Organising AI training: who learns what, in which format and when

What Art. 4 of the AI Act requires and what you should document, with slides and templates, is covered in the guide AI literacy under Art. 4 EU AI Act. This guide shows how to organise the training in your organisation.

Who needs which training?

Not everyone needs the same thing. Splitting by role saves time and addresses the participants’ questions:

Role	Content	When
Everyone who uses AI	What AI can and cannot do, checking results against the source, which data is allowed, the organisation’s rules, exercises with the approved apps	before access is switched on
People who build apps for others	good instructions, choosing documents, testing apps, sharing with groups	before the first build
Administrators	permissions and groups, approvals, log settings, updates	during setup
Managers	goals, rules, what AI does not replace, how to support their team	before the start in the team
Areas with sensitive data, such as health, social or personnel data	confidentiality, data protection, limits of use	before the apps are shared
New staff	basic training, adapted to their tasks	during onboarding

In what order do you proceed?

1. Clarify responsibility: One person coordinates; human resources development or continuing education organises dates and records.
2. Assess needs: Which groups get access, which apps, which data.
3. Set rules: Permitted systems, permitted data, contact person and reporting channel for errors. The rules are part of every training session.
4. Name contact persons: One person per department who has more in-depth training and helps colleagues with questions.
5. Schedule dates: Basic training shortly before access is switched on, not weeks in advance. What you do not apply straight away, you forget.
6. Follow up: After a few weeks, a short round with questions from practice.
7. Refresh and onboard: Once a year, when there are new apps or rules, and for every new person.
8. Document: Date, content, participants, next refresher.

Which formats work?

• Practice on your own tasks: Anyone who works on a real letter, set of minutes or enquiry from their daily work during the training takes away more than from any slide.
• Virtual: Saves travel, good for distributed sites and short refreshers.
• On site: Good for the start and for groups with little experience of technology.
• Short learning units: One topic in 15 minutes, for example before the shift starts or in the team meeting.
• Office hours: A fixed slot where questions are welcome. Lowers the threshold for asking.
• Contact persons in the departments: They pass on what the team needs and report where the training needs to improve.

Mixing groups across professions is possible, but separate groups address the questions more precisely: nursing staff ask different questions from administrative staff.

Does the staff representation have to be involved?

Often, yes. In Germany, the works council has co-determination rights in the implementation of in-house training measures (§ 98 BetrVG, the Works Constitution Act). If tasks change so that existing skills are no longer sufficient, it also co-determines the introduction of the training (§ 97 para. 2 BetrVG). Under federal staff representation law, co-determination applies to general questions of continuing education (§ 80 para. 1 no. 10 BPersVG, the Federal Staff Representation Act); the federal states and church law have comparable rules. The easiest way is to cover training in the agreement on AI itself, see Co-determination and AI.

How can you tell whether the training is working?

• Questions in office hours and support become more specific: fewer “How does this work?”, more “Does this also work for …?”.
• Contact persons report that colleagues check results before using them.
• Departments propose new apps themselves.
• There are few reports of data entered incorrectly, and the reports come through the agreed channel.

Usage figures for individual people are not a measure of this. Collecting them would amount to performance monitoring, see AI and performance or behaviour monitoring.

How does this work with basebox?

basebox trains your staff on the use cases of your organisation, virtually or on site: four hours, any number of participants. One session costs €900 virtually, €1,400 on site plus travel costs. Support over six months comprises twelve training sessions and a monthly office hour and costs €10,800. Setup in your IT includes two training sessions. Booked training sessions also cover AI literacy under Art. 4. All prices are list prices, net. basebox partners also offer training, for example Consileon (Partners).

Ready-made apps need little prior knowledge: your staff choose the task and get the result. The training can therefore focus on checking, data and rules.

Not legal advice: Information on co-determination and the AI Act reflects the status as of 11 October 2026. Have agreements reviewed for your organisation.