Engines do not share a fan-out. Each one rewrites, retrieves, and cites on its own logic, which is why a single content plan produces uneven citation across surfaces. These are the behavior profiles we build libraries against.
An engine retrieval profile is a documented model of how one AI engine rewrites a user question, which sources it prefers, and what document properties raise a passage's chance of being cited. Profiles are the input that makes a subquery library engine-specific rather than generic.
An average across seven engines describes none of them.
Coverage decisions are made per engine, because the fragment that wins a Perplexity citation is often a fragment Copilot never issues.
ChatGPT issues search strings that share roughly 13% of their vocabulary with the prompt, and about 91% of those strings change between runs of the same question. It favors recent, densely-structured pages and will quote a single passage rather than summarize a whole document.
Write standalone passages that name the entity inside the sentence. A block that depends on the heading above it gets dropped at extraction.
Gemini resolves the entity first and retrieves against that resolved understanding, which means it is unusually sensitive to whether your brand is consistently described across the open web. Contradictory descriptions suppress inclusion more than thin content does.
Publish one canonical entity summary and repeat it verbatim across owned surfaces, directories, and profiles.
Google states outright that AI Mode issues multiple related searches per question across subtopics and data sources. Overviews lean heavily on pages that already rank for the fragment, so classical ranking still functions as an admission ticket to the fragment space.
Cover the fragment with a page that can rank conventionally, then engineer the passage for lift.
Perplexity exposes its sources and rewards documents that read like references: dated claims, named numbers, explicit comparisons. It cites more sources per answer than any other major engine, which makes it the most winnable surface for a challenger brand.
Ship comparison tables, dated benchmarks, and methodology notes. Perplexity cites what looks checkable.
Copilot stays closer to the original wording than ChatGPT does and draws on the Bing index, so Bing coverage gaps translate directly into citation gaps. Enterprise and procurement phrasing survives its rewrites unusually well.
Verify Bing indexation, and write to procurement vocabulary: security, compliance, total cost, migration.
Claude retrieves less often and cites fewer sources, preferring documents with visible reasoning and explicit caveats. Marketing superlatives are actively counterproductive; qualified statements are treated as more usable.
State limits and conditions in the same block as the claim. Hedged, specific writing outperforms confident copy.
Grok weights live discussion far more than the other engines, so practitioner conversation about your category moves its answers within days. Owned content alone rarely moves it.
Seed and participate in real practitioner threads. This is the one engine where placement beats publishing.
See your own category fanned out across these engines.