Iro AIIro AI Blog
When we published the first 50 results of our free AI IQ test, the average score was 5.2 out of 10. Ninety-seven completions in, it's 7.4. People didn't get smarter in ten weeks. Who takes the test changed, and that turns out to be the more interesting finding.
Across 97 completions of our free AI IQ test, from 90 people over four and a half months, the average score is 7.4 out of 10 and the median is 8. Scores span the full range from 1 to 10, with a standard deviation of 2.3.
The top is genuinely hard to reach: only about a quarter of people score 9 or higher, and about 1 in 10 post a perfect 10. The bottom is still very much occupied: about 1 in 4 people land at 6 or below, and 10% score below 4.
Median 8. A quarter of people score 9 or higher. A quarter land at 6 or below.
When we published the first cut of this data, the average across 50 completions was 5.2 and the median was 5. Ten weeks later the average is 7.4. Nobody improves that fast at population scale, so what happened?
The audience changed. The early completions came from cold traffic: people landing on the site from search with no particular AI background. The later completions increasingly come from people who found us through AI content, follow AI accounts, or read this blog. They were already the 8-out-of-10 crowd before they clicked.
Retakes count. 97 completions from 90 people means some people took it more than once, and people who retake tend to be chasing a better number.
And full disclosure: a few of the perfect 10s are our founder testing his own quiz. We can do the math on that instead of hiding it: strip three founder runs out and the average moves from 7.4 to about 7.3, and the median does not move at all. We left them in because we count everything, and because the honest version of a small dataset is more useful than a polished one.
This is the quiet lesson of the whole post: with self-selected data, who shows up moves the number far more than how good people are. Any published statistic about "the average person's AI skills", ours included, is really a statistic about who took that particular test.
Iro AI turns ideas like the ones in this post into 5-minute exercises with feedback. Free tier, Pro from $4.17 a month ($49.99 a year, 7-day free trial).
The shape of all 97 completions, as of August 2026:
| Stat | Value |
|---|---|
| Average | 7.4 / 10 |
| Median | 8 |
| Middle 50% of scores | 6 to 9 |
| 10th percentile | 3.6 |
| 90th percentile | 10 |
| Range | 1 to 10 |
| Standard deviation | 2.3 |
For comparison, the first 50 completions averaged 5.2 with a median of 5, and 56% of that cohort scored 5 or below. In the current data the middle has moved up two full points while the low tail never went away.
Two things are true at once in this data, and both survive the cohort shift.
AI fluency is learnable and plenty of people have it. A median of 8 among people who engage with AI content shows the skills are absorbable: prompting, tool choice, knowing when a model is likely to hallucinate. These are not talent, they are reps.
The spread refuses to close. Even in this warmer audience, a quarter of people land at 6 or below and one in ten scores below 4. The distance between someone who scores 9 and someone who scores 3 is not trivia knowledge, it is the difference between reliably getting trustworthy output from AI and pasting a vague question into a chat box. That divide showed up in the cold-traffic cohort at 5.2 and it shows up in the warm one at 7.4.
If AI skills are becoming a workplace dividing line, this is what it looks like in miniature: the people who practise pull ahead, and the gap widens rather than narrows.
What this is and is not:
We'll keep refreshing these numbers as the sample grows, and we'll keep showing the previous cohorts when we do.
The fastest way to see your own AI literacy is to take the test: free, about two minutes, no signup, and you get a per-topic skill profile with the area you're weakest on.
The median is 8, so that's the bar. If you land under it, that's not a verdict, it's a starting point. Iro is built to close exactly these gaps: short, gamified daily lessons that take you from "I've used ChatGPT" to genuinely fluent across ChatGPT, Claude, Gemini, Perplexity, prompting, and more, in about five minutes a day.
Iro AI turns ideas like the ones in this post into 5-minute exercises with feedback. Free tier, Pro from $4.17 a month ($49.99 a year, 7-day free trial).
Across 97 completions of the Iro AI IQ test as of August 2026, the average score is 7.4 out of 10 and the median is 8. Only about a quarter of people score 9 or higher. The first 50 completions averaged 5.2, and the difference reflects who takes the test rather than people improving.
It is just under the median of 8 in our current data, so it is a solid middle score with clear room to grow. The gaps at that level are usually judgment skills: knowing when a model is likely to hallucinate, choosing the right tool, and writing prompts that constrain the output.
The audience changed, not the population. Early quiz takers arrived as cold search traffic; later takers increasingly come from AI-focused content and were already skilled. Retakes also count as separate completions. With self-selected data, who shows up moves the number more than anything else.
From 97 completed Iro AI IQ tests by 90 unique people between March and August 2026. The test is 10 questions scored 0 to 10. It is self-selected data with retakes included, disclosed contamination quantified, and it is a snapshot rather than a representative study.
Take the free Iro AI IQ test to see your score, a per-topic skill profile, and your weakest area, then close the gaps with short daily lessons in the Iro app. It is designed to move you from casual AI use to genuine fluency in about five minutes a day.