Iro AIIro AI Blog
We've published the results of our free AI IQ test since the first 50 completions. This refresh is the cleanest cut yet: 196 people, one first attempt each, internal test runs removed, April 1 to October 8, 2026. The median is 7 out of 10. And where people came from moves their score more than almost anything else, which turns out to be the more interesting finding.
Across 196 people who took our free AI IQ test between April 1 and October 8, 2026, the median score is 7 out of 10, the average is 6.91, 33% scored 9 or higher (about 1 in 3), and only 14% got a perfect 10 (about 1 in 7). Scores span the full range from 0 to 10.
The top is genuinely hard to reach: a third of people score 9 or higher, and only about 1 in 7 post a perfect 10. The bottom is still very much occupied: about 4 in 10 people land at 6 or below, and 14% score 3 or below.
Median 7. A third of people score 9 or higher. Only 1 in 7 get a perfect score.
When we published the first cut of this data, the average across 50 completions was 5.2. A later cut that counted every completion, retakes and our own test runs included, came out at 7.4. This version is stricter, and it lands in between at 6.91. Nobody's AI skills swing that much at population scale, so what is moving?
The audience. Where someone arrives from predicts their score better than almost anything else in this data. Visitors who came to the quiz from Reddit averaged 8.4. Visitors who came from Google averaged 6.1. A plausible reason: Reddit visitors tend to arrive from AI-focused communities, while search visitors are closer to cold traffic.
Retakes and test runs used to count. Earlier cuts included repeat attempts, and people who retake tend to be chasing a better number. They also included a handful of perfect 10s from our founder testing his own quiz. This time each person counts once, by their first attempt, and internal test runs are out.
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.
Practice this, don't just read it.
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).
Every score from 0 to 10, counting each person's first attempt (n=196, April 1 to October 8, 2026):
| Score | People | Share |
|---|---|---|
| 0 | 3 | 1.5% |
| 1 | 3 | 1.5% |
| 2 | 1 | 0.5% |
| 3 | 20 | 10.2% |
| 4 | 9 | 4.6% |
| 5 | 15 | 7.7% |
| 6 | 27 | 13.8% |
| 7 | 21 | 10.7% |
| 8 | 32 | 16.3% |
| 9 | 37 | 18.9% |
| 10 | 28 | 14.3% |
And the summary:
| Stat | Value |
|---|---|
| People | 196 |
| Average | 6.91 / 10 |
| Median | 7 |
| Middle 50% of scores | 5 to 9 |
| Scored 9 or higher | 33% (about 1 in 3) |
| Perfect 10 | 14% (about 1 in 7) |
| Range | 0 to 10 |
Method: one first attempt per person, internal test runs excluded, self-selected sample, web only.
The shape is lopsided. The single most common score is 9, and 8 to 10 together account for nearly half of everyone. But there is a second cluster at 3, a reminder that a real group of people are still largely guessing.
Two things are true at once in this data, and both survive every cut we have made of it.
AI fluency is learnable and plenty of people have it. A median of 7, with a third of people at 9 or higher, 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. About 4 in 10 people land at 6 or below and 14% score 3 or below. 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 early cold-traffic cohort, and it shows up again in the gap between Reddit and Google visitors.
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 7, 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.
Practice this, don't just read it.
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 196 people who took the Iro AI IQ test between April 1 and October 8, 2026 (first attempt each, internal test runs excluded), the median score is 7 out of 10 and the average is 6.91. 33% scored 9 or higher, and only 14% (about 1 in 7) got a perfect 10. Scores depend heavily on who takes the test: Reddit visitors averaged 8.4, Google visitors 6.1.
It is exactly the median 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.
Mostly because the audience changed, not the population. The first 50 completions averaged 5.2, a later cut that counted retakes and internal test runs averaged 7.4, and the current first-attempt data averages 6.91. Traffic source alone moves the average from 6.1 (Google visitors) to 8.4 (Reddit visitors). With self-selected data, who shows up moves the number more than anything else.
From 196 people who completed the Iro AI IQ test between April 1 and October 8, 2026. Each person counts once, by their first attempt, and internal test runs are excluded. The test is 10 questions scored 0 to 10. It is self-selected, web-only data and 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.