support the AI literacy of their staff. Since the Digital Omnibus took effect on 27 July 2026, the law asks for measures, not a guaranteed level. The European Commission lists four minimum steps: a general understanding of AI, your role as provider or deployer, the risks of the systems you use, and actions fitted to what each group already knows.

The agenda above maps onto those steps. Blocks 1 and 3 cover understanding and risk, block 4 covers your role and data, and the tracks fit the training to each group. The Commission also writes that relying on instructions for use or "asking the staff to read them might be ineffective", and that there is "no need for a certificate". Keep an internal record of who attended what. I covered the law in detail in AI Act Article 4: what the AI literacy duty asks of you.

Common mistakes when companies plan AI training

1. Choosing a course before choosing the tools. The data rules depend on the account type. Decide which tools are approved first.
2. Teaching only the tool. People forget menus. They keep workflows that save them an hour on Friday.
3. Generic examples. A recipe or a poem proves nothing to an accountant. Use the team's documents.
4. No checking habit. One invented figure in a client report costs more trust than the time AI saved.
5. No follow-up. Set the review date before the workshop ends.

Plan the day for your team

Tell me how big the team is, which tools it uses and what slows it down. On a 30-minute call you hear whether training fixes it or something else does. If it does, we agree in writing what I deliver, by when and at what price. Sessions run in Lithuanian or English, on site in Vilnius or remote. See the formats on my AI training page, read the daily habits in how to use ChatGPT at work, or book a call.