Build an AI literacy exercise around the job

Build an AI literacy exercise around a decision someone must make in their actual role. Ask the person to handle a realistic but safe task, explain what they would check, and show when they would stop or ask for help. The useful result is a clearer learning need and a better working instruction, supported by a short record of what was practised.

This guide offers a practical learning design for Irish workplace teams. It does not determine compliance with the EU AI Act or other laws. The European Commission’s AI literacy FAQ emphasises the context of use and the knowledge of the people involved; it does not prescribe one training format or require a specific certificate. The exercise below is our suggested implementation method, not a Commission-approved course. 1

Choose one work decision, use approved fictional or public information, and make the expected response observable. Give preparers, reviewers and approvers different responsibilities. Use feedback to improve both learning and instructions. Keep a proportionate record without collecting unnecessary employee or customer data.

Start with a task that staff already understand: preparing an internal briefing, drafting a customer response for review, or organising a public supplier notice. Write one sentence explaining what AI may contribute and one explaining what remains the person’s responsibility. A learner needs to know what a usable result looks like before being asked to critique it.

Choose the error that matters to that task. For an internal briefing it might be a missing exception; for a customer draft it might be an unsupported promise. Avoid a collection of spectacular but irrelevant failures. Ask the process owner which misunderstanding would most likely lead to a bad decision, then build the exercise around that point. Our guide to an AI workflow with meaningful human checks helps define that boundary.

The following card is a suggested design aid. Complete it before inviting staff, and have a competent colleague check that the expected response is defensible. A facilitator should not improvise an answer key from whichever response the AI happens to produce.

Keep the source pack fixed during the session so participants can compare reasoning. If live AI is unnecessary, use a prepared sample output. A static sample can expose a learning gap without uploading documents or giving a tool access to operational systems. Label it clearly as an exercise, including any deliberately incorrect statement.

Ask the preparer to identify permitted inputs and separate supported facts from assumptions. Ask the reviewer to explain which evidence changes the recommendation. Ask the approver to decide whether the output can be released, needs revision or should be held. These roles may belong to one person in a small team, but the decisions still need to be distinguished.

Offer an accessible explanation of the task and allow staff to ask questions about unfamiliar terminology. A person who has never worked with the underlying process may need process training before AI practice. NIST’s generative AI profile treats human–AI interaction as a risk area; our practical inference is that learning should include the human decision, not just operating the interface. 2

Assume an invented Dublin team prepares an English-language briefing for colleagues in Ireland and the UAE. The fictional source says a supplier’s proposed service change is open for comment. A prepared AI-style summary incorrectly states that the change is already mandatory. No real supplier, employee performance or regulatory notice is represented.

The preparer should flag the difference between a proposal and an adopted requirement. The reviewer should point to the relevant source wording and correct the summary. The approver should hold any operational instruction based on the false statement. Ask each person to explain what would change if the source were unavailable or the affected country were unclear.

A correct answer copied from another participant is weak evidence of understanding. Give a second short example with a different surface detail but the same decision problem. Use the response to decide whether further explanation or supervised practice would help; do not present the exercise as a validated assessment of overall competence.

Use this fictional source statement: “The service provider is consulting customers on a possible revised support window. Comments close on 20 November. No implementation date has been agreed.” The deliberately flawed sample summary says: “The revised support window becomes mandatory on 20 November.” Neither statement describes a real provider or requirement.

A corrected output is: “The provider is consulting on a possible support-window change. Comments close on 20 November; implementation has not been agreed.” This is an answer key, not a record of employee performance or an exercise already delivered.

For a second case, remove the source's jurisdiction or affected service. Ask participants what remains uncertain and which clarification is needed. Record observed reasoning only after the exercise occurs. Completing this local task can reveal a learning need; it does not establish general professional competence, certify AI Act compliance or justify a claim that everyone has achieved a prescribed level of literacy.

Retain the task version, date, roles involved, material used, main learning gaps and agreed follow-up. Record the minimum personal information needed for the organisation’s legitimate learning process, with appropriate access and retention. Do not turn a learning log into an undisclosed employee scoring system.

Separate attendance, observed task performance and authority to use a live system. They answer different questions. Where several people make the same mistake, revise the operating instruction or source pack as well as the training. Repeat practice after a meaningful change to the task, tool or review responsibilities; choose the interval according to the work rather than inventing a universal annual rule.

It can record participation or an assessment defined by its provider. Check what it actually assessed. This suggested exercise examines one local task and does not establish legal compliance or broad professional competence.

No. Use invented or suitable public material whenever it can reproduce the decision problem. Any move to sensitive information requires the organisation’s approved handling arrangements.

From AI to SI is a QSE Advisory product and related-brand resource. It is not independent evidence for this article or a claim of regulatory endorsement. Assess any service against the task and learning need you have identified.

Prepared with AI-assisted research and drafting. Suggested methods and hypothetical examples are editorial guidance, not reported client results. QSE Advisory is part of QSE Advisory. General information; obtain competent review for consequential use.

1. European Commission AI literacy questions and answers — Live FAQ checked 2 October 2026; used for contextual learning and no mandatory certificate, not disputed enforcement dates.

2. NIST Generative Artificial Intelligence Profile — Final NIST AI 600-1, July 2024; voluntary risk-management guidance, including human–AI configuration.

This article is general information. Verify current requirements and seek qualified advice for your specific circumstances.

Spotted an issue? Contact our editorial desk.