What we measure, how we score it, and what it can’t tell you.
Most AI readiness quizzes ask you to rate yourself, and people are poor judges of their own AI skill. This one also tests what you choose when it comes to it, and asks you to show real work. Here is exactly how it works, including its limits.
Last updated 28 October 2026 · Score version 3 · Take the score
In one paragraph
You answer questions in five areas of AI at work, called CRAFT. Some ask how you work. Others put you in a situation written for your seat and see what you do, or check one fact. At the end you name your tools and show what you have automated. Each area scores 0 to 20, the five combine into 0 to 100, and the result is a tier, a personal AI type and a plan. Wherever we compare what you say with what you do, the report shows you the gap.
See it in a minute
From Ammar’s post on LinkedIn about the score.
What it measures
Five dimensions, scored separately so the report can say what to work on rather than give you one number.
- Communication: how you brief AI, what you hand it and how.
- Readiness: what you do when AI gets it wrong, and how you keep learning.
- Adoption: how far AI has got into your actual week, not just your browser.
- Fluency: which tools you reach for, and whether they fit the job.
- Trust: when you check AI output, and when you let it run.
If you lead people, there is a second number, Leading AI (0 to 100): how ready the people you lead are to follow you into this, based on what you ask of them and what you show them yourself. It is kept apart from your CRAFT score, which measures you.
Four kinds of question
How you work
Five-step ladders, from “I haven’t done this” to “I do this systematically and help others do it”. The steps score 0, 2, 9, 15 and 20. These are self-ratings, and we treat them as that.
What would you do?
A situation written for your seat (leading, managing, or doing the work), with five answers. Every answer is one of five fixed types: the best response (20), thorough but slow (10), shallow (6), overreach (4) and avoidance (2). When we rewrite a scenario from your own description of your job, the model writes the words and the types and points stay fixed. A model never decides your score on these.
Spot checks
One right answer, and the reason is explained as soon as you pick. They test what you know, not how you rate yourself.
Show your work
At the end: name the AI tools you actually use (required), say whether you have automated anything, and, if you have, describe it and optionally upload a screenshot. A model reads the evidence against a written rubric and scores it 0 to 20. If you have not automated anything yet, you design a first automation for a task from your own week, which is scored step by step against a fixed answer key. On the Full and Deep versions there is also the case: a real decision in a business like yours, where you filter eight ideas, choose safeguards and build a six-step automation.
A few questions are written from your own answers (“follow-ups”). They probe the places where your answers disagree or where one answer could mean two things.
How it adapts to you
- Your seat: eight levels, from founder to individual contributor. Leaders get Leading AI and leadership versions of the scenarios.
- Your function and industry: fifteen of each. They change the wording, the examples and the case.
- Your own words: one question asks what your work actually involves. It is used to rewrite a few questions and scenarios around your real week, with the scoring rules above unchanged.
- Your company: on a company edition, the organiser gives us the company website and the case is set in that business.
- Adaptive follow-ups: an answer in the ambiguous middle of a dimension brings a second question on it, so a dimension is never decided by one tap.
Three depths: Quick (about 25 questions, 6 minutes), Full (about 45, 15 minutes) and Deep (about 55, 24 minutes). Leaders get a few more.
How it is scored
Each dimension scores 0 to 20 from the questions you were actually asked, so a dimension that got two questions is not compared against one that got five. The five combine into 0 to 100 with these weights:
| Dimension | Weight | Why |
|---|---|---|
| Readiness | 1.5 | How someone reacts to a bad output predicts progress better than any tool knowledge. |
| Adoption | 1.3 | It is where the time actually comes back. |
| Trust | 1.2 | It is where the risk sits: unchecked output in front of a client. |
| Communication | 1.0 | The base every other skill sits on. |
| Fluency | 1.0 | The most learnable dimension: a map, not a talent. |
These weights are a judgement from running training programmes, not a statistical fit. We say so because it matters.
Tiers
80 and over is AI Multiplier, 60 to 79 AI Integrator, 40 to 59 AI Practitioner, 20 to 39 AI Curious and under 20 AI Bystander. Once 100 people have finished, the boundaries are set from the real spread of results instead. The report leads with the tier and its band rather than the exact number, because with five-step answers the number is less precise than it looks.
Comparisons with other people
A percentile only appears once 30 people have finished (or 15 in your seat or role). Before that you see no percentile at all, rather than an invented one.
What you say against what you do
This is the part most assessments cannot show you. Studies keep finding that how good people say they are with AI barely tracks how good they are, and that people using AI tend to overestimate how well they did (a 2026 preprint meta-analysis, Aalto University). So the score keeps the two apart:
- Say vs do: for each dimension, your self-ratings and your scenario and spot-check answers are scored separately. A gap of 6 points or more is named in your report.
- The mirror: just before your result you guess your own tier, and the report shows how close you were.
- Confidence: every result carries High, Medium or Low, from how many answers you gave, how many behaviour checks you took and how much real work you showed. A Quick run cannot reach High; showing your work is what raises it.
Your AI type
The tier says how far along you are. The type says how you use AI. It comes from the shape of your profile, never from a model:
- The Briefer: Communication is your strongest dimension.
- The Experimenter: Readiness is your strongest.
- The Operator: Adoption is your strongest.
- The Scout: Fluency is your strongest.
- The Editor: Trust is your strongest.
- The Builder: you showed an automation that runs today, and it scored 13 or more out of 20.
- The Starter: you are at the very beginning; the type names your first step instead.
It comes with one of three notes from the say-vs-do comparison and the mirror: ahead of your habits (you rate yourself above what you chose), quietly capable (you chose better than you rated yourself), or well calibrated.
What it cannot tell you
- It is a self-assessment with behaviour checks, not an exam. It does not certify skill in any one tool.
- Scenario answers can be gamed by someone who knows what a good answer looks like, or who asks a chatbot. Real evidence is harder to fake, which is why it raises confidence.
- Model-read evidence can be wrong. Screenshots and written answers are judged against a written rubric, and the judgement is shown to you, but a model can misread an image.
- We have not yet published retest or validity statistics. We will once there are enough completions to make them meaningful, and this page will carry them.
- A result is a snapshot. AI work changes fast, so results carry a re-check date six months out.
Sharing and credentials
Your result page has a share card built around your AI type. It shows the shape of your CRAFT profile, not your score. If you verified your email, you can also turn the result into a credential: a page on mycocoon.life with your first name, AI type, tier, CRAFT shape, issue date and re-check date, which you can add to LinkedIn under Licences & certifications. Credential pages are not listed in search engines.
Team and company results
When an employer runs the score for a team, the organiser sees the group first: averages, the spread, results by seat and by function (only for groups of three or more, so nobody can be singled out from a group), and the tools the room uses. Named results sit behind a separate button, for coaching conversations rather than ranking. Individual team rows are kept for seven days and then folded into the team’s overall figures. Private audits keep results for an agreed period, then delete them.
Your data
We keep your answers and your result so your report is there when you come back, and so Cocoon can see what people need to learn. Getting the report signs you up for one weekly AI tip, and every email has an unsubscribe link. The Privacy Policy has the detail, including how to have everything deleted.
Questions about the method
If something here looks wrong, or you want to use the score for research, write to [email protected] or [email protected].