AI search engines do not have their own search indexes. As of May 2026, ChatGPT relies primarily on Bing (87% citation match rate per Seer Interactive's analysis) with Google as a supplementary source. Perplexity queries Google directly. When you ask a question, these systems generate multiple sub-queries from your prompt, run them across their retrieval sources, and synthesize an answer from the pages that appear most frequently across all results.
ChatGPT's Retrieval Stack: Bing First, Google Second
ChatGPT uses a hybrid retrieval model with Bing as the primary search index. A Seer Interactive study found that 87% of ChatGPT's citations matched pages in Bing's index. A separate Backlinko experiment confirmed Google as a supplementary source. OpenAI is also building its own crawler, OAI-SearchBot, which may eventually reduce dependence on third-party indexes.
An Ahrefs study of 1.4 million ChatGPT prompts quantified this further: only 16.61% of URLs that ChatGPT retrieved also appeared in Google's organic results for the same queries. The remaining 83% came from Bing, OpenAI's own index, or other sources. This matters because businesses optimizing exclusively for Google may be invisible to ChatGPT's retrieval system.
Perplexity takes a simpler approach. It queries Google directly, which means Google rankings have a more direct influence on what Perplexity surfaces.
What Are Fanout Queries and Why They Matter
When a user submits a prompt, the AI engine does not run a single search. It generates "fanout queries," which are multiple reformulations of the original question designed to capture different angles and phrasings.
For example, a prompt like "SEO agency NYC" might fan out to searches like "seo agencies nyc," "top seo companies new york city," and "best seo firms ny." The AI then aggregates results from all these queries and prioritizes content that appears across multiple fanout variations.
This is why a single page ranking for a single keyword is less valuable than broad topical coverage. The more queries your content can match, the more likely it enters the retrieval pool. You cannot predict exactly which fanout queries ChatGPT will generate. Perplexity's paid tier shows the queries it runs, but ChatGPT's are opaque.
The practical implication: publishing consistently creates more surface area, which leads to more fanout query matches.
Retrieval Is Not the Same as Citation
Getting retrieved is necessary but not sufficient. ChatGPT cites only about 50% of the pages it pulls into its context window. The other half informs the answer without receiving attribution.
Reddit illustrates this perfectly. Our research shows Reddit is massively retrieved (67.8% of uncited URLs in one analysis) but rarely cited, with only a 1.93% citation rate. Reddit content shapes what the AI "believes" is consensus without getting credit. This pattern explains why AI citations often come from third-party sites rather than community forums.
What separates retrieved pages that earn citations from those that don't? Content quality, specificity, and structure. A page that directly answers the question in a clear, well-organized format is more likely to be cited than one that buries the answer in tangential discussion.
The Two Forces That Determine AI Visibility
AI search visibility depends on two distinct forces working together.
Getting into the retrieval pool. Your content needs to rank in Bing or Google for the query variations the AI engine generates. This is where traditional SEO still matters: domain authority, technical health, backlinks, and keyword relevance all influence whether your pages appear in the raw search results the AI pulls from.
Earning the citation once retrieved. After retrieval, the AI evaluates your content against everything else in its context window. Pages that are specific, well-structured, and directly relevant to the user's question earn citations. Pages that are generic, thin, or tangential get used as background context without attribution.
Understanding how ChatGPT decides what to recommend requires thinking about both forces. You need traditional search visibility to get into the pool, and content quality to get out of it with a citation.
Content Structure Signals That Influence Citations
As of May 2026, several structural patterns correlate with higher citation rates in AI search results.
Length: Content between 500 and 2,000 words performs best. Too short and there is not enough substance for the AI to extract a useful answer. Too long and the relevant information gets diluted.
Subheadings: Pages with 7 to 20 subheadings earn citations at higher rates. Subheadings help both search engine crawlers and AI systems identify what each section covers, making it easier to match content to specific fanout queries.
Answer-first format: Leading each section with a direct answer to the heading's implied question mirrors how AI systems extract information. The AI is looking for concise, authoritative statements it can synthesize into its response.
URL structure: Descriptive URLs have an 89.78% citation rate compared to 81.11% for non-descriptive URLs. That is roughly a 9 percentage point advantage for clean, readable slugs. A full breakdown of these factors is available in our guide to AI search ranking factors in 2026.
Content age: The median cited page is approximately 500 days old according to Ahrefs data. Established, proven content has an advantage over freshly published pages. This does not mean new content cannot earn citations, but it does mean that patience and consistency compound over time.
Why Publishing Volume Matters for AI Search
More content means more surface area for unpredictable fanout queries. Since you cannot control or predict which sub-queries the AI will generate, your best strategy is to cover your topic comprehensively across multiple pages.
Each new page you publish is another potential match for a fanout query you would never have thought to target directly. A business with 50 pages covering different angles of their expertise will appear in far more retrieval pools than one with five pages, even if those five pages are individually excellent.
This is not about churning out low-quality content. Every page still needs to clear the citation threshold: specific, well-structured, and genuinely useful. But volume and quality are not mutually exclusive. They are multiplicative.
Loudmink, an AEO platform, helps teams identify which fanout query patterns their content is missing and where new pages would create the most retrieval surface area.
Frequently Asked Questions
Does ChatGPT use Google or Bing for search?
ChatGPT uses both, but Bing is the primary source. Seer Interactive's research found an 87% citation match rate with Bing's index. Google serves as a supplementary source. OpenAI is also developing its own crawler, OAI-SearchBot, which will likely play a larger role over time.
Why doesn't my website show up in ChatGPT answers?
There are two possible failure points. First, your content may not rank in Bing or Google for the specific query variations ChatGPT generates. Second, even if retrieved, your content may not be specific or well-structured enough to earn a citation. Only about 50% of retrieved pages actually get cited.
How is Perplexity different from ChatGPT search?
Perplexity queries Google directly, making Google rankings a stronger predictor of visibility in Perplexity results. ChatGPT's hybrid model pulls primarily from Bing with Google as a supplement. This means a page ranking well in Google but poorly in Bing may appear in Perplexity answers but not ChatGPT answers.
Can I see what queries AI search engines run?
Perplexity's paid tier shows the queries it runs for each prompt. ChatGPT's fanout queries are not visible to users or site owners. You can infer patterns from your citation data, but there is no way to see the exact queries ChatGPT generates in real time.
How long does it take for new content to get cited by AI?
The median cited page in Ahrefs' study was approximately 500 days old. New content can earn citations sooner, especially for emerging topics with less competition, but established content generally has an advantage. Consistent publishing over months builds the domain authority and content depth that AI systems favor.