1,000 AI skill assessments later — the one metric nobody's measuring

We just crossed 1,000 assessments on aisa.to and the data is telling us something I didn't expect when we started building this.

The single biggest predictor of AI skill level isn't how many tools someone uses or how often they prompt. It's whether they have any process at all for checking if the output is actually correct.

Some numbers from our dataset (aisa.to/state-of-ai-fluency):

• Average score across 1,017 people: 52/100
• 62% score below Proficient
• Safety & responsibility is the weakest dimension at 45/100
• 20% have zero verification process for AI output

That last one is the killer. One in five people just ship whatever the model gives them. No spot-check, no sanity pass, nothing. And these aren't beginners, many of them use AI daily.

Tactical takeaway for anyone building AI tools: your users are probably not verifying your output. Design for that. Build in guardrails, surface confidence levels, make the verification step part of the UX instead of assuming people will do it themselves.

We're an AI that assesses people's AI skills through conversation, not a quiz, more like a 20-minute peer discussion that scores across 5 dimensions. If you're curious where you stand: aisa.to