Open reference for AI fluency

Not how much you use AI. How well.

AI fluency: knowing what to hand to AI, how to ask, whether to trust what comes back, and what you answer for. This site describes it in four competencies and four levels, and lets you place yourself in ten minutes.

Take the test Understand it in 10 minutes

Works with Copilot, ChatGPT, Claude or any other tool.

The idea in 60 seconds

The 4D framework

The prompt is one of four competencies

The 4D framework splits working with AI into four competencies, in the order they come up. The prompt is only the second.

Using AI well, the “how well” of the title, takes all four. A conversation shows only part of it: of the framework’s 24 indicators, 11 can be seen there.

Animation: the framework’s 24 indicators; 11 move into a conversation window, the other 13 stay outside.

Sources: AI Fluency Framework · Anthropic, The AI Fluency Index, February 2026

How they fit

Before, during, after.

Delegation, Description, Discernment, Diligence: two nested loops LONG LOOP · AROUND THE CONVERSATION BEFORE Delegation AFTER Diligence DURING · SHORT LOOP Description Discernment describe · judge · describe again what you answer for changes what you delegate next time Quality is decided in the short loop. Risk is carried by the long one.

The four levels

One request, at four levels

Read the four levels as text

Using AI more does not raise your level: you can use it all day and stay at level 1. You become a Champion by writing down a way of working that others reuse, whatever your job.

Why measure it

When the result looks finished, people check it less.

In early 2026 Anthropic observed eleven of these indicators across 9,830 real conversations with an assistant. When the answer is a finished document, people give more instructions and check less.

Source: Anthropic, The AI Fluency Index, February 2026. Out of 100 conversations. All eleven indicators →

Where to start

10 minPlace yourself

Eight gestures from your job

You act, then say what you did last time. Nothing is stored.

Take the test → 10 minUnderstand

The four competencies in detail

Twelve behaviours, the loop, and a 48-cell grid with an example for each cell.

Open the framework → DemoSteer a team

Where a team stands

What each role needs, and whether the team is ready for its AI projects. A demo on fictitious data.

See the team view →

What you get

An estimate

No badge, no pass mark. A measurement needs real work read by two people, which a pilot in an organisation provides.

A result that stays yours

This site stores nothing: your result lives in a link you keep.

A profile, with no average

Your level, competency by competency. No average and no ranking: the shape of the profile says what to work on.