There is a gap between what marketers assume about AI search and what the data actually shows.
For years, the dominant belief was straightforward: rank high in Google, and you will appear in AI-generated answers. Build authority, earn backlinks, optimize for keywords — and visibility will follow, regardless of the platform generating the answer. That belief is not entirely wrong, but it is incomplete in ways that now carry real consequences for brands relying on organic discovery.
A large-scale analysis by AirOps, examining 548,534 pages retrieved by ChatGPT across 15,000 original prompts, makes the mechanics of AI citation selection measurably clear for the first time. The findings reshape how content teams, SEOs, and digital marketers should think about the path from “being discoverable” to “being cited” — two stages that are far less connected than most assume.
This post breaks down what that research found, explains the mechanics of retrieval and fan-out query expansion, examines the specific role Google rankings play in ChatGPT’s citation process, and draws out the practical implications for any brand that wants to appear in AI-generated answers across ChatGPT, Perplexity, Google AI Mode, and similar platforms.
What Actually Happens When ChatGPT Answers a Question
Most users experience ChatGPT as a single, clean response to whatever they type. Behind that response is a multi-stage process that is considerably more complex than most content strategies account for.
When a user submits a query, ChatGPT does not simply draw on its training data and produce an answer. In search-enabled mode, it initiates a retrieval process — querying the web, pulling pages, evaluating them, and then synthesizing a response that may cite some of what it found. The key word there is “may.” Retrieval and citation are two distinct events, and the gap between them is where most brands quietly disappear.
The AirOps dataset puts a number to that gap: only 15% of retrieved pages were ultimately cited in a final response. The other 85% were found, evaluated, and discarded — never surfacing for the user at all. That means a brand can be technically discoverable by ChatGPT and still have no visible presence in the answer that a user actually reads.
This distinction changes the framing of what content optimization for AI search actually means. It is not enough to ensure that your pages enter the retrieval pool. The more meaningful question is whether your pages survive the selection step — whether they are chosen over the dozens of other sources ChatGPT evaluated for the same answer.
The mechanics of how that selection works, and what factors correlate with being chosen, are what the rest of this analysis examines.
Understanding Retrieval Fan-out: How One Query Becomes Dozens
The concept of “fan-out” is central to understanding why most brand-level keyword tracking fails to capture the full citation picture in AI search.
When ChatGPT receives a query, it typically does not search for that phrase once and stop. Instead, it generates a series of follow-up searches — what researchers call fan-out queries — that expand the retrieval surface well beyond the original keyword. A user asking “What should I look for in VDR vendors?” may trigger ChatGPT to search for vendor evaluation criteria, document security features, pricing structures, compliance requirements, and comparison frameworks — all as part of building a single answer.
The AirOps data quantifies this: 89.6% of the 15,000 original prompts in the study triggered two or more fan-out queries. The total query set expanded from 15,000 original prompts to 43,233 queries — nearly a 3x increase. That expansion is not random. It reflects how ChatGPT decomposes a question into its component parts, each of which requires independent retrieval before the synthesis stage.
The significance for brands is direct. If a brand tracks only its primary target keyword and monitors whether its pages rank for that term, it is missing the majority of the search surface that actually determines ChatGPT citation. 32.9% of all cited pages in the AirOps study appeared in the results for fan-out queries only — not the original prompt. They were never discovered through the primary keyword at all.
That means nearly one-third of citation opportunities exist entirely outside the tracking scope of a conventional keyword strategy.
How Fan-out Queries Are Structured
Fan-out query behavior is not uniform across query types. The AirOps research found meaningful differences in how ChatGPT expands different categories of questions.
For informational queries — definitions, how-to guidance, evaluations, and explorations — ChatGPT tended to stay relatively close to the original phrasing. 39.9% of informational fan-out queries were near-verbatim restatements of the original prompt. Definition queries were the most stable: 51.6% stayed near-verbatim. How-to queries rewrote at 42.6%, usually by adding category or tool-specific context to the original phrasing.
Commercial queries behaved differently. Instead of restating the original question with modest adjustments, ChatGPT more often decomposed commercial prompts into component-level sub-searches. A comparison query like “HubSpot vs Salesforce” was frequently broken into separate searches for pricing, features, user reviews, and integration compatibility — each one a distinct retrieval event. Comparison queries split into sub-queries 38.4% of the time, the highest rate of any category.
Research queries (best or top options) stayed near-verbatim 50.6% of the time but were the only category where year-qualifier modifiers appeared at meaningful volume — 9.7% of research fan-out queries included a year, suggesting ChatGPT actively looks for freshness signals in research-oriented contexts.
This structural difference in how fan-out queries are generated has direct content strategy implications. Informational content benefits from depth on the core topic. Commercial content, by contrast, requires modular coverage that addresses pricing, alternatives, feature-level specifics, and comparison angles — because those are the component searches ChatGPT uses to build answers in commercial categories.
The Volume Gap That Makes Fan-out Invisible
One of the most operationally significant findings in the AirOps study concerns the relationship between fan-out queries and traditional keyword volume: 95% of fan-out queries had zero monthly search volume by conventional keyword tool metrics.
This is not a minor limitation. It means that the standard tooling most content teams use to identify keyword opportunities is structurally blind to the vast majority of the search surface that determines ChatGPT citation. A keyword research workflow built on Ahrefs, Semrush, or Google Search Console will correctly identify primary target terms but will systematically miss the follow-up queries ChatGPT generates — the ones responsible for nearly a third of all citation opportunities.
Brands that rely exclusively on traditional keyword tracking are, in effect, optimizing for a subset of the retrieval environment. The rest of the map is invisible to them.
How Google Rankings Correlate with ChatGPT Citations
Given that 95% of fan-out queries have no traditional search volume, and given that ChatGPT clearly retrieves from multiple sources rather than Google alone, it would be reasonable to question whether Google rankings matter at all for AI citation. The data says they do — significantly — but the relationship is more nuanced than the binary “SEO = AI visibility” assumption implies.
The Position-1 Advantage
The AirOps study mapped all cited and retrieved-but-uncited pages against Google’s top 20 results for both original and fan-out queries. The overall finding: 55.8% of all cited pages ranked in Google’s top 20 for at least one query. That is a meaningful correlation, but it leaves substantial room for citation paths that run entirely outside of Google’s rankings.
More instructive is the citation rate by ranking position. Among pages holding position 1 in Google, 43.2% were cited by ChatGPT — a citation rate 3.5 times higher than pages ranking outside the top 20. This is not a trivial advantage. Being the top-ranked result in Google does not guarantee a citation, but it produces a fundamentally different citation probability than ranking on page two or beyond.
The practical implication is that brand content teams should think about Google rankings not just as a traffic metric but as an input into ChatGPT’s citation process. Moving high-value pages from position 10 to position 1 is not just a click-through-rate improvement — it is a citation probability upgrade.
The Hybrid Reality: ChatGPT Does Not Run on Google Alone
The correlated picture is complicated by a separate Ahrefs analysis that examined ChatGPT’s actual fan-out query results against what Google would return for the same queries. The findings create a more accurate picture of the underlying architecture.
Only 6.82% of ChatGPT search results appeared in Google’s top 10 for the same queries. Only 16.61% appeared anywhere in Google’s organic results. That means 83.39% of pages ChatGPT retrieved did not appear in Google’s SERPs at all for the equivalent query.
This strongly suggests ChatGPT uses a hybrid retrieval architecture — drawing from multiple sources including Google results, Bing, its own web index, and potentially third-party APIs — and then applies its own re-ranking algorithm on top of the combined retrieval set. OpenAI has an incentive to reduce dependence on any single provider, and the data is consistent with that interpretation.
The practical takeaway from combining both studies: Google rankings are a meaningful input into ChatGPT’s citation process, but they are not the whole picture. A page that ranks #1 in Google enters the citation competition with a structural advantage. A page that does not rank in Google at all is not automatically excluded — but it faces harder odds. Content that sits at positions 10–20 in Google is particularly worth examining, because improving those pages toward the top 5 or top 3 carries a compounding benefit: more Google traffic and a meaningfully higher ChatGPT citation probability simultaneously.
Domain Authority and the Mid-Tier Citation Advantage
One of the less expected findings in the AirOps dataset concerns domain authority distribution. The common assumption in the SEO world is that high-authority domains — news publishers, large media brands, Wikipedia, major platforms — dominate the citation landscape. The data draws a more differentiated picture.
Approximately 74% of citations in the dataset went to sites with domain authority under 80. The majority of citations did not cluster at the top of the authority spectrum. Sites in the DA 20–80 range accounted for 63.6% of all citations. Most strikingly, the DA 20–40 tier alone contributed a larger citation share than the DA 80–100 tier — 26.0% versus 25.4%.
The citation rate by retrieval further complicates the expected hierarchy. From DA 0 through DA 80, the share of retrieved pages that earned a citation held steady — between 21.5% and 23.6%. The only tier that significantly underperformed its retrieval frequency was DA 80–100, which had a citation rate of only 15.0% despite being retrieved more often than any other tier.
High-authority domains are retrieved frequently. But once retrieved, they are selected at a lower rate than mid-authority sites. That finding runs counter to the assumption that the largest publishers have a structural citation lock — they are present, but their presence does not automatically translate into selection.
This creates a genuinely usable competitive opening for mid-size brands and specialized publishers. The DA 40–80 range — where many sector-focused media companies, B2B brands, and professional blogs sit — shows citation rates comparable to or better than those of the largest authority domains. The determining factor is not primarily authority but rather how well a retrieved page matches the specific information need that ChatGPT is trying to answer.
A domain authority of 45 combined with strong topic coverage, clear title-to-query alignment, and readable content structure can compete meaningfully with a DA 90 publication that covers the same topic less specifically. The data supports that outcome at scale.
What Separates Cited Pages From Retrieved-But-Ignored Pages
With 85% of retrieved pages never reaching a citation, the question of what actually separates cited from uncited pages is the most practically important question in this analysis. The AirOps data identifies several measurable differentiators.
Title-to-Query Alignment
The most clear-cut signal in the dataset relates to how closely a page’s title matches the query — original or fan-out — that led to its retrieval.
Pages where 50% or more of title words matched the query had a 20.1% citation rate. Pages where title-query overlap was below 10% had a citation rate of 9.3%. That is a 2.2x difference driven primarily by a title-level signal.
This is actionable. It means that page titles written for broad brand appeal or general SEO purposes — titles that gesture at a topic without directly naming the query terms — are likely underperforming their citation potential. Titles that precisely reflect the specific question a page answers are cited more than twice as often.
For content teams, this suggests that existing high-ranking pages with generically written titles may benefit from revisions that increase title-query alignment. This is not about keyword-stuffing; it is about making explicit what a title’s page actually answers, using the language a searcher (or, now, an AI system) would use to look for that answer.
Readability as a Selection Signal
Readability scores also correlated with citation behavior. Pages with Flesch Reading Ease scores of 50 or higher were more commonly found among ChatGPT’s cited pages. A Flesch score of 50 roughly corresponds to content that is accessible to a general adult audience — not academic or technical in complexity, but substantive.
This does not mean that technical content cannot be cited. It means that readability is part of the selection calculus. When ChatGPT evaluates retrieved pages, content that is well-structured and clearly written appears to have an edge over content that is dense, jargon-heavy, or structured for a narrow specialist audience.
Content Positioning Within the Page
A separate study of ChatGPT citation patterns found that 44.2% of citations drew from content located in the first 30% of a page. The middle section (30–70% into the page) contributed 31.1% of citations, and the final third contributed only 24.7%, with a marked drop toward the page end.
For structure, this means that the most important information — the direct answer to the query a page is designed to address — should appear early. ChatGPT’s retrieval process appears to front-load its evaluation of page content. Long preambles, extended context-setting sections, and buried answers reduce the probability of being cited, regardless of how good the underlying content is.
Freshness
Content recency matters, particularly in commercial and research query contexts. According to SE Ranking’s analysis, content updated within the past three months is twice as likely to be cited by ChatGPT as older, outdated pages. For research queries, which showed the highest rate of year-modifier use in fan-out searches (9.7% including year qualifiers), freshness is not just a bonus signal — it is a competitive variable.
Brands should treat content refresh schedules as part of their AI citation strategy, not just their traditional SEO maintenance. Pages that cover high-intent topics but haven’t been updated in 12–18 months are likely leaving citation opportunities to competitors who are publishing fresher versions of the same information.
Query Intent and Citation Rate Variation
Citation probability is not uniform across query types. The AirOps breakdown by intent category reveals meaningfully different citation dynamics depending on what kind of question is being asked.
Product-discovery queries — awareness-stage searches asking what tools or solutions exist — had the highest citation rate at 18.3%. This is notable: awareness-stage content, often deprioritized relative to conversion-stage content in traditional content strategies, appears to have a higher likelihood of earning a citation in AI-generated answers.
How-to queries were close behind at 16.9%. Procedural content that walks through specific steps performs well in ChatGPT’s selection process, likely because the clear action structure makes it easy for the model to extract and incorporate specific guidance.
Validation queries — searches asking for specific product details like pricing, compatibility, or feature specifics — showed the lowest citation rate at 11.3%. Comparison queries followed at 13.1%. This lower rate for validation queries makes sense in context: ChatGPT may be more likely to synthesize validation information across multiple sources rather than citing a single page, reducing the per-page citation probability even when the content is high quality.
For content prioritization, this means that building or refreshing product discovery and how-to content may yield stronger citation returns than comparable effort spent on comparison or pricing content — though both categories still matter.
The Practical Content Strategy Implications
Taken together, the data from the AirOps study and corroborating research from Ahrefs, SE Ranking, and ekamoira.com creates a fairly precise picture of what a brand needs to do to improve its ChatGPT citation probability. It involves neither ignoring traditional SEO nor treating it as the complete answer.
First, think beyond primary keywords. Traditional keyword tracking captures the original query but misses the 43,000+ fan-out queries that ChatGPT generates in the course of answering those original queries. Brands need to understand which sub-topics and follow-up questions their primary content topics generate — and whether their existing pages cover those angles. Content gaps around adjacent questions are citation gaps in disguise.
Second, audit existing pages at positions 10–20 in Google. These pages are within reach of the top-SERP citation advantage but are not yet receiving its full benefit. Improving depth, updating freshness, tightening title-query alignment, and expanding coverage of adjacent fan-out topics on these pages offers one of the highest-leverage citation improvement opportunities available.
Third, optimize page structure for early extraction. Given that nearly half of citations come from the first 30% of a page, leads should be direct and answer-first. If a page answers the question “What are the main types of document management software?”, that answer should appear in the first paragraph, not after two sections of background context.
Fourth, treat content freshness as a citation signal. For any page targeting commercial, research, or validation queries, a content review and update cadence of every 3–6 months is now part of competitive AI citation maintenance — not just good editorial hygiene.
Fifth, expand into topic clusters that serve both original queries and fan-out structures. A single pillar piece may rank well in Google but fail to cover the 8–12 sub-questions ChatGPT generates around that topic. A well-structured cluster of supporting pages — each answering a specific sub-question — increases the probability that at least one page appears in the retrieval set for each fan-out query, multiplying citation exposure across the full query expansion.
Sixth, rewrite title tags for specificity, not just keyword density. Pages with strong title-query alignment are cited at more than double the rate of pages with weak alignment. The investment required to rewrite a title tag is minimal compared to producing new content — and for pages that are already ranking and being retrieved, it may be the highest-impact single optimization available.
Why This Matters for the Broader Shift Toward AI-Driven Search
The AirOps data and its corroborating studies arrive at a moment when the structural shift in how users retrieve information is well past the experimental phase.
ChatGPT crossed 800 million weekly active users. Google AI Overviews appear on approximately 25% of all queries. Perplexity, Claude, and Gemini are being used for research tasks that would previously have generated multiple Google clicks. Zero-click searches — where users receive answers without visiting any external page — now account for an estimated 69% of search sessions.
That shift changes the economics of visibility. Traffic driven by a top Google ranking is still valuable, but it is no longer the only form of useful search visibility. A brand cited in a ChatGPT answer that 800 million weekly users read is reaching an audience that may never click through to any page. Citation becomes a form of brand presence independent of click-through rate.
This is particularly important for the ~28% of ChatGPT’s top 1,000 cited pages that, per Ahrefs data, have zero organic visibility in Google. Those pages are being seen in AI-generated answers without ever generating a traditional SEO traffic signal. For brands measuring content ROI exclusively through Google Analytics or Search Console sessions, this means that some of their most impactful citation-earning content is invisible to their own measurement systems.
The emergence of “Answer Engine Optimization” (AEO) as a discipline alongside traditional SEO reflects this reality. AEO extends the logic of SEO — structured content, topical authority, technical accessibility — toward an additional objective: being selected by an AI synthesis engine as a reliable, citable source. The mechanics differ. The underlying commitment to producing specific, accurate, well-structured content that directly answers the questions people actually ask is the same.
The Fan-out Opportunity Most Brands Are Still Missing
Perhaps the most actionable finding in this entire body of research is the one that is easiest to overlook: 32.9% of cited pages were found only through fan-out queries, not the original prompt. Nearly one-third of all citation opportunities exist outside the primary keyword — in follow-up searches that conventional analytics cannot see and that standard keyword tools register as having zero monthly volume.
That is not a niche edge case. It is a structural feature of how ChatGPT answers questions. Any brand that optimizes only for its primary target terms and ignores the question-expansion layer is, by design, unreachable for roughly a third of the citations it could otherwise earn.
Addressing this gap does not necessarily require building an entirely new content library. In many cases, it means strengthening existing pages to cover the adjacent questions that ChatGPT generates while researching the primary topic. A page on “enterprise data room software” may already rank for that core term. Whether it also covers the follow-up questions — regulatory compliance features, granular permission structures, audit trail requirements, vendor comparison criteria — determines whether it gets cited not just for the primary query, but for the fan-out searches that make up the bulk of ChatGPT’s retrieval process.
Expanding topic coverage on high-performing existing pages, informed by an understanding of which fan-out queries they could plausibly rank for, is one of the most direct and cost-efficient paths to improving AI citation visibility.
Frequently Asked Questions
What is retrieval fan-out in ChatGPT? Retrieval fan-out refers to the process by which ChatGPT expands a single user query into multiple follow-up searches during answer generation. Instead of retrieving pages for just the original question, ChatGPT generates additional queries — typically 2 to 12, depending on question complexity — that cover sub-topics, adjacent questions, and related angles. These fan-out queries allow ChatGPT to build more comprehensive answers but also mean that the retrieval surface is far larger than any single keyword. In the AirOps study covering 15,000 prompts, fan-out expanded the total query count to 43,233 — nearly three times the number of original prompts.
How does ChatGPT decide which pages to cite? ChatGPT retrieves many more pages than it ultimately cites. Of 548,534 pages retrieved in the AirOps study, only 15% appeared in a final answer as a citation. Selection appears to be influenced by several factors: how closely a page’s title aligns with the specific query that triggered its retrieval, the page’s readability (Flesch Reading Ease of 50+ correlates with higher citation rates), how well the page’s content directly answers the question, whether the content appears early in the page structure, and the page’s Google ranking position. Pages at position 1 in Google were cited 3.5 times more often than pages outside the top 20 results.
Does ranking high in Google guarantee citation in ChatGPT? No. While Google ranking position strongly correlates with ChatGPT citation likelihood, it does not guarantee it. 85% of retrieved pages are never cited regardless of their Google ranking. Additionally, a separate Ahrefs analysis found that only 16.61% of pages ChatGPT retrieved even appeared in Google’s search results at all for the same queries — meaning ChatGPT draws substantially from sources beyond Google. Strong Google rankings increase citation probability, but they are one input into a multi-factor selection process, not a sufficient condition on their own.
What is the relationship between domain authority and ChatGPT citations? High domain authority does not dominate citation selection in the way many assume. The AirOps data found that approximately 74% of citations went to sites with domain authority under 80. Sites in the DA 20–80 range earned 63.6% of all citations. More surprisingly, the DA 80–100 tier had a citation rate of only 15% after retrieval — lower than every other authority tier. Mid-tier sites with DA between 20 and 80 were cited at rates between 21.5% and 23.6% of retrievals, comparable to or better than the highest-authority domains. The evidence suggests that topic relevance, content quality, and query alignment matter more than authority tier alone.
What percentage of ChatGPT fan-out queries have zero traditional keyword search volume? According to the AirOps study, 95% of fan-out queries generated by ChatGPT had zero monthly search volume according to standard keyword tools. This means that the vast majority of the query expansion that drives ChatGPT’s retrieval process is entirely invisible to conventional keyword research workflows. Brands relying solely on traditional keyword tracking will systematically miss the search surface that produces nearly a third of all ChatGPT citation opportunities.
How does query type affect citation rates in ChatGPT? Citation rates vary meaningfully by query intent. In the AirOps dataset, product discovery queries (awareness-stage) had the highest citation rate at 18.3%, followed by how-to queries at 16.9%. Comparison queries came in at 13.1%, and validation queries (seeking specific product details like pricing or compatibility) had the lowest citation rate at 11.3%. This variation suggests that content strategy for AI citation should prioritize query types with higher citation conversion rates — particularly product-discovery and procedural how-to content.
Does content freshness affect ChatGPT citation probability? Yes, substantially. SE Ranking’s analysis found that content updated within the past three months is approximately twice as likely to be cited by ChatGPT as older content. Research queries, which had the highest rate of year-modifier inclusion in fan-out searches (9.7%), are particularly sensitive to freshness. For any brand targeting competitive commercial or research queries, maintaining a regular content review and update cadence — roughly every 3–6 months — is now part of active citation management, not just general content maintenance.
What role does title-query alignment play in ChatGPT citations? It plays a measurable and significant role. Pages where the title matched 50% or more of the query terms had a citation rate of 20.1%. Pages where title-query overlap fell below 10% had a citation rate of just 9.3% — a 2.2x difference. This suggests that page titles written for general brand messaging rather than precise question-matching are likely underperforming their citation potential. Revising titles on existing high-ranking pages to more directly reflect the specific questions those pages answer is one of the most accessible citation-optimization moves available.
How can brands find and optimize for fan-out queries? Traditional keyword research tools cannot surface most fan-out queries because 95% have zero monthly search volume. Specialized tools and platforms — including AirOps Insights — can identify the fan-out queries ChatGPT generates for specific prompts, allowing brands to understand the follow-up search landscape around their core topics. Without such tools, a practical approach is to manually map the sub-questions a primary topic generates: what would someone need to know to fully answer the main query? Each sub-question represents a potential fan-out target, and creating or expanding content to cover those questions builds citation surface in areas that competitor keyword strategies typically ignore.
Is ChatGPT primarily powered by Google for its search results? The evidence suggests no. An Ahrefs analysis found that only 6.82% of ChatGPT’s actual search results appeared in Google’s top 10, and only 16.61% appeared in Google’s organic results at all. 83.39% of pages ChatGPT retrieved did not appear in Google’s search results for the same queries. The most accurate characterization is that ChatGPT uses a hybrid retrieval approach — drawing from multiple sources including Google, Bing, its own web index, and potentially third-party APIs — and applies its own re-ranking logic. Google rankings influence citation probability, but ChatGPT is not a Google proxy.
What content structure best supports ChatGPT citation? Research on citation location within pages found that 44.2% of ChatGPT citations came from content in the first 30% of the page. Citations from the final third of content dropped sharply. This indicates that answer-first content structure — where the direct response to the page’s core question appears early — is favored by ChatGPT’s retrieval evaluation. Long introductions, extensive background-setting, and buried answers reduce the probability of being cited, even if the underlying information is strong. Formatting choices that make key answers extractable — clear headers, numbered lists for procedural content, short paragraphs for definitions — also reduce the cognitive cost for the model to identify and select a page.
How does ChatGPT treat pages that rank well in Google but score low on content quality signals? The citation data suggests that strong Google rankings improve retrieval frequency — pages ranking #1 in Google are more often included in ChatGPT’s retrieval pool — but that the selection from retrieved pages depends on quality signals independent of ranking position. High-authority pages with DA above 80 had a lower citation rate (15%) after retrieval than mid-authority pages (21.5–23.6%), suggesting that authority alone does not compensate for weaker content quality or lower topic specificity. A page ranking in position 1 for a broad query but failing to directly address the specific sub-question ChatGPT is evaluating may be retrieved but not cited.
How many fan-out queries does ChatGPT generate per prompt on average? The AirOps study found that ChatGPT generates two or more fan-out queries on 89.6% of searches. Ahrefs’ separate analysis found that ChatGPT pulls an average of 1.78 search queries per prompt, with 75% of prompts triggering exactly two searches. More complex queries — particularly commercial queries in the comparison and research categories — tended to generate a higher number of fan-out branches, with decomposition into sub-queries occurring at rates of 38% or higher for comparison queries.
What is the citation advantage of ranking in Google’s top 20 versus beyond? The AirOps study found that 55.8% of all cited pages ranked in Google’s top 20 for at least one query (original or fan-out). Pages at position 1 had a 43.2% citation rate — 3.5 times higher than pages outside the top 20. The advantage tapers from position 1 downward but remains measurable through the top 20. Beyond position 20, pages enter a much more competitive retrieval environment where citation rates drop substantially, and where the page is more likely to be retrieved only through a hybrid non-Google source rather than through Google’s organic results.
The body of evidence on retrieval fan-out, Google SERP influence, and ChatGPT citation mechanics points toward one overarching shift: the optimization target has moved from search result appearance to selection in a synthesis process. Brands that continue to measure success exclusively by keyword rankings and organic traffic will have an increasingly incomplete view of their actual visibility. A page can rank on the first page of Google, be retrieved by ChatGPT, and still reach no one — because it lost the selection competition to a page that covered the topic more specifically, structured its answer more clearly, or appeared in the fan-out query that ChatGPT actually relied on to build the response. Understanding the gap between retrieval and citation, and the mechanics that determine which side of that gap a page falls on, is the foundational intelligence for any content program operating in the current search environment.
About ALM Corp
ALM Corp is a data-driven digital marketing and AI search strategy firm that helps brands build measurable visibility across both traditional search and AI-generated answers. As retrieval fan-out, citation selection, and answer engine dynamics reshape how audiences discover brands online, ALM Corp works with clients to audit their current citation performance, identify content gaps across fan-out query landscapes, and build structured content strategies that earn selection — not just retrieval. Whether your brand needs to improve its position in ChatGPT answers, Google AI Overviews, or Perplexity citations, ALM Corp’s team applies the same evidence-based approach documented in the research covered here: understanding where your content currently sits in the retrieval-to-citation pipeline, and systematically closing the gaps. Learn more at almcorp.com.
