AI Literacy is not prompt training: from AI policy to safe behaviour in the workplace

Many organisations start today with a logical question: how do we make sure our employees learn to use AI well?

But that is exactly where the first misconception lies.

Because “we need prompt training” is rarely the real question. Usually something else is going on. Employees are already using AI, but not always consciously. Teams experiment with customer emails, internal documents, meeting notes or HR notes. Legal writes guidelines. IT looks at tooling and security. HR and L&D are asked to provide “something on AI”.

But on the work floor, the doubt remains.

Can I have this text rewritten? Can I summarise this document? What information can I enter? When do I need to check the output? And who remains responsible if AI helped me?

If you then mainly look at tools, prompts or productivity, you solve a skills question. Not the underlying behavioural question.

And that is exactly where Keep Growing iDesign takes a different view.

, ,
7–10 minutes

We don’t start from tools, but from behaviour

For many organisations, AI Literacy starts with explanation. What is generative AI? How does a prompt work? Which tools are there? What are the risks?

That is a logical starting point, but it is not enough.

AI Literacy only becomes relevant when employees make better decisions in their own work. Not just working faster, but working more safely. Not just producing nicer output, but dealing more critically with what AI gives back. Not just knowing that privacy matters, but recognising in the moment which information does not belong in an AI tool.

That is why we don’t start with the question of which AI tool someone needs to learn to use.

We start with the question of what needs to happen differently in practice.

Which choices do employees need to make better? Which mistakes do we want to prevent? Which risks need to be recognised faster? Where should AI be used in a useful, safe and responsible way? And when should someone stop, check or escalate?

That then determines everything. The content of the training. The scenarios. The feedback. The job aids. And the extent to which an AI Literacy programme needs to be more than a short awareness module.

AI Literacy is about professional judgement

Prompt training teaches employees to write better instructions. That can be valuable. Especially if you want to use AI for structure, language, summaries or ideas.

But prompt training is no guarantee of safe AI use.

An employee can know perfectly well how to write a good prompt and still enter confidential customer information. Someone can quickly improve an AI output and still not recognise that the content is wrong, biased or incomplete. A team can experiment enthusiastically and at the same time create new risks around privacy, compliance or reputation.

So AI Literacy calls for a different bar.

The real question is not: can employees use AI?

The real question is: can employees judge when, for what and under which conditions AI use is responsible?

That is professional judgement. And professional judgement doesn’t develop just by giving explanations. You have to practise it in recognisable situations.

A simple model helps employees decide

AI policy only becomes usable when employees can apply it to their own work. That is why a simple decision model helps.

Not as extra theory, but as something to hold on to at the moment someone wants to use AI.

The 5 questions for safe AI use

1. What do I want to use AI for?

Rewriting a text is different from preparing input for customer advice. Speeding up a brainstorm is different from summarising HR notes. The closer AI gets to people, customers, decisions or sensitive processes, the more critical the trade-off needs to be.

2. What information am I entering?

This is often the point where policy and behaviour drift apart. Employees think they are “just asking for a bit of help”, but meanwhile copy customer names, internal figures, contract information or HR context into an AI tool.

3. Who or what could be affected by the output?

A wrong answer in a brainstorm carries different weight from a wrong answer in a customer email, appraisal conversation, compliance process or safety instruction. The possible impact determines how much checking is needed.

4. How do I check the result?

AI can sound convincing and still be wrong.

“Check the output” is too vague. Employees need to know what checking means in their context: content, source, completeness, tone, bias, legal sensitivity or possible harm.

5. Who remains responsible?

AI can support, speed up and structure. But AI does not take over responsibility. The employee, manager or organisation remains the owner of what is ultimately used, shared or decided.

This model is deliberately simple. Not because AI is simple, but because employees don’t need a legal handbook in the moment of work. They need a usable thought process.

Good learning design makes policy workable

An AI policy is important. But a policy does not change behaviour on its own.

The translation happens in learning design.

That means you don’t just tell employees what is and isn’t allowed, but let them practise with situations they recognise. Rewriting a customer email. Summarising an internal document. Preparing feedback. Structuring meeting notes. Translating a text into customer language.

Most of the learning value doesn’t lie in the easy examples. It lies in the grey areas.

Can you use a document if you remove the names? When is information sufficiently anonymised? When do you only use an approved internal tool? When should Legal, IT or a manager be involved?

That is where policy becomes practical.

That is why a generic AI module usually has limited effect. HR, finance, sales, operations and customer service use AI in different ways. They work with different data, different risks and different output. A basic layer can be the same for everyone, but behaviour change calls for recognisable cases per role, process or risk level.

And it doesn’t stop there. Behaviour doesn’t only change during a training. Employees need support at the moment they use AI: a short decision tree, a 5-question card, a data checklist or team agreements that make clear what is normal, safe and desirable.

So good AI Literacy is not a one-off knowledge session. It is a learning intervention that translates policy into daily decisions.

What organisations notice

For organisations, that usually means three things.

  • More safety, because employees recognise sensitive situations faster.
  • More consistency, because teams don’t each invent their own AI rules.
  • More business value, because AI doesn’t just deliver faster work, but becomes better integrated into how people think, decide and collaborate.

This also shifts the role of L&D. No longer “we need to provide an AI training”, but “we need to help design safe and useful AI behaviour”.

That is a more fundamental conversation. And a more valuable one for management, Legal, HR and IT.

The real difference

The difference between AI Literacy approaches rarely lies in who can show the most tools or collects the best prompt formulas.

The real difference lies in how you look at it.

At AI as a tool, or at AI as a behavioural question. At policy as a document, or at policy as a daily choice. At training as information transfer, or at learning as a lever for safe, critical and responsible action.

If you approach AI Literacy as prompt training, you mainly get more skilled AI users.

If you approach AI Literacy as behaviour design, you make different choices. Then you start from risks, work context, scenarios, feedback and transfer to practice.

That is exactly where good learning design begins for us.

Checklist: is your AI Literacy approach behaviour-focused enough?

Use this checklist as a quick reality check.

  • Does your AI training start from real work situations?
  • Is it clear which AI tools employees may and may not use?
  • Do employees learn which information they must not enter?
  • Does the training include grey areas, not just easy examples?
  • Is there a distinction between roles, processes or risk levels?
  • Do employees practise concretely how to check AI output?
  • Is it clear when employees need to escalate?
  • Does responsibility remain explicitly with the user?
  • Are job aids available after the training?
  • Do you look at application, not just participation?

If you mostly answer “no”, you probably don’t have an AI Literacy programme yet. You have an awareness moment.

That can be a start. But it is not a mature approach.

FAQ

Is AI Literacy mandatory?

Under the EU AI Act, AI Literacy is included as an obligation for providers and deployers of AI systems. Organisations must take measures to support a sufficient level of AI literacy among employees and other people who work with AI systems on their behalf. How that is put into practice depends on context, role, level of knowledge and risk.

So is prompt training useless?

No. Prompt training can be useful, but only as part of a broader approach. Employees also need to learn when AI use is responsible, which information is sensitive, how to check output and when to escalate.

Does everyone need to follow the same AI Literacy training?

Not necessarily. A basic layer for everyone makes sense. But roles with sensitive data, external communication, HR processes, customer advice or decision-making need additional scenarios.

AI Literacy Design Scan

Curious how to translate your AI policy into safe and responsible behaviour in the workplace?

With an AI Literacy Design Scan, we map which AI use cases are already alive today, where the main behavioural risks lie and how you support employees in making better decisions when using AI.

Share on:

Related blogs