Documenting AI literacy: an employee record template

Start with how people actually use AI

A useful record describes work with the tools in use. Filing a course title and eight signatures tells a future reader little about whether the session covered quotes, customer conversations or employment decisions. Start by recording what people do and where mistakes could have consequences.

The template below connects that assessment to completed measures and supporting records. It can serve as a summary with links to further material. Documentation does not replace guidance or practice: a neatly labelled folder has never explained an incorrect AI answer to anyone.

Article 4 calls for measures suited to the use

Article 4 applies to providers and deployers of AI systems. As amended by Regulation (EU) 2026/1744, it requires measures to support AI literacy, taking account of technical knowledge, experience, education, training, context of use and affected people. The amended wording does not require a guaranteed level of literacy for each individual.

The European Commission’s questions and answers explain that a certificate is not required and organisations may keep internal records of training and other guidance. There is no single training format. Copying its practice examples does not automatically establish a presumption of compliance either. None of this establishes a fixed number of minutes or pages as sufficient.

Three questions shape the training plan

Match the learning need to the task. Our practical recommendation is a short assessment with the people using AI or approving its output. Walk through one actual work process together before planning the session.

• What does the person do with which system? Distinguish drafting text, approving a quote and setting up an automated workflow. Also record the business’s role in relation to each system.
• What does the person already know? Ask about tools and prior experience. Have them explain how they would spot and check an invented claim in a sample; this is a planning aid, not a claim that the law requires an exam.
• What could go wrong, and who would be affected? Record risks such as incorrect prices, unauthorised data entry or unchecked promises. Use these to choose learning content, named contacts and approval steps.

Template: document employee AI literacy measures

The template links each measure to its purpose. Copy these fields into an internal document and complete them for the actual use case. Supporting documents can contain more detail; this is a starting point, not an authority-approved record.

• Date and ownership: [date], [responsible role or person], [contact for questions].
• System and use: [tool and relevant version], [task], [permitted use], [excluded use].
• Business’s role: [role in relation to this system and basis for that assessment].
• People involved: [names or traceable participant list], [tasks], [external people acting on the business’s behalf, where relevant].
• Starting point: [prior knowledge, experience, relevant education and training], [specific learning needs identified].
• Risks and affected people: [possible errors], [people affected], [checks and approvals].
• Completed measures: [date and actual duration], [content], [person providing guidance], [examples and materials used].
• Practical application: [task completed], [errors discussed], [questions still unresolved].
• Supporting records: [confirmation of attendance], [location of materials], [applicable usage policy and its version date].
• Next steps: [further guidance needed], [responsible person], [date], [trigger for another review].

A completed example makes open questions visible

This example describes a fictional trades business. Two employees use an approved chat tool to draft quotes from anonymised notes, with the office manager approving the finished documents. One employee already uses the tool; the other is new to it. The details are illustrative.

• Learning need: Both employees should recognise invented descriptions of work. The new employee also practises entering data. Prices come from the approved price list.
• Measure on 2 October 2026: Work through a sample quote together and compare it with the input list. File the materials and record actual attendance.
• Observation: An invented delivery date is identified and removed from the practice draft. Approval follows a check of scope, price and timing.
• Open question: The new employee cannot yet formulate a clarification request independently. The office manager schedules another exercise before the next independent draft.

Build guidance around a real task

Good guidance connects rules to a manageable work process. For quote drafting, we would put a deliberately flawed draft alongside the correct input and the business’s usage policy. Discuss missing information, additions made by the tool and who decides what can be sent.

• Tool and limits: Demonstrate the approved application and use an example to explain why a fluent answer may still be wrong.
• Data: Define which information may be processed in the chosen system. Practise using placeholders with the guide to redacting customer data before prompting.
• Review: Use the quote prompt and checking checklist to compare the draft with supported information.
• Errors and questions: Have participants practise stopping and reporting a problem. Assign responsibility for unanswered questions.

Let tasks and risks determine the depth

Different responsibilities need different levels of guidance. Drafting text involves different steps from configuring integrations or using AI suggestions in employment decisions. Our example focuses on text drafts with human approval. It does not cover every application or the additional obligations that may apply to particular systems.

For sensitive decisions, automatic actions or a system whose classification is unclear, we recommend qualified help with the intended use and risks first. External training can also be useful in a small office if nobody internally can explain the application or work through errors reliably. Choose relevant content and opportunities to practise; a particular course length is not a quality mark.

Keep the record connected to everyday work

Revisit the record when something changes. New tools, new tasks, new employees and notable errors are practical reasons to review learning needs. Keep completed measures separate from planned actions. Scheduled training has not yet taken place.

We would complete the template while introducing the first workflow. The German-language guide to a first AI workflow trial helps define responsibilities and limits; our internal AI workflow services explain how we can support that work. This practical guidance does not replace a legal assessment of the specific case.

Common questions about AI literacy records

Is one page enough? It can provide a summary that points to supporting records. What matters is whether the measures, learning needs and open questions are traceable; page count alone cannot establish that.

Does everyone need the same course? Planning around responsibilities and prior knowledge is more useful. In the example, both employees practise checking quotes while the new employee gets extra guidance on entering information.

What about external people? Article 4 also covers people operating or using AI systems on the business’s behalf. Include the relevant guidance and records for those tasks too.

Where do we start? Pick one application people actually use, complete the first six fields in the template and arrange the practical guidance it calls for with the responsible person.