Data, analysis, and findings from scoring AI prompts across 7 verticals.
I scored 500 software engineering prompts against an 8-dimension quality rubric. Every single one failed. Average score: 13.3/80. Zero graded C or above. Here's what that looks like and why it matters whether a human types the prompt or an agent generates it.
Why no single evaluation framework captures prompt quality, how PQS reconciles five of them into one score, and what that gives you that single-framework tools cannot.
Mapped 43 AI Actors on Apify Store. 74% use Claude. One in three is an MCP server. Only 5 publish their prompts. The AI input quality problem in concrete form.