Whiteship Research Report
What makes AI models cite one source and ignore another?
A Whiteship research study based on 69,642 AI answers and 1.72 million visited source rows across OpenAI, Google, Anthropic, Grok, and Perplexity.
What this Whiteship study shows
Citation behavior follows a recognizable source hierarchy.
This study covers 69,642 analyzed answer runs and 1.72 million visited source rows across five providers. Metrics are correlation-based and describe observable citation patterns, not hidden causal reasoning inside model ranking systems.
Read the condensed articleQualitative patterns
The strongest source traits are visible before citation even happens.
Brand-owned, documentation, and product pages outperform social and aggregator sources because they read as canonical, attributable, and easier to trust.
Pages whose title and URL slug strongly match the originating query are more likely to convert from visited to cited.
Deep pages with descriptive paths outperform flat category URLs when the model needs a precise fact or recommendation anchor.
OpenAI runs in this Whiteship study retain visited sources much more aggressively than Anthropic, which suggests provider-specific sourcing policies or retrieval behavior.
Brief summary
What this research brief establishes.
This study maps the structural and qualitative patterns behind citation behavior: which pages get visited, which domains actually make it into final answers, and which source types convert from visited to cited most reliably.
Brand-owned, documentation, and product pages outperform social and aggregator sources because they read as canonical, attributable, and easier to trust.
Pages whose title and URL slug strongly match the originating query are more likely to convert from visited to cited.
Deep pages with descriptive paths outperform flat category URLs when the model needs a precise fact or recommendation anchor.
Book a demo
Book a demo today with our team.
See the workflow live with our team and understand how Whiteship turns AI search visibility and source-citation signals into something your team can act on.