What AI Literacy Actually Means for Your Team

The European Union’s AI Act made AI literacy an obligation from 2 February 2025, alongside its prohibitions on certain practices. Organisations covered by it are expected to ensure that people dealing with these systems have a sufficient understanding of them.

For a business in Quebec or the United States that is not directly binding unless your systems reach the European market. The underlying problem is entirely real regardless, and most companies have it.

Your team is already using it

Somebody is drafting client emails with it. Somebody is summarising a call. Somebody pasted a contract into a chat window last week to get help understanding a clause.

None of them were trained. None of them were told what is acceptable. They were being resourceful with a tool that appeared in their life, and the business found out later or has not found out yet.

That is the actual state of adoption in most companies. Not a strategy, a drift, and the risk sits entirely in what nobody explained.

What people get wrong, specifically

Four misunderstandings account for most of the damage, and none of them are stupid.

Assuming it knows things it does not. A system answering confidently about your pricing, your contract terms or your process is generating something plausible unless it has been given that material. Confidence is not evidence, and it reads exactly the same either way.

Not knowing where the information goes. Most people have no idea whether what they paste is retained, whether it might be used to improve a model, or which jurisdiction it lands in. They are not being careless. Nobody told them there was a question.

Treating output as finished. A draft is a draft. The failure is not that a system produced something imperfect, it is that somebody sent it without reading it properly because it looked polished.

Not recognising when to stop. The moment a task moves from drafting into deciding something that affects a client is the moment a person has to be involved, and that boundary is invisible unless somebody draws it.

What useful training looks like

Not a course. Half an hour and a page.

Show them what it is good at, with examples from your actual work. Show them a confident wrong answer, which is the single most valuable thing you can demonstrate. Tell them which tools are approved and for what. Tell them what must never be pasted anywhere, named specifically rather than described as sensitive information. And tell them who to ask.

People retain a demonstration of failure far better than a list of rules. Watching a system state something incorrect about your own business, fluently, changes how somebody treats it permanently.

Write the one page policy

A business of this size does not need a framework. It needs one page somebody will actually read.

Which tools are approved and for what. Which categories of information never leave your systems. What has to be reviewed by a person before it reaches a client. Who to ask when something falls outside the list.

A twelve page document produced by a lawyer will be filed and ignored, and the behaviour you were worried about continues exactly as before.

Give them something better than what they found

Whatever the policy says, assume some of this is already happening outside it.

People use what helps them finish the day. If the approved route is slower or more awkward than the one they discovered themselves, they will keep using theirs, on a personal account, where you have no visibility.

Which makes the sanctioned option a design problem rather than a policy problem. A system built on your own material, that gives better answers about your business than a general tool can, wins on merit. Businesses that ban without replacing end up with the same activity and less oversight than before.

Where the literacy question is really settled

The most reliable way to make a team safe with these tools is not training. It is building systems where the dangerous options are not available.

A system that answers only from approved material cannot invent your pricing. One with permissions carried through cannot surface something the person asking should not see. One with a review threshold cannot send a commitment without somebody approving it.

Training closes the gap between what people know and what they can reach. Design closes the gap by moving what they can reach. The second is more reliable, because it does not depend on somebody remembering a rule at four on a Friday.

What to do this month

• Ask what they are already using. Not as an audit with consequences. The list will be longer than expected and it is your real picture.

• Run the thirty minute session, including one live demonstration of a confident wrong answer about your own business.

• Write the page. Plain language, given to everybody.

• Provide the better option, because a rule without an alternative is a rule people route around.

The client facing question your team will be asked

One thing worth preparing for, because it is coming and most teams have no answer ready.

A client asks whether their work was done with AI. That question is asked with genuine curiosity roughly as often as with suspicion, and an evasive answer converts the first into the second immediately.

Decide the position and give everybody the same words. Something honest and unremarkable: we use it for drafting and research, a person reviews everything that reaches you, and here is what we never put into it. Delivered plainly, that answer usually ends the conversation and occasionally starts a better one.

What damages you is three people in your business giving three different answers, which is what happens when nobody decided.

Refresh it, because the tools move

A session run once becomes out of date faster than most training, because the capabilities and the interfaces genuinely change.

Twenty minutes twice a year is enough. What changed, what is now possible that was not, and what mistakes have shown up internally since last time. That last item is the most valuable and it costs nothing to collect if somebody is paying attention.

Sources

• European Commission, Regulatory framework for AI. The AI Act entered into force 1 August 2024; prohibitions and AI literacy obligations applicable from 2 February 2025.

AI Optimize builds systems trained on your own material, with permissions and review boundaries designed in, so the safe option is also the useful one. That work sits under Custom AI Integrations.