LUKA SANDVOSSBLOG
← ALL POSTS

Digital literacy

AI literacy at work: practise reviewing the answer

Run an AI literacy exercise with a real source document. A practical review record helps staff track evidence, unresolved questions and release decisions.

Luka Sandvoss · · 4 min read

Auf Deutsch lesen →
A person working on a laptop beside documents at a desk by a window.

Before booking an AI workshop, ask a colleague to show you the last piece of AI-assisted work they checked. A grant summary, a draft email or an internal briefing will do. Look at what happened between generating the text and letting somebody rely on it.

That interval is where I would focus a team's training budget. People need enough understanding to decide what deserves checking, find the relevant evidence and recognise when the tool is unsuitable for the task.

What the EU guidance asks of employers

For organisations operating in the EU, the Commission's July 2026 guidance matters. It explains that the amended Article 4 still requires providers and deployers to support the development of AI literacy, while no particular individual proficiency level is mandated. Training should reflect people's background and the circumstances in which AI is used. European Commission: AI talent, skills and literacy.

The exercise below is my proposal for putting review skills into practice.

Try a document with an exception in it

Imagine a small charity preparing an overview of a public funding opportunity. Use a published call for applications, with no personal information, and ask one colleague to produce an AI-assisted summary. Give another colleague the original.

The second person's job is to identify the conditions that could change the charity's decision to apply. They might find an eligibility restriction in the final paragraph or a deadline that applies only to one stage. Compare those findings with the generated summary.

Keep the disagreement on the page. If the notice says that only registered charities may apply, a summary that leaves out that restriction is incomplete. Ask the reviewer to mark the original sentence and correct the draft.

The output can be grammatically excellent and still fail this exercise.

Keep a short review record next to the draft. For the fictional funding example, it could read:

  • Claim: our organisation is eligible.
  • Evidence: link to the eligibility paragraph in the original call.
  • Unresolved question: does our legal status meet that condition?
  • Release decision: hold the eligibility statement until the project lead has checked it.

The next reviewer can pick up the question without reconstructing the whole chat. Save the corrected summary with this record.

In an IHK Düsseldorf statement dated 24 August 2026, continuing-education director Dr Kerstin Vorberg argues that knowing numerous tools is a poor measure of AI competence. Her emphasis is on applying them to work. This exercise gives a team something specific to assess.

Decide who may rely on the result

An exercise becomes useful when it changes the workflow. In this example, the person writing the summary would attach the original source and flag anything they couldn't verify. A named colleague would check the eligibility criteria before the summary went to the project team.

Keep those instructions beside the working document. Include the approved tool and a contact for questions about acceptable inputs. Staff shouldn't have to improvise the rules every time they open a chat window.

Someone using AI to brainstorm headings needs a different level of review from someone preparing material that affects access to support. The Commission's FAQ also discusses context-specific measures and does not prescribe a universal training certificate. AI literacy questions and answers.

Check the habit after the workshop

A month later, introduce a different funding notice. Watch whether people find its exceptions without being prompted. Record where they hesitate and improve the instructions at those points.

Keep a record of the points people missed. That gives the next session a purpose; attendance alone does not tell you what to teach next. This remains an internal exercise, rather than an assessment of legal compliance.

I would spend the first session on that review, even if it leaves less time for trying new tools.

For practice with deceptive messages and AI-generated imagery, see Medienkundig's security awareness pilot. The German-language offer sets out how an organisation can agree a training focus and repeated sessions. The source-review exercise above can be run with a public document you already use.

Sources checked on 10 September 2026. Cover: photograph by Windows / Unsplash. Illustrative photograph, not a Medienkundig project or training session.