One Engine Is an Opinion. Nine Engines Are a Market.

Run the same buyer question across engines and you will see it immediately: one recommends you, another names three competitors, a third describes the category without naming anyone. This is not noise — each engine retrieves from different sources, weights them differently, and answers with a different disposition.

That disagreement is information. It tells you that the sampled outputs differ; citations can show which pages appeared alongside each output, without proving why the engines disagreed.

Different engines can expose different citation mixes — Reddit, listicles, vendor docs — and can recommend different products on the same question. If your buyers use Perplexity and you only check ChatGPT, your sample covers one engine, not the whole observed set.

Breadth without replication is just more screenshots. The method that holds up is both: multiple engines, sampled on disclosed cadences and aggregated over the week — so a brand that appears in one engine once does not get confused with a brand repeatedly recommended in the observed sample.

That is the standard we hold our own report to: provenance-separated answer paths, only substantial answers allowed into verdict claims, coverage disclosed, and every answer movement traceable to a sampled run. The exact engine set depends on plan, keys, and current registry.