Five questions, a point for each yes. They take about ninety seconds and most people score lower than they expect, including people who open one of these tools every working day. What the number is worth is set out further down, and the short version is that it is a good way to start a conversation and not a measurement of anything.
The answer, in one line
Five yes-or-no questions, a point for each yes. Have you set up a Project, a Gem or saved instructions so it knows who you are without being told.
The five questions#
1. Have you ever set up a Project, a Gem or saved instructions, so it knows who you are without being told?
2. Have you ever run a deep research task, waited twenty minutes, and read what came back?
3. Have you ever asked it to argue against you, and then changed your mind because it was right?
4. Do you know, specifically, what your own organisation allows? Not what you assume. What the policy says.
5. Have you ever built anything with it that you still use?
Reading the score#
Zero or one. You are most of any room, including people who use these tools every single day. Frequency of use and capability with them turn out to be close to unrelated, which is the finding that makes this worth asking at all. The next step is the second rung rather than more reading about it.
Two or three. The habit exists and the gap is consistency rather than knowledge. People in this band usually know what they should be doing and do it when the week is calm.
Four or five. The tool questions are settled, and nothing further about features will help. What is left is judgement: what to delegate, what to keep doing yourself, and how you would notice if your own capability were slipping while your output stayed good.
The bands were written for a room of students and have been softened here for a general reader, which is a change of address rather than of content. The original wording put the lowest band at "nothing to be embarrassed about", and that instruction survives: the point of asking is not to sort people.
What each question is actually testing#
One and five are about rungs. Saved instructions is the second of five rungs of use, and building something you still use is the third or above. Most people sit on the first and do not know there are five. That ladder is set out at the five rungs of AI use.
Two is about patience, the one people fail most often. Waiting twenty minutes and then reading the whole thing is a different activity from asking a question and skimming an answer, and the second one is what nearly everybody does. It is also where the free-tier limit does its quiet damage, described at running out of messages makes you worse.
Three is the one that matters, the only one here you cannot acquire in an afternoon. Setting up a Project is a feature. Asking a system to argue against you and then changing your mind requires having held a position first and being willing to lose it. Four of these five questions measure fluency. This one measures whether judgement is still in the loop, and the practice behind it is at how do I get AI to challenge me.
Four is the one people are most confident and most wrong about. Almost everybody believes they know their organisation's position. Very few have read it. The gap between the assumed rule and the written one is where most trouble starts. That is covered at using AI when your department bans it.
What this score is not#
It is not validated, and the distance between this and a validated instrument is worth being precise about. Nothing tests whether the five questions measure one underlying thing. Nothing establishes that the bands separate people meaningfully, or that the boundaries fall in the right places. Above all, nothing shows the score predicts any outcome: not performance, not judgement quality, not whether somebody's capability holds up when the tool is removed.
What can be said is narrower. It has been put to live audiences, it sorts a room in a way the room recognises, and it moves a session from "how do I use this" to the question worth an hour. That is a claim about a conversation rather than about a person.
This estate publishes an evidence base so that the difference between a measurement and a good question stays visible, and an instrument with a number attached is the easiest place in the world to lose it. A five-point score looks like data. It is five yes-or-no questions somebody found useful on stage.
The organisational version, and why it is a different instrument#
Scoring individuals tells an organisation almost nothing, because the thing an organisation needs to know is not how fluent its people are but whether it could still do its work unaided. Those come apart: fluency is what this page measures, and retained capability is what a capability audit is for.
Counting tool adoption is the third thing, and the weakest of the three. A rising adoption number is compatible with every person in the organisation sitting on rung one. That is the argument of usage theatre and how do you measure AI adoption properly.
If you scored low#
Do question one this week. It takes about ten minutes and it is the single change that moves the most, because everything after it stops beginning from nothing. Then question two, once, on something you actually care about, so the waiting has a point.
Question three is the one to keep returning to, and it does not get easier with practice. It requires a position to defend, which is the habit set out at Human at the Start, and the willingness to be argued out of it.
Key sources
- Hirji, R. (2026). Mastering AI, teaching deck, September 2026, slide 89. The five questions and the banded reading, delivered to live audiences before being published here.
- Hirji, R. (2026). SuperSkills: The Seven Human Skills for the Age of AI. Kogan Page.
Related SuperSkills research#
The ladder behind questions one and five is the five rungs of AI use. The organisational instrument is the capability audit, and the individual one is am I becoming dependent on AI. On why a high score is not the same as a safe position, automation bias and capability debt. The plain starting point for anyone scoring zero is how to use AI at work.
About this research#
Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. The five questions and the bands are his, from the September 2026 teaching deck, and no earlier dated publication of them exists, so they anchor to the deck with no claim of first use. The instrument is unvalidated and this page says so in its own section rather than in a note.
Essay · SS-2026-215
Hirji, R. (2026). How good are you at using AI?. The SuperSkills evidence base, SS-2026-215. https://thesuperskills.com/research/how-good-are-you-at-using-ai. Last reviewed 11 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work