Citability Is Not a Checklist: Why Your Ranking Content Still Gets Ignored by AI Engines

Steve Palomares

11 Aug 2026 — 11 min read

Photo by Glenn Carstens-Peters / Unsplash

The TL;DR

My article "What Makes a Page Citable to AI? The 7 Markers" ranks #1 organically on Google for its target query. Google's AI Overview for that same query doesn't cite it. It cites a competitor instead. Citability and ranking are not the same mechanism, and this is the proof, from my own site. Citability is a property of individual passages, not pages, and it is measured by different signals than the ones that get you to position one.

What You'll Learn

Why a page can rank #1 and still lose the AI Overview citation to a lower-ranked competitor
The precise, content-level definition of citability, and how it differs from the two other definitions currently in circulation
How query fan-out has decoupled ranking from citation at the mechanism level, with the data to show it
Why RAG chunking is the technical reason passage structure determines extractability
What to actually fix on a page that ranks but doesn't get cited

The Breakdown

Citability is the measurable likelihood that a specific passage of your content, not your whole page, gets pulled into an AI-generated answer. It is not a synonym for ranking well, and it is not a synonym for having a good domain. A page ranks because Google's classic algorithm scores the page as a whole: backlinks, domain authority, on-page relevance, user signals. A passage gets cited because an AI system's retrieval layer scores that specific chunk of text against a specific sub-query, in isolation from the rest of the page.

Those are two different scoring systems evaluating two different units of content. A page with strong aggregate authority can rank #1 while every individual passage on it fails the extraction test. That is not a contradiction. It is what happens when you optimize for one mechanism and assume it covers the other.

Main Content

Why does my page rank #1 but still get skipped by the AI Overview?

I have a direct answer, because it happened to me. My article "What Makes a Page Citable to AI? The 7 Markers" holds the number one organic spot on Google for the exact query "what makes a page citable to AI." I checked. It's there.

Ask Google's AI Overview the same question and it doesn't pull from my article. It pulls from bradleebartlett.com.

I built that page to rank. It has backlinks, it answers the query, it has been live long enough to earn trust signals. All of that got it to position one. None of it guaranteed a citation. That gap between "ranks first" and "gets cited first" is the whole subject of this article, and it's not a fluke specific to my site. It's a structural feature of how AI answer engines work now.

What is citability, actually?

Here's the problem: ask five different sources what "citability" means and you'll get three different answers, and none of them agree with each other.

Some content treats citability as a brand-level property: can HumanizeAI, as an entity, be cited by name. Some content treats it as a step in a funnel: Visibility leads to Citability leads to Retrievability, three sequential stages a piece of content has to clear. Neither of those is wrong, exactly, but neither of them is useful if you're staring at one specific page trying to figure out why it isn't showing up.

The definition that actually helps you fix something is this: citability is a property of a passage's extractability, evaluated independently of the page it sits on. It's not about your brand. It's not a funnel stage. It's whether a self-contained block of your text, 100 to 300 words, can be lifted out by a retrieval system and used as a clean answer to a specific question, with no other context required.

That's the operative definition for anyone who has to actually go fix a page. Everything else in this article uses it.

Why has ranking stopped predicting AI citation?

Because the two systems are no longer answering the same question, and there's now data showing the gap widening in real time. In July 2025, roughly 76% of pages cited in Google's AI Overviews also ranked in Google's own top 10 results for that query. By March 2026, that overlap had fallen to approximately 38%, according to Ahrefs' analysis of 863,000 keywords and 4 million AI Overview URLs (Search Engine Journal, reporting Ahrefs data, 2026). That's not a small shift. That's the primary predictor of AI citation losing more than half its predictive power in eight months.

The mechanism behind that drop is query fan-out. When you type one question into Google's AI Overview or ChatGPT, the system doesn't just retrieve pages for the query you typed. It decomposes your question into a set of related sub-queries, runs retrieval against each one separately, and assembles the final answer from whichever passages surface consistently across that whole set. Your page might win the exact query you typed and still lose the fan-out, because it never surfaces for the three or four adjacent sub-queries the system also ran.

This is why domain authority alone has stopped being a reliable predictor. A page can be the single best answer to the literal search string and still be the wrong answer to the four sub-questions the AI actually asked behind the scenes.

What is RAG chunking and why does it decide whether your content gets extracted?

Most AI answer engines run on some version of retrieval-augmented generation, or RAG. Before your page ever gets evaluated for an answer, it gets broken into chunks, typically somewhere in the 512 to 1024 token range depending on the system and the query type, and each chunk is indexed and scored independently.

That single technical fact explains almost every AEO best practice you've ever read, and almost none of the corpus actually says so directly. Where the chunk boundary lands determines what gets retrieved together and what gets split apart. A paragraph that depends on the paragraph before it to make sense might get chunked away from that context entirely. If your key claim is in sentence four of a paragraph, and the retrieval system pulled a chunk that starts at sentence five, your claim is gone.

This is the real, mechanical reason BLUF writing works, why modular self-contained paragraphs outperform flowing narrative prose, and why sentences that rely on "as mentioned above" or "this approach" fail extraction even when a human reader would follow them just fine. The AI reading your chunk has no idea what came before it. Neither does the scoring system deciding whether to cite it.

A page can rank #1 while every individual chunk on it fails independently, because ranking scores the page and chunking evaluates each fragment in isolation.

Why do 70 to 88 percent of top-ranking pages get skipped by AI citations?

Because most content on the internet, including a lot of content that ranks extremely well, was written for one scoring system and is being evaluated by another. Depending on which study you look at, somewhere between 70% and 88% of Google's top-ranking pages for a given query are not cited by AI search engines answering the same question. Position Digital's 2026 analysis found that only 12% of URLs cited by ChatGPT, Perplexity, and Microsoft Copilot ranked in Google's top 10 at all (Position Digital, 2026, reported via AuthorityTech). That means roughly 88% of AI-cited pages weren't top-10 Google rankers, and the inverse holds too: the large majority of top-10 rankers aren't showing up in the citations.

The existing content on this problem treats it almost entirely as an optimization problem for pages that don't rank. Write better content, get better backlinks, earn the ranking, and citability will follow. That advice is fine for a page starting from zero. It is useless for a page that already has the backlinks, already has the authority, and still isn't being cited, because the thing actually blocking it isn't authority. It's passage structure. Nobody in the corpus is telling operators with strong-ranking content that they have a second, separate problem to solve.

How does HumanizeAI's Generative Engine Optimization Visibility Framework apply to citability specifically?

Our own Generative Engine Optimization Framework runs three stages: Prompt Audit, Content Gap Analysis, and Iterate and Monitor. Inside the Content Gap Analysis stage, every page gets evaluated against the Three Citation Tests: can AI see it, does AI have proof, does AI trust the brand.

Citability, as defined in this article, sits squarely inside the first test, "can AI see it," but at a resolution the framework hadn't spelled out until now. It's not enough to ask whether a page is technically crawlable. You have to ask whether each individual passage on that page would survive being lifted out on its own, at chunk size, with zero surrounding context. A page can pass every technical crawlability check and still fail this test paragraph by paragraph.

What is a competitor trying to do with the term "citability" right now?

A tool called citability.dev has built a scoring product, CiteRank, around the term. That's a signal worth paying attention to, not because the tool itself matters much yet, but because it means the term is up for grabs. Nobody has cleanly and rigorously defined "citability" as content-level passage extractability, distinct from brand mentions and distinct from a generic visibility funnel, in a way that's actually useful to a practitioner sitting in front of a page that ranks and still isn't cited.

That's the gap this article is closing. Own the definition, and you own the framing every future conversation about the term has to reference.

HumanizeAI Framework References

This article extends two HumanizeAI frameworks at once. We built something called The Generative Engine Optimization Visibility Framework and the Three Citation Tests. Both of these help us get a more precise operating definition for "Can AI see it?" passage-level extractability, not just technical crawlability. We also built the H.E.A.R.T. framework, which has an “Answer First” principle. This is the direct writing-level fix for the RAG chunking problem described above: if every section leads with its answer in the first two to three sentences, you're structurally aligning your content with how chunk boundaries actually get drawn.

Founder Observation

My article "What Makes a Page Citable to AI? The 7 Markers" ranks #1 organically in Google for the exact query "what makes a page citable to AI." I built it to rank, and it worked. When I checked what Google's AI Overview actually cites for that same question, it wasn't my article. It was bradleebartlett.com.

That stung a little, honestly, given the subject matter. But it's also the cleanest piece of evidence I have that ranking and citability are genuinely separate mechanisms, not two names for the same outcome. I have a page with real authority signals, real backlinks, a real top position, and it still lost the citation to a competing page. If it can happen on an article specifically written about citability, it can happen on any page. That's not a hypothetical for HumanizeAI. It's a live data point sitting on our own domain right now, and fixing it is next on my list.

Research & Supporting Evidence

The share of AI Overview citations that also ranked in Google's top 10 fell from roughly 76% in July 2025 to approximately 38% by March 2026, based on Ahrefs' analysis of 863,000 keywords and 4 million AI Overview URLs, reported by Search Engine Journal (2026).

Only 12% of URLs cited by ChatGPT, Perplexity, and Microsoft Copilot rank in Google's top 10, and roughly 80% of ChatGPT's most-cited pages don't rank in Google's top 100 at all, according to Position Digital's 2026 data as reported by AuthorityTech.

RAG systems commonly chunk content in the 400 to 1024 token range depending on query type, with factoid queries favoring smaller chunks around 256-512 tokens and analytical queries favoring larger chunks of 1024 tokens or more, per Firecrawl's 2026 guide to chunking strategies for retrieval-augmented generation.

Mini Case Study

This is an illustrative case study, composed from patterns we've seen across HumanizeAI's own AEO/GEO work with 3 early stage founders of SaaS companies. A SaaS client's pricing comparison page ranked in the top 3 for its target keyword for over a year. Strong backlinks, solid domain trust, consistent traffic. When we ran a Prompt Audit against it using our own Generative Engine Optimization Visibility Framework, the page showed up in zero AI Overview answers and zero ChatGPT responses to the same question.

The page itself wasn't weak. The individual paragraphs were the problem. Every pricing claim depended on a comparison table two sections above it. Every "as shown above" reference broke the moment a retrieval system pulled that paragraph as an isolated chunk. We rewrote the same information as self-contained blocks, each one restating the plan name and the specific number instead of pointing back to the table, without changing the page's ranking-relevant elements at all. Within the next monitoring cycle, the page started surfacing in AI answers for two of the five tracked prompts. Nothing about its Google ranking moved. That's the tell: the ranking signal and the citation signal responded to two completely different fixes.

Key Takeaways

Citability is a property of a passage's extractability, evaluated independently of the page it sits on, not a property of the page as a whole.
Ranking and AI citation have measurably decoupled: top-10 overlap with AI Overview citations fell from roughly 76% to 38% between July 2025 and March 2026.
Query fan-out is the mechanism behind that decoupling. AI systems answer decomposed sub-queries, not just the literal string a user typed.
RAG systems chunk content into 400-1024 token blocks and score each chunk independently, which is why context-dependent sentences fail extraction even on high-ranking pages.
A page can pass every ranking signal and still fail citability at the paragraph level. Fixing one does not automatically fix the other.

FAQ

Why does my page rank on Google but not show up in AI Overviews or ChatGPT? Because ranking and AI citation are scored by different mechanisms. Google's classic ranking evaluates your whole page: backlinks, domain trust, relevance signals. AI citation evaluates individual passages in isolation, often after your page has been broken into retrieval chunks. A page can win on page-level signals and still lose on passage-level extractability.

What is citability in AI search? Citability is the measurable likelihood that a specific passage of your content can be extracted and used as a clean, self-contained answer by an AI system, independent of the rest of the page it sits on. It's a passage-level property, not a brand-level or page-level one.

What is the difference between AEO, GEO, and citability? AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) describe the broader discipline of optimizing content and technical infrastructure for AI search visibility. Citability is one specific, measurable factor inside that discipline: whether your passages survive extraction. You can be technically AEO-ready and still fail citability if your paragraphs depend on surrounding context to make sense.

Does good content automatically mean citable content? No. Research suggests that somewhere between 70% and 88% of Google's top-ranking pages are not cited by AI search engines for the same queries. Good content in the traditional SEO sense, meaning content that earns backlinks, satisfies search intent, and ranks well, is not automatically structured for passage-level extraction. Those are two separate qualities a page has to earn independently.

What is query fan-out and why does it matter for citations? Query fan-out is when an AI system decomposes one user question into multiple related sub-queries, runs retrieval against each one, and builds its final answer from whichever content surfaces consistently across that whole set. It matters because your page might be the strongest match for the literal question a user typed and still lose the citation because it never surfaces for the adjacent sub-queries the system also checked.

How does RAG chunking affect whether my content gets cited? RAG systems split content into chunks, typically 400 to 1024 tokens, before scoring it for retrieval. Each chunk is evaluated with little or no awareness of the rest of the page. If your key point depends on a sentence, table, or claim from earlier in the page, and the chunk boundary falls after that dependency, the extracted chunk won't make sense on its own and is unlikely to get cited.

How do I make my content more citable without rewriting the whole page? Focus on the passages carrying your most important claims first. Rewrite them as self-contained units of roughly 100-300 words: restate the specific number, name, or fact instead of referring back to an earlier table or paragraph, and lead each unit with its answer instead of building up to it. You're not rewriting the page for ranking. You're rewriting individual passages so they survive being lifted out alone.

Additional Resources

About the Author

Steve Palomares has spent 25+ years building software companies. Now owner of HumanizeAI, he writes about AI content strategy for marketing, AEO, GEO and growing software businesses with AI. Based in North Texas.