The 5 Levels of AI Literacy

I run a short AI literacy check in my training sessions. Fifteen questions, answered on a phone, about five minutes. It is not a knowledge quiz. It asks people what they actually do with AI in their work, and it puts each person on a five level ladder at the end.

Earlier this year, in one session, three people came out as Workflow Builders. Level four of five.

The same room had also filled in a tool bingo card before the check: tick every AI tool you have used. All three of those people had ticked ChatGPT. One had also ticked Copilot. Nothing else. No image tools, no note takers, no research tools, no assistant they had set up themselves. They were working from their phones, in a chat window, one question at a time.

They had not exaggerated on the check. They really do use AI a lot, most days, for real work. But the tool had told them they had rebuilt their work around AI, and they had not. They had a habit. They did not have a workflow.

That was my own tool getting it wrong, and I have since tightened the scoring. I am writing about it anyway, because it is the same mistake almost everyone makes about themselves.

Frequency is not fluency

When people ask themselves "how good am I with AI?", they nearly always answer a different question: "how often do I use AI?"

Frequency is easy to notice. It feels like progress. And it is progress, up to a point. It just stops telling you anything useful surprisingly early.

The June 2026 AI Proficiency Report from Section is the clearest illustration I have seen. They tested 5,026 knowledge workers on live tasks rather than asking them to rate themselves.

• 67% use AI at least once a week.
• 5.5% meet the bar for proficiency.
• In an earlier round, 54% described themselves as proficient and around 10% tested that way.

So weekly use is now normal. Skill is not. The gap between those two numbers is the entire subject of this article.

Using AI every day tells you that you have a habit. It does not tell you what that habit is worth.

AI literacy vs AI fluency: what is the difference?

The two words get used as if they mean the same thing. They do not, and the difference is the reason a five level ladder is useful at all.

• AI literacy is understanding. You know what these tools are, what they are good at, where they fail, and what the rules are for using them at work. You can hold a sensible conversation about AI.
• AI fluency is practice. Your work has changed shape because of AI. You have built things you reuse, you move between more than one kind of tool, and you check what comes out before it leaves your desk.

On the ladder below, literacy covers levels one to three. Fluency starts at level four. Most AI training sold to companies teaches literacy and then measures attendance, which is why so many teams finish the year better informed and working in exactly the same way.

The five levels of AI literacy

This is the ladder I draw for rooms, and the one the check scores against.

The percentages on the drawing are my own rough sense of the rooms I stand in, not research. Section's tested split came out at 20.9% novices, 73.5% experimenters and 5.5% practitioners and experts combined, which is harsher than mine at the top. Different tests draw the lines in different places. What matters is the shape, and every version of it agrees: crowded at the bottom and the middle, very thin at the top.

The wall between level 3 and level 4

Almost everyone I meet who is worried about being behind is at level three. And almost everyone at level three thinks the way up is to use AI more.

It is not. Level three to level four is the only jump on this ladder that is not about volume. It is about whether the work itself has changed shape.

Three questions sort this quickly. Answer them honestly, in your head, right now.

• Name one task in your week that runs differently now. Not faster. Differently.
• Name one thing you have built and use again: a saved prompt, a project with your documents in it, a custom assistant, an automation.
• Name the last time you used AI for something that was not text in a chat window.

Most people get one out of three. That is not a failure, it is a position, and a position is the only thing you can move from.

The half of level four that nobody wants

There is a second condition, and it is the one people skip. Level four and five are not only about doing more with AI. They are about checking it.

In the largest global study on this, run by KPMG and the University of Melbourne across 48,000 people in 47 countries:

• 66% rely on AI output without checking whether it is accurate.
• 56% have made mistakes at work because of AI.
• 57% hide their AI use, or present AI-generated work as their own.

Speed without checking is not level four. It is level three carrying more risk. The reason level five is thin is not that the tools are hard. It is that judgement is slow to build, and quietly, most people have not built it yet.

Where the EU AI Act fits

There is now a legal reason to care about AI literacy, not only a commercial one. Article 4 of the EU AI Act has applied since 2 February 2025. It asks providers and deployers of AI systems to make sure their staff have a sufficient level of AI literacy, taking account of people's training, their role, and the setting they use AI in. Deployer covers almost any employer whose people use third party AI tools at work. The supervision and penalty framework behind the Act becomes enforceable on 2 August 2026.

I am a trainer, not a lawyer, so take the wording to your own advisers. Two practical things follow from it either way:

• Sufficient is measured against the role. A finance team putting client numbers through AI needs more than a marketing team drafting captions.
• You cannot evidence a level you have never measured. A register of who attended a webinar is not the same as knowing where people actually sit.

A five level ladder answers both in a plain way. It gives you a level for each person, a next step for each person, and a record you can show.

If your whole team is stuck at level three

Everything above is written to one person. There is a point where that stops being fair.

If one person in a department is at level three, the next step is theirs to take. If the whole department is at level three, that is not a coincidence and it is not a motivation problem. It is a condition of the place they work in.

Level four costs things an individual usually cannot spend on their own:

• Access to more than one tool, with an approval route that takes days rather than quarters.
• Permission to change how a task is done, not just how fast it is done.
• Time that has not already been sold to a client.

Nobody builds a workflow in software they cannot get approved, on a process they are not allowed to alter.

This is the pattern I see most often in organisations that have bought the licences and run the training and are confused about why nothing moved. The people are not the blocker. The conditions are. I have written separately about what actually decides whether AI training sticks, and it comes down to permission and accountability, both of which sit above the person taking the training.

What to do with your level

Nobody needs to be at level five. Most roles do not require it, and chasing it is a good way to spend a year on tools instead of on work.

What is worth knowing is which rung you are actually on, because the next move is completely different depending on the answer.

• At level two, the answer is repetition. One task, five days running.
• At level three, it is building one thing you reuse, and one tool that is not a chatbot.
• At level four, it is judgement and scope. Bigger pieces of work, and a checking step you could defend.

Being wrong about your own level is expensive, because you spend effort on the wrong step and then conclude that AI does not help you much.

I run this check live in sessions, with the room's results coming up on the screen as people finish. The most useful moment is never the individual score. It is the shape of the room appearing behind it, and everyone realising at the same time that the confident daily users and the quiet ones are closer together than anyone expected.