How It Works
Cross-Model Normalization
How a tool combines results from AI systems that do not answer the same way, and the reason one blended visibility score can mislead. The surfaces differ structurally, not only in score. Measured on 8 August 2026, ChatGPT returned no cited sources at all for definitional questions across nine uncached runs in three unrelated categories, while citing twelve to fifteen domains for the buying question in each of those same categories. Both Google surfaces cited either way. Averaging a surface that cites nobody with one that always cites produces a number no engine would recognise. Normalization worth trusting keeps surface, question type, market and run count separate, and says which of them it held fixed.