2.9 vs 1.0
Engines search a different number of times.
So the Search lens reads each engine's own searches, in the index it uses.
The numbers
Searches an engine ran for one answer, on average. Perplexity and Le Chat do not show theirs.
Answer engines
They search different indexes, cite different sources and name different brands. So Vercatus asks each engine on its own and keeps its figures apart.
none sold separately, never an add-on
put the question with web search on
read from the live results page
The index is where Vercatus looks up the searches an engine ran, to see who ranks there. One not checked is shown as "not checked", never as absent.
| Engine | How it is reached | Index checked | Asked in |
|---|---|---|---|
| ChatGPTSearches the web for each answer | AskedIts app, as a person sees it, with web search on | Bing, in part | Every poll |
| PerplexityFresh pages, many citations | AskedSearches the live web for every answer | Its own index | Every poll |
| GeminiGrounded on Google Search | AskedGrounded on Google Search; a source counts when the answer's text rests on it | Every poll | |
| ClaudeA few searches, few sources | AskedWeb search on; it cites the passages it quotes | Brave Search | Every poll |
| GrokLive web search | AskedLive web search on | Not checked | Every poll |
| Le ChatWeb search, cited references | AskedWeb search on | Not checked | Every poll |
| You.comResearch, many sources | AskedIts research interface, in a full run's audit | Not checked | Full runs only |
| Google AI OverviewsGoogle's own AI summary | ObservedThe Overview on Google's live results page; its references are its citations | Every poll | |
| Google AI ModeGoogle Search, conversational | ObservedThe answer on Google's AI Mode page; its references are its citations | Every poll | |
| Copilot Search (Bing)Bing index with an LLM summary | ObservedThe summary on Bing's live results page; its references are its citations | Bing | Every poll |
Asked engines are put the question with web search on; observed engines cannot be asked, so Vercatus reads the live results page in the question's country and language. A page with no AI answer is recorded as "no AI answer shown" and counts for nobody.
One question, every brand any answer named, and which engines named it. This is why Vercatus never averages engines into one picture.
21 English questions in 3 markets, asked of 6 engines with search on and off, on 1 October 2026. Each finding changes how Vercatus asks.
2.9 vs 1.0
So the Search lens reads each engine's own searches, in the index it uses.
Searches an engine ran for one answer, on average. Perplexity and Le Chat do not show theirs.
30 to 58%
So answers without search are kept out of Answer Share, and read in the Memory lens.
How many of an engine's brands with search off it also names with search on.
4%
So ChatGPT is read through its app, as a person sees it.
On the same 10 questions, the two shared this much of what they named and cited.
Findings, not laws: one day, three markets, a few hundred answers. All six findings and the method
Kept apart everywhere in Vercatus, because an engine that reads your page is not an engine that credits it.
The answer mentions the brand.
The answer links one of the brand's own pages as a source.
The brand is the answer's first or only recommendation.
The engine retrieved the page and the answer does not credit it. Never counted as a citation; it counts in the Sources lens.
Not measured. Engines that answer without their sources through any interface Vercatus can read, such as Meta AI and Siri. For those, Vercatus reports whether your site lets their crawlers in: the precondition, not the citation. Answer engines in the docs
Add your domain. Every plan reads every engine, and shows each one's answers apart.
Cancel any time · Data stored in the EU