How the half day runs

A morning or an afternoon, at your office in Ghent or remotely. Everyone works on their own laptop, with the account they will use tomorrow.

  1. Beforehand, a 30-minute call. Which tools and licences you have, and which tasks take the most time. You pick three or four everyday documents. Anything sensitive we anonymise together first.
  2. Where your text goes. What happens to what you type into ChatGPT, Copilot or Claude, and why the account you use makes the difference.
  3. Practice on real work. Summarise a long email, draft a reply, pull a table out of a PDF, put a report into your house style.
  4. Where it goes wrong. We let AI make mistakes on your own documents: an invented figure, a source that doesn't exist, a legal reference that is wrong.
  5. Finish the policy. In the last half hour we write the rules that fit your team, together.

After the session you get the policy as an editable document, with the attendance list.

What you can paste where

The account matters more than the tool. The same ChatGPT can be safe or unsafe for client data.

What you can paste into which kind of AI account
Account What can go in Why
Personal: ChatGPT Free, Plus or Pro, Claude Free, Pro or Max No data about clients or colleagues. Public text and your own notes without names. In ChatGPT, training on your chats is on by default; you can switch it off. In Claude, each user chooses. Your company has no data processing agreement for it.
Business: ChatGPT Business or Enterprise, Claude Team or Enterprise, Microsoft Copilot with your work account Work documents, within what your AI policy allows. No training on your data by default, and the provider offers your company a data processing agreement.
Nowhere Passwords, API keys, national ID numbers. Health data only if you decide so explicitly. A business account also keeps what you type in its history.

Sources: OpenAI, business data and data controls in ChatGPT; Anthropic, updates to consumer terms (August 2025); Microsoft, data, privacy and security for Microsoft Copilot. Checked on 25 October 2026.

ChatGPT, Copilot or Claude: we work with what you have

If you work in Microsoft 365, Copilot is often the obvious choice: it works in Outlook, Word and Teams and finds your own files. For summarising long documents, rewriting text or analysing something you hand it, ChatGPT and Claude do well too. Paying for two tools side by side is rarely needed.

Copilot sees what you can see

Copilot only shows files the user already has access to. That sounds safe, until it turns out the payroll folder on SharePoint is open to everyone. Then Copilot finds it too. Before you roll out Copilot, check who can open what.

Your admin can also switch on Anthropic models inside Copilot. According to Microsoft, those are currently outside the EU Data Boundary.

Source: Microsoft, data, privacy and security for Microsoft Copilot, checked on 25 October 2026.

The policy fits on one page

Which tools are allowed, what never goes in, what you always check and who to ask when in doubt. Plus the date of the training and who attended.

Made-up example AI policy · Example Office BV · version 1, October 2026

How we use AI at work

  1. Which tools. Microsoft Copilot Chat, signed in with your work account. ChatGPT only in the office Business workspace. Personal accounts are not for work.
  2. Never paste. National register numbers, health data, passwords and access codes. Nor any document a client asked us to keep confidential.
  3. Always check. Figures, names, dates, legal references and sources. What AI writes goes out under your name.
  4. If in doubt. Ask Els (office manager) before you paste anything. One question too many is fine.
  5. On record. Training held on 14 October 2026, 11 attendees, list attached. Review in April 2027.
From the one-page policy your team leaves the session with. The office is made up.

What the AI Act actually requires

Article 4 of the AI Act has applied since 2 February 2025 to every company that deploys AI systems, including staff using ChatGPT for work. Since Regulation (EU) 2026/1744 came into force on 27 July 2026, you have to take measures to support AI literacy. You don't have to guarantee that everyone reaches a set level, and no certificate is required.

Do write down what you did: when, with whom, what was covered. This half day with the policy is such a measure, and the attendance list is your record. If you use AI to make decisions about people, in hiring for example, stricter rules apply that are outside this session.

Sources: Regulation (EU) 2024/1689 (AI Act), Article 4; Regulation (EU) 2026/1744; KI-Campus, AI Omnibus recalibrates Article 4.

Half a day, up to 12 people

  • The 30-minute call beforehand
  • Half a day on your own documents, at your office in Ghent or remotely
  • The one-page AI policy, as an editable document
  • The attendance list as your record

With more than 12 people we plan a second half day. We work at a fixed day rate or at a fixed price per outcome. You know the price before we start.

What comes next