Click a citation inside Google’s AI Mode and watch what happens. You do not land at the standard header of the page. Instead, the URL carries a text-fragment directive ending in #:~:text=…, prompting your browser to automatically scroll down and jump directly to a passage highlighted in purple. Google didn’t cite the page as a whole. It cited roughly 117 words extracted from a specific section of that page.
When tracking purple text highlights across client sites, a clear pattern emerges: generative search engines do not treat URLs as single units of content. Instead, AI Mode evaluates, extracts, and cites individual passages. To understand the scale and mechanics of this behavior, a massive study was conducted, analyzing 15,699,298 AI Mode citations across 148 distinct industries. Nearly half of those citations—47.7%—were scroll-to-text highlights rather than standard, top-of-page URLs.
This reality fundamentally changes how search marketers and editorial teams must approach content optimization. AI Mode’s primary unit of indexation and retrieval is not the webpage; it is the passage. When you analyze citations at the passage level rather than the page level, clear, highly actionable structural patterns begin to surface.
Google Recycles the Passages It Likes Relentlessly
The study of 15.7 million citations ultimately resolved to 4.6 million unique highlighted passages spanning 2.7 million individual pages. While the vast majority of passages (80.9%) were cited only once, a small subset performed far above average. Approximately 2,300 passages were reused 61 or more times by Google’s generative models. In fact, the single most-recycled passage within the dataset was cited a staggering 661 times.
This behavior illustrates a key characteristic of Google’s AI retrieval architecture: when Google identifies a passage it trusts as an optimal answer, it does not use it once and discard it. It returns to that exact same highlighted passage repeatedly across numerous contextually related queries.
One Strong Passage Answers Dozens of Unique Questions
A common misconception is that a passage cited hundreds of times simply answers a single high-volume query repeated by different users. However, the data proves otherwise. The top-performing passage in the study was reused across 483 distinct search queries. Similarly, a single paragraph from a promotional-products retailer answered 221 different questions, while a structured passage explaining how to build a free HOA website was cited across 91 unique queries.
A single well-crafted paragraph can become Google’s universal go-to answer for an entire cluster of search intent. This insight overturns a decade-old SEO impulse: instead of publishing 10 thin, slightly varied pages to capture 10 long-tail keyword variations, publishers need one highly authoritative passage capable of satisfying all of them simultaneously.
Highlight Accumulation Tracks Directly with Top Organic Rankings
As pages accumulate distinct highlighted passages over time, their total citation footprint correlates strongly with traditional organic rankings. Within the dataset:
- Pages containing 1 to 4 highlighted passages had a median organic rank of 11.
- Pages containing 21 or more highlighted passages had a median organic rank of #1, with 67% ranking first outright in standard search results.
- Among passages recycled more than 100 times, 76% originated from pages that held the #1 organic position.
It is important to maintain nuance regarding causality versus correlation. AI Mode relies heavily on pages that have already established authority, trustworthiness, and strong ranking signals in classic organic search. Generative visibility does not operate in isolation from traditional search engine optimization; rather, it compounds directly upon existing SEO strength.
What the Most Recycled Passages Have in Common
To determine why certain blocks of text succeed while others are ignored, over 1,000 actual highlighted passages were reconstructed. Because URL text fragments only preserve the starting and ending anchor words, reaching these insights required pulling the live web pages and isolating the exact highlighted text spans. Evaluating their format, syntax, and sentence structure revealed four distinct characteristics shared by winning passages:
1. They Are Full Paragraphs, Not Brief Snippets
The median highlighted passage length across the data was 117 words. AI Mode does not merely lift single-sentence definitions or isolated bullet points. It extracts complete, multi-sentence contextual paragraphs capable of providing comprehensive answers.
2. They Lead with the Direct Answer
In 80% of extracted client passages and 81% of the web’s most-recycled benchmark passages, the direct answer was positioned in the very first sentence. This inverted-pyramid style proved to be the single most consistent structural pattern in the entire study. If an answer is buried in the middle or end of a paragraph, the likelihood of passage extraction declines significantly.
3. They Are Contextually Standalone
Roughly 85% of all highlighted passages were completely self-contained. They avoided dependent phrases such as “as previously noted,” “in the previous step,” or “see above.” If a passage requires the surrounding text on the page to make logical sense, retrieval models are far less likely to pull it as an isolated snippet.
4. Question-Led Structures Get Recycled Far More Frequently
Among passages that were repeatedly cited, 48% opened with an explicit question—typically formatted as a subheading (H2 or H3)—compared to only 22% among single-use passages. Question-and-Answer (Q&A) and step-by-step how-to formats dominated the highest recycling tiers. Narrative prose and isolated statistical callouts were far less effective at earning repeated citations across multiple queries.
Anatomy of a Winning Passage
To understand how these four traits function in practice, consider a real-world example from the study dataset that earned 158 citations while ranking #1 organically:
How to create an HOA website for free? Creating an HOA website for free can be accomplished using platforms like WordPress, Wix, or Google Sites. These platforms offer free templates and hosting services. Start by selecting a template…
This passage illustrates all four core traits:
- Literal Question Match: The heading explicitly mirrors natural user search queries.
- Answer-First Format: The first sentence provides the core solution before expanding on details.
- Self-Contained Logic: It operates independently without relying on preceding introductory paragraphs.
- Ideal Word Count: It forms a complete, multi-sentence paragraph between 75 and 150 words.
The text is deliberately formatted to facilitate algorithmic retrieval and seamless extraction by generative models.
Actionable Steps for SEO and Content Teams
Optimizing content for Google’s AI Mode does not require a complete rewrite of your existing library. Instead, editorial teams should implement targeted structural adjustments across high-value assets:
- Format Subheadings as Explicit Questions: Replace vague section headers like “HOA Website Options” with direct user queries such as “How do you create an HOA website for free?”
- Adopt the Answer-First Writing Style: Provide a complete, standalone answer in the first sentence directly beneath every core subheading, followed by 75 to 150 words of supporting detail.
- Ensure Passage Independence: Review primary informational sections independently. Remove relative statements, backward references, and pronoun dependencies so each passage stands firmly on its own.
- Prioritize Q&A and How-To Structures: While narrative essays serve brand storytelling, structured Q&A blocks drive recurring AI citations.
- Maintain Traditional SEO Fundamentals: Because passage recycling heavily favors top-ranking URLs, continuing to build page authority, technical health, and strong backlink profiles remains critical.
- Audit Content at the Passage Level: Select your top 10 most critical landing pages and evaluate each H2/H3 section against four core criteria: question-led header, direct answer first, self-contained context, and multi-sentence paragraph length.
The Paradigm Shift in Search Engine Optimization
For over two decades, search engine optimization focused primarily on page-level signals: meta tags, overall keyword density, comprehensive URL depth, and site architecture. However, modern AI engines operate by breaking pages down into smaller text chunks, embedding them vectorially, and retrieving specific passages to construct real-time generative answers.
This shift represents a positive evolution for digital publishers and content strategists. It favors clarity, rigorous formatting, and concise writing over sheer content volume. Restructuring web pages to satisfy passage-level retrieval not only improves visibility within Google’s AI Mode, but also improves performance across traditional featured snippets and user experience overall.
To evaluate how your existing digital footprint performs in generative search, check your domain’s visibility with a free AI citation readiness report. This tool analyzes whether major AI engines cite your content, identifies competing domains occupying those citation spots, and scores your site’s overall readiness for passage extraction.
For a detailed breakdown of the research data, including full statistical distributions, methodology notes, and study boundaries, read the complete text-fragment research study.