Six-check manual audit can verify if AI visibility scores are reproducible

A tutorial published on DEV Community outlines a six-field framework for auditing AI visibility scores, arguing that such scores are only meaningful when the underlying prompt, engine, raw answer, date, and counting rule are all disclosed. The core claim is that an AI visibility score is not a fixed brand property but a test result tied to a specific question set, engine panel, and run date — making it inherently non-reproducible without those inputs. The method involves selecting three buyer-style questions, running them across two AI surfaces, and recording results in six cells without attempting to force consistency between tools. A real example from Webappski's June 14, 2026 tracker report is cited, showing the brand appeared in 2 of 39 answer cells, with one named result from Claude and none from ChatGPT or Gemini. The tutorial warns that scores from different tools are not directly comparable since they measure different question-and-engine grids, and recommends expressing results explicitly — such as 'present in two of six answers on this date' — rather than as a standalone percentage.