Why ChatGPT, Gemini, and Perplexity Choose Some Brands and Ignore Others
The source selection behaviour of major AI assistants is not random or entirely opaque. Observable patterns in which brands get cited and which do not reveal specific content and entity properties that influence selection. This article examines those patterns.
This article examines observable differences in how ChatGPT (with browsing), Gemini, and Perplexity select sources for generated responses — what patterns are consistent across systems, where they diverge, and what content properties are associated with higher citation inclusion rates across all three.
The Common Selection Logic Across Systems
Despite having different architectures and training approaches, all three systems share a common citation selection structure: they retrieve candidate content from an index or real-time search, evaluate candidate chunks against the query intent, apply trust signal filters at the source level, and select the subset of chunks that satisfy both the query and the trust threshold. The shared structure means certain content properties influence citation probability across all three: factual specificity, entity clarity, structural accessibility, and source authority signals all affect citation selection in all three systems, albeit with different weights.
Where the Systems Diverge
The systems diverge primarily in how they weight recency and in the depth of their real-time retrieval. Perplexity performs fresh real-time search for every query and cites sources from those search results — recency and current crawl accessibility are high-weight factors. ChatGPT with browsing combines training knowledge with targeted browsing and tends toward fewer, higher-authority citations per response. Gemini integrates heavily with Google's Knowledge Graph, meaning entities that are well-represented in Google's entity index receive higher citation probability for entity-related queries.
The Entity Recognition Factor
Across all three systems, entity recognition has a consistent effect on citation probability: sources whose primary entities are clearly recognised and well-defined in the relevant AI system's knowledge infrastructure receive systematically higher citation rates than sources with ambiguous or poorly defined entities. For Gemini, this means entities with strong Google Knowledge Graph entries. For ChatGPT, entities that appear frequently and consistently in training data. For Perplexity, entities that are clearly disambiguated in the content retrieved during real-time search. The mechanism is the same across all three: the system is more confident attributing information to a source when it can confidently identify what entity the source represents.
●The highest-consistency finding across all three systems is that contradictory entity descriptions — different descriptions of the same entity across different pages or between schema and body text — reduce citation probability. The systems interpret inconsistency as a reliability signal. Consistent entity description is the single most reliable citation probability investment.
Building Cross-System Citation Presence
Building citation presence across all three systems requires addressing the union of their signal sets rather than optimising for any single system. This means: recency maintenance and crawl accessibility for Perplexity, entity clarity and factual density for ChatGPT, and Knowledge Graph integration and entity schema accuracy for Gemini. The intersection of all three — clear entity definition, consistent cross-page entity description, factual specificity with attribution, and regular content maintenance — is the baseline investment that improves citation probability across all three simultaneously.