BACKGROUND
This project was an internal quality tool for a content team producing AI-assisted content. The goal was to give editors a structured framework for deciding whether AI-generated drafts were ready to publish, with clear thresholds that distinguished between items requiring a fix, items requiring a return to the writer, and items that should block publication entirely.
I built the framework across four evaluation dimensions in order of stakes: accuracy first because accuracy failures can cause active harm, followed by tone, originality, and brand fit. Each dimension has specific criteria, how-to-check instructions, and pass, warn, or fail thresholds, with a decision guide at the top that mapped any combination of results to a clear action. The framework closes with a named-reviewer scorecard, making accountability explicit rather than diffuse.
The result was a single document that could be used consistently across editors and content types, with the rationale for the ordering and thresholds built in so the framework could be applied with judgment rather than just ticked through.
