It used to, but not anymore. A page holding the top spot on Google can vanish from the AI Overview without any page edits or ranking changes.
Ranking and AI citation are judged by two different systems. Google’s ranking algorithm scores whole pages against a query and returns a ranked list. An LLM breaks pages into passages, retrieves the ones that match the question semantically, and decides what to quote before generating an answer.
If your page is ranking first but not appearing in AI Overviews, it probably means you need to go back and check your chunking. Rewrite some paragraphs to have standalone answers to target queries, and check back after a few days to see if you’ve moved into the citations.
What earns an AI citation after ranking
Ranking #1 means Google’s algorithm judged the whole page best against a query. Citation is a narrower bar: one passage inside that page has to clear a separate check for accuracy and quotability.
| Attribute | Google ranking | AI citation |
| Unit judged | The whole page | A single passage |
| Main signal | Backlinks and domain authority | Clarity, specificity, sourcing |
| When it’s decided | Before the query is asked | At the moment of generation |
| How stable it is | Changes slowly, over weeks | Can shift from one session to the next |
Ahrefs’s analysis of 863,000 SERPs found that only 38% of AI Overview citations now come from a page’s top 10 results, down from 76% a year earlier. A page ranked in the top 10 can disappear from the AI Overview answering the same query while a lower-ranked competitor gets quoted word for word. Citation goes to whichever passage answers the question most cleanly, regardless of which page it lived on.
That doesn’t mean the page failed. It means citation gets decided on a different unit of text, after ranking is already settled.
AI answer engines retrieve passages, not pages
That vanishing act happens because AI systems run a retrieval pipeline before generating anything. Before an LLM writes a single word of its answer, it does the following:
- chunk the page into passages,
- convert each chunk into a numerical embedding, then
- retrieve whichever chunks score highest for semantic similarity to the question asked.
Generation happens last, and only on what the retrieval step handed over. A systematic review of retrieval-augmented generation traces this exact sequence, and a survey of RAG evaluation methods confirms it holds across the major frameworks in production.
AI models never read the page in full, instead building their answers from retrieved fragments.
That’s why a page can rank #1 on document-level relevance and still lose every citation. A claim buried three sentences into a long paragraph, or one that only clicks once you’ve read the heading above it, often can’t stand alone once it’s sliced out of context and handed to a model with nothing else attached. It still works fine for a human scrolling top to bottom, but the isolated chunk doesn’t hold up on its own. Ranking rewards the whole document, while retrieval rewards the sentences within it.
This also reconciles something that looks contradictory at first glance. Google’s AI Overviews layer retrieval on top of traditional search rather than replacing it, so their citations still track ranking more closely than ChatGPT’s or Gemini’s do. Different systems lean on ranking signals to different degrees, which is why citation rates can vary sharply between them even for the same query.
Why passage quality beats backlinks for citation
Ranking authority is still substantially a link graph of who points to a page, how many links they provide, and how trusted they are. None of that graph gets consulted during AI retrieval or generation.
A model deciding whether to quote a passage isn’t checking who linked to it. It’s checking whether the passage is internally consistent, clearly sourced, and specific enough to restate without distortion; that’s a different exam entirely.
A Princeton GEO study found specific content changes, citing sources and statistics, adding quotations, using clear factual phrasing, can lift visibility in generative engine answers by up to 40%.
None of those levers touch a backlink profile. A page with thin link equity but a passage that states a fact plainly, names its source, and stops there can out-cite a page with a stronger link history buried in hedged, qualifier-heavy prose.
E-E-A-T still matters here, by the way, just not through Google’s original mechanism. Expertise and sourcing make a passage machine-extractable long before they make it link-worthy.
Freshness and specificity break ties
Google’s rankings tolerate an undated page just fine, as long as the backlinks and on-page signals hold up. A model generating an answer doesn’t work that way. It favors content that reads as current, with a visible date, a reference to this year, and figures that haven’t gone stale.
Specificity settles the rest. A passage naming an exact figure, a year, or a named organization gives a model something it can restate with confidence. A passage that hedges instead, speaking only in categories like “many organizations” or “figures vary,” gives it nothing safe to quote.
Domain authority doesn’t rescue a page that hedges where a competitor states plainly.
The domains AI tools trust keep shifting
Research on the domains AI tools cite most found sharp drops in citations from Reddit and Wikipedia, sources that sat near the top of that list a year earlier.
Google’s rank tracker has never had to measure that kind of shift, because nothing comparable happens to conventional search rankings. A page holding position ten often stays there until its (or the pages’ around its) backlinks or content change.
Model trust in a domain moves on its own timeline, independent of anything the domain does.
A citation pattern documented six months ago may already be stale strategy today. Building a content plan around which sources a model currently favors is building on a foundation that shifts without warning.
Structure and specificity hold up regardless of which domains fall in or out of favor.
Write the passage before you write the page
Most content briefs start with a keyword and build outward. A better starting point for AI citation is the extractable passage itself: Write the two or three sentences that would answer the question completely if a model pulled only those sentences and showed nothing else. Then build the rest of the page around that passage, not the other way around.
This reverses the usual drafting order. Instead of an introduction that warms up to an answer, the answer comes first, phrased so it survives being lifted out of context. Everything after it exists to support, extend, or qualify that one passage, not to build up to it.
Google’s own guidance on structured data markup points the same direction: content that’s unambiguous about what it’s claiming, and who’s claiming it, is easier for machines to parse correctly. AI citation rewards the same discipline search structured data always has:
Say the thing plainly, attribute it clearly, and let the rest of the page do the explaining.
Run this three-step audit on your best page
- Copy a single paragraph from your best-ranking page, paste it somewhere with no heading and no surrounding text, and read it. Does it answer one specific question completely, or does it lean on the sentence above it to work?
- Open the same page in a browser’s inspect tool or an accessibility checker and look at the heading hierarchy and schema markup. A page that fails a basic structure check is very likely failing AI parseability for the same reason.
- Then run your actual target queries in ChatGPT, Gemini, and AI Overviews, several phrasings, several sessions. When a competitor gets cited instead, check their structure and freshness, not their backlinks.
Make edits as needed and keep an eye on future citations with your favorite AI listening tool.
And celebrate the wins! It’s always fun to see my work reflected in these tools. I want it to be fun for you too.
Frequently asked questions
Do AI answer engines cite the same pages that rank on page one of Google?
Sometimes, but not reliably. A page can rank well in traditional search while never showing up in an AI-generated answer, because one ranks a full page against a query and the other extracts a passage that answers a question cleanly.
What makes a passage easy for an AI model to cite?
A self-contained claim, a named source or number attached to it, and language that doesn’t depend on the sentence before or after it to make sense.
Is schema markup required to get cited by AI tools?
Not strictly. It removes ambiguity and raises the odds, but clearly structured, specific prose can get cited without it. Markup mostly confirms what a model is already inferring from the text.
Is llms.txt worth implementing now?
It’s an emerging, non-standardized convention. Low cost to add, worth watching, but not yet a proven lever for citation.
References
- A systematic review of key retrieval-augmented generation (RAG) systems: Progress, gaps, and future directions. (2017). In Arxiv.org. https://arxiv.org/html/2507.18910v1
- Aggarwal, P., Vishvak Murahari, Tanmay Rajpurohit, Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative engine optimization. arXiv (Cornell University), 33, 5–16. https://doi.org/10.1145/3637528.3671900
- AI overview citations now 54% from organic rankings. (2024a). In Brightedge.com. https://www.brightedge.com/resources/weekly-ai-search-insights/rank-overlap-after-16-months-of-aio
- Harsel, L., Aleksandr Drozdov, & Skopec, C. (2025). The most-cited domains in AI: A 3-Month study. In Semrush Blog. Semrush. https://www.semrush.com/blog/most-cited-domains-ai/
- Hava Salsi, Go, S., & Barcik, B. (2017). Organization schema: What it is & how to implement it. In Semrush Blog. Semrush. https://www.semrush.com/blog/schema-markup-for-company-corporations/
- HTML: A good basis for accessibility – learn web development | MDN. (2024b). In MDN Web Docs. https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/Accessibility/HTML
- Linehan, L. (2026). Update: 38% Of AI overview citations pull from the top 10. In SEO Blog by Ahrefs. https://ahrefs.com/blog/ai-overview-citations-top-10/
- Retrieval augmented generation evaluation in the era of large language models: A comprehensive survey. (2025). In Arxiv.org. https://arxiv.org/html/2504.14891v1
