Assistants walk a page's HTML structure and keep it in chunks, not as one document
PrincipleContent & Social
Assistants walk a page's HTML structure and keep it in chunks, not as one document
Google AI is described as reading a page with a tree-walking algorithm that follows the semantic HTML structure from top to bottom. Assistants then chunk the content into smaller pieces rather than processing it whole, and decide which chunks are worth keeping.
Source
The New SEO Playbook for AI Search (Top GEO Ranking Factors) - Ahrefs, presented by Sam Oh (2025)
Ahrefs; underlying studies published at ahrefs.com/blog/ai-overview-brand-correlation
00:03:11 and 00:03:41, Ahrefs, The New SEO Playbook for AI Search, November 2025
Go to the originalMore from this source
StatBrand mentions correlate with AI Overview visibility about three times more strongly than backlinksStatThe three strongest AI Overview signals are all off-site brand signals, not site metricsPrincipleEvery credible mention of a brand is another training example tying it to a topicPrincipleChase links-to-page for Google AI Overviews and traffic-to-page for ChatGPT and PerplexityPrincipleLong-tail is now a retrieval mechanism rather than a coverage tacticPrincipleWrite sections that stand alone and still connect, rather than FAQ-style fragmentsStatContent cited by AI assistants is about a quarter fresher than what ranks in organic searchPrincipleFreshness is a retrieval signal now, because retrieval only fires on topics the model does not already knowCaseOne content refresh produced over a thousand new AI mentions to a single pageStatOnly one in seven of the most-cited sources is shared across all three AI assistantsPrincipleEach AI assistant has its own source diet, so one visibility strategy cannot serve all threeCaseThe only sources cited by all three assistants are Wikipedia, Google, Amazon, Apple, Microsoft and two Brazilian portals
We wrote about this