How to Improve Your Chances of Being Cited by ChatGPT and Perplexity

This article examines the citation selection mechanism in two AI assistants that generate source citations — ChatGPT with browsing enabled and Perplexity — and identifies what measurable content properties most consistently influence citation inclusion. The characterisation is based on observable behaviour patterns rather than claims about proprietary system internals. All probabilistic estimates are labelled as such.

How ChatGPT and Perplexity Differ in Citation Selection

ChatGPT with browsing and Perplexity are both retrieval-augmented generation systems, but they use retrieval differently. Perplexity performs real-time web search for each query and assembles citations from the search results it returns — its citation selection is tightly coupled to its search ranking decisions. A page that ranks well in Perplexity's search layer is likely to appear as a citation. ChatGPT's browsing behaviour is more selective: it follows a chain of queries to gather information, evaluates retrieved content for relevance and reliability, and tends to select fewer, more authoritative sources per response. The practical implication is that optimising for Perplexity citations has more overlap with traditional search ranking, while optimising for ChatGPT citations requires higher trust signal density and entity authority.

The Citation Selection Factors That Are Measurable

Based on observable citation behaviour, the following factors are consistently associated with higher citation probability in both systems:

• Factual specificity: claims that include specific quantities, dates, named entities, or verifiable comparisons are more citation-worthy than general assertions. "AI Overviews appear on approximately 15–25% of queries according to tracking studies" is more citation-worthy than "AI Overviews appear frequently."
• Source attribution: claims that attribute their evidence to a named source give AI systems a verification pathway. Unattributed claims are harder to cite because they offer no chain of evidence.
• Direct answer structure: content organised around the query form (definitional, procedural, comparative, factual) rather than narrative flow produces chunks that answer the question directly, which AI systems prefer as citation candidates.
• Entity authority: the primary entity behind the content should be clearly identified and externally validated. AI systems are more likely to cite sources from entities they can recognise and verify.
• Recency signalling: content with accurate datePublished and dateModified schema, and with substantive rather than cosmetic updates, signals active maintenance — which increases citation probability for time-sensitive queries.
• Topical specificity: a page that addresses one topic or entity deeply is more likely to be cited for queries on that topic than a page that covers many topics superficially.

What Specifically Reduces Citation Probability

• Entity inconsistency across the domain: if different pages on the same site describe the primary entity differently, AI systems assign lower confidence to entity-related claims from that domain
• Schema misalignment: schema that asserts attributes not present in body text creates a detectable inconsistency that reduces source trust
• Absence of external validation: a domain with no sameAs links to verifiable knowledge sources has no external anchor for its entity claims — AI systems treat it as self-asserting authority rather than validated authority
• Thin factual content: pages that consist primarily of general assertions without specific, verifiable claims rarely appear as citations because there is no specific fact to cite
• Stale content on time-sensitive topics: content on topics where recency matters (technology, policy, market data) with no dateModified signal is increasingly likely to be superseded by more recently maintained sources

The Highest-Impact Single Intervention

Across the factors listed above, the single intervention with the highest expected impact for most content sites is improving factual specificity: replacing general assertions with specific, attributable, verifiable claims. This addresses the citation selection test directly — AI systems need a specific fact to cite — and it is a content improvement that improves the page for human readers simultaneously. A page rewritten to replace vague generalisations with specific data points, named examples, and attributed claims will typically see citation probability improvement without any structural or technical changes.