1,000+ AI skill assessments later — the skill everyone obsesses over is actually the easiest

We crossed 1,000 AI skills assessments on aisa.to last month. Here's the thing that surprised me most.

Everyone worries about prompting. "Am I writing good enough prompts?" It's the first question in almost every assessment conversation. And honestly, most people are fine at it. Prompting scored highest across the board.

The skill that actually separates competent from excellent? Verification. How you check whether AI output is correct before you use it.

The average score across all 1,017 assessments was 52/100. But when we broke it down by dimension, Safety and critical evaluation scored lowest at 45/100. Meanwhile 62% of people scored below what we'd call Proficient.

The pattern: people who are great at getting output from AI are often terrible at evaluating whether that output is actually good. They've optimised the input side and completely neglected the output side.

We published the full breakdown here: aisa.to/state-of-ai-fluency

Curious if other builders are seeing this too. If you're building AI tools — are your users checking output, or just shipping whatever the model gives them?