AI literacy training records describe who an initiative was intended for, what guidance it covered, when it ran, and what follow-up was chosen. Keep participation evidence separate from knowledge-check answers and practical skill observations. The template below is an optional internal documentation tool, not a prescribed Article 4 form or a compliance certificate.
Key takeaways
- Record the audience, AI use case, approved guidance version, delivery and follow-up.
- Explain how you established participation instead of treating a display name as verified identity.
- Preserve the conditions of any knowledge check, including collaboration and source access.
- Decide retention and access with your responsible owner; avoid collecting unnecessary personal information.
What does Article 4 guidance say about documentation?
The European Commission's AI literacy Q&A says organisations can keep an internal record of training or other guiding initiatives and that a certificate is not needed. It also describes a context-sensitive approach to AI literacy. This does not turn a generic attendance list, quiz score or purchased course into proof of compliance.
This article provides practical documentation suggestions, not legal advice. Have your responsible legal or compliance adviser check the rules that apply to your organisation and AI systems. Use the current official guidance rather than assuming that an old deadline summary or checklist remains sufficient.
What can you put in an internal record?
Use only the fields that serve your purpose. The following is an original suggested template, not a required list from a regulator.
| Field | What to document | Why it helps |
|---|---|---|
| Initiative and owner | Name, responsible team and contact | Someone can explain the record later |
| Intended audience | Roles, relevant experience and AI tasks | Shows whom the guidance was designed to support |
| Tools and context | Approved tool/use case and relevant information limits | Connects the activity to actual work |
| Objectives | Decisions or skills the initiative addressed | Defines what any check can examine |
| Source version | Policy title, version, effective date and reviewer | Preserves the basis for the answers |
| Delivery | Date, format, materials and facilitator | Describes what was offered or delivered |
| Participation | Your approved attendance evidence and identity-matching method | Keeps verified attendance distinct from submitted names |
| Checks and observations | Questions/tasks, conditions, results and limitations | Makes the evidence interpretable |
| Follow-up | Clarifications, unresolved issues, owner and next review | Records what the organisation chose to do next |
| Access and retention | Who may view the record and the approved retention rule | Keeps recordkeeping proportionate |
Do not collect personal identifiers just because there is space in a spreadsheet. Your record owner should decide the minimum needed for the purpose and the appropriate access, retention and deletion process. Restricted incident details and employee assessments do not belong in public game questions or AI prompts.
How do you distinguish the different kinds of evidence?
Attendance evidence describes participation under your organisation's process. A knowledge check samples responses to particular questions. A practical observation examines an action under stated conditions. Keep those meanings visible even when the information is stored together.
For example, an employee may correctly choose to verify a source but still need help finding the approved reference. Another may know the process but be unable to access the tool. Record the unresolved question or operational obstacle rather than labelling every difficulty a lack of knowledge.
For broader programme design, the NIST AI Risk Management Framework is a voluntary framework for managing AI-related risks. It can inform a risk discussion; it does not prescribe this training-record template or establish Article 4 compliance.
What does a worked entry look like?
The following entry is entirely fictional. It illustrates the template and is not a customer case or recommended policy.
| Field | Fictional entry |
|---|---|
| Initiative | Customer-summary AI workshop, owned by Learning Operations |
| Audience and context | Support agents drafting internal summaries in an approved tool |
| Source | Example AI Use Guide v2, reviewed by its fictional policy owner |
| Objective | Choose permitted source material and the required human-review route |
| Delivery | November 12, 2026; facilitated walkthrough plus three questions |
| Participation | Separately recorded using the company's approved attendance process |
| Check conditions | Individual answers, approved guide available, no speed requirement |
| Observation | A participant selected the approved review route but could not locate its instructions |
| Follow-up | Facilitator demonstrated the route; system owner to repair the outdated guide link |
| Record handling | Access and retention governed by the company's approved record process |
Suppose the fictional guide allows public information only in the AI tool. A customer complaint contains account details. Which action stays within that rule?
- A. Paste it because the tool is approved.
- B. Remove the customer's name but retain the account details.
- C. Use a fictional example without customer information, within the approved rules.
Answer: C. Tool approval does not remove the stated information restriction. A ignores that limit, and B still includes restricted details. Keep this explanation with the facilitator's source record. A correct choice alone does not show what a participant would do in a real customer workflow.
What does a BrainFusion report contribute?
Use BrainFusion for the question-practice part of the initiative. Trainers can enter questions manually, import a reviewed CSV or review AI-generated drafts from approved material. Learners can join a session without learner accounts; the creator needs an account.
A saved scored-session result includes the submitted display name, score, maximum possible score, optional duration, server submission time and game type. When supplied, question responses include the text, submitted answer, correctness, order and round. Review the full reporting and identity limitations before using results in your records.
Names are self-entered. They are not verified employee identifiers. Best-score summaries group by submitted name, while question statistics can include repeated submissions. Shared answers also do not show each person's understanding. An organisation that needs verified participant records must maintain its own appropriate mapping and evidence.
Reports can be printed and sessions exported as .bfsess files. Neither is a signed attendance certificate, a policy sign-off or proof of practical proficiency. The facilitator supplies the explanations and separate observations. The public AI literacy sample is useful for inspecting content; it does not create a saved employee report.
How should you review the record after training?
Check that the record says what happened, not what the activity was hoped to achieve. Verify that a future reader can identify the source version, conditions and unresolved work without inferring attendance or competence from a score.
When a policy, approved tool or reporting route changes, decide whether the guidance and activity need revision. Keep the change and any resulting follow-up with the record. The timing should follow the actual change and your organisation's requirements, rather than an invented universal refresher interval.
Frequently asked questions
Is there one mandatory Article 4 record template?
The Commission's documentation answer describes internal records and does not prescribe the template above. It says a certificate is not needed. Ask your adviser what evidence is appropriate for your particular duties and circumstances.
Can a leaderboard stand in for attendance?
No. Submitted names are not authenticated employee identities, and best-score grouping can merge different people using the same name. Keep attendance evidence under the organisation's approved process.
How long should we keep these records?
There is no universal period supplied by this article. Your record owner should set an appropriate rule with the relevant legal, privacy and operational requirements in mind. Check the provider's data-handling terms separately from your own record-retention policy.
Start with one reviewed decision and a clear follow-up. Use the employee knowledge-check template, then create a practice game when the source and answer key are ready.
Sources and method
Official Commission Q&A and NIST's framework overview were checked November 12, 2026 and linked where used. The template, fictional entry and worked question are BrainFusion's original suggestions, not regulatory instructions or a validated assessment. Product fields and limits were checked against current submission, reporting, ranking and export code. BrainFusion publishes this guide and provides the optional practice tool described.
