Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

How AI ranks content: retrieval, passages and citations

Updated · 7 min read

Short answer AI search keeps no separate leaderboard. It runs ordinary searches, often several for one question, reads passages from the pages those searches return, writes an answer and links the pages it used. To be picked, a page must be crawlable and indexed, rank for the sub-questions, and state its facts plainly in text.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

How does AI rank content?

In four steps, and the middle two are ordinary search. Google says its AI Overviews and AI Mode sit on top of the search engine you already know:

"our generative AI features on Google Search are rooted in our core Search ranking and quality systems." (Google Search Central: Optimizing for generative AI search)
StepWhat the AI doesWhat decides whether your page is used
1. Split the questionTurns one question into several related searchesWhether you answer the smaller questions buyers ask
2. Search an indexPulls pages from a search index: Google's for AI Overviews and AI Mode, Bing's for Copilot; ChatGPT and Perplexity send their own search crawlersThe crawler can reach the page, it's indexed, and it ranks
3. Pick passagesReads those pages and keeps the parts that answer each smaller questionA section that states the fact in plain text
4. Write and citeThe language model writes the answer and links the pages it drew onWhether your passage adds something the other sources don't

Google calls step 1 "query fan-out". For "how to fix a lawn that's full of weeds", its example extra searches are "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn" (Google Search Central: Optimizing for generative AI search). Step 2 is what Google calls retrieval-augmented generation, or grounding: it relies on "our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index."

So a page that isn't indexed, or doesn't rank for any of those searches, never gets read. If that's your problem, start with website not showing on Google or pages not indexed.

Does AI match keywords or meaning?

Both. Search systems use two kinds of matching, and Google's regular search already uses both.

MethodHow it worksGood at
Keyword matching (TF-IDF, BM25)Scores a passage by how often the search words appear in it, giving rare words more weight and adjusting for lengthNames, model numbers, towns, prices
Meaning matching (embeddings)Turns the question and each passage into a long list of numbers; close lists mean close meaning"Furnace blowing cold air" finds "heater not putting out heat"

OpenAI's developer guide explains the second one:

"An embedding is a vector (list) of floating point numbers. The distance between two vectors measures their relatedness." (OpenAI API docs: Embeddings)

Meaning matching finds answers keyword scoring misses. In a 2020 study, a meaning-based retriever put a passage with the answer in its top 20 results for 9 to 19 percentage points more test questions than BM25 (Karpukhin et al.). No AI search product publishes which methods it uses or how it weighs them, so treat any exact formula you read as a guess.

"Old SEO is keywords, AI is meaning" is wrong. Google's regular ranking already uses AI language systems such as RankBrain, BERT and neural matching (Google Search Central: ranking systems guide). For AI answers, Google adds:

"AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words." (Google Search Central: Optimizing for generative AI search)

In practice: use the words your customers use, once, where they fit. Repeating variations doesn't help; see keyword density.

Does AI rank whole pages or passages?

Pages get you retrieved; passages get you quoted. Google's regular search already looks at passages:

"Passage ranking is an AI system we use to identify individual sections or "passages" of a web page to better understand how relevant a page is to a search." (Google Search Central: ranking systems guide)

Microsoft describes Copilot the same way: it breaks pages "into smaller, structured pieces that can be evaluated for authority and relevance", then assembles the answer from several sources (Microsoft Advertising Blog, October 2025).

Google and Microsoft differ on what you should do about it:

TopicGoogleMicrosoft (Bing, Copilot)
Splitting pages into small chunks"There's no requirement to break your content into tiny pieces for AI to better understand it."Avoid "long walls of text"; use clear headings, question-and-answer pairs, lists and tables
Schema markupNot required; "no special schema.org markup you need to add"Says schema helps AI systems understand content
Key facts in images or PDFsMake sure "important content is available in textual form"Facts only in images are read less accurately; PDFs lack the headings and structure HTML gives

One habit satisfies both: a heading that names the question, an answer in the first sentence or two below it, then the detail. Don't build a separate page for every variation of a question. Google says doing that "primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy."

How does AI choose which sources to cite?

No platform publishes a formula. What they do publish points the same way:

  • Rank first. Retrieval starts from the search index, so the usual signals count. Google says one sign of quality is whether "other prominent websites link or refer to the content" (Google Search: How Search works). See E-E-A-T for a small business.
  • Say something others don't. Google: "Our AI systems take a look at a variety of sources, so it can be helpful to have a unique viewpoint that stands out." It contrasts a first-hand review with a summary that restates what's already online.
  • Show evidence. Bing: "Examples, data, and cited sources help build trust when content is reused in AI-generated answers."
  • Stay current where it matters. Google runs "query deserves freshness" systems for searches where people expect recent content; Bing says regular updates help AI use the current version of a page.
  • Skip fake buzz. Google says chasing inauthentic "mentions" across the web "isn't as helpful as it might seem."

One lab study tested this. Its answer engine had GPT-3.5 write answers from the top 5 Google results. Adding citations, credible quotes or statistics to one source raised its share of the answer by 30 to 40%; keyword stuffing gave "little to no improvement" (Aggarwal et al., KDD 2024). It's one study on 2023-era engines: a direction, not a promise.

How can you see whether AI uses your pages?

WhereWhat it shows
Google Search Console, generative AI performance reportsHow often your pages appeared in AI Overviews, AI Mode and Discover's AI features, by page, country and device. Available to all sites worldwide since August 31, 2026
Bing Webmaster Tools, AI Performance (public preview since February 2026)Citations in Copilot and Bing's AI summaries, which pages were cited, and "grounding queries": a sample of the phrases the AI searched with when it pulled your content
ChatGPT and PerplexityNo owner report. Allow their search crawlers (OAI-SearchBot, PerplexityBot), ask the questions buyers ask, and watch referral visits in your analytics

Bing's grounding queries are step 1 made visible: the smaller searches that led the AI to your page. Be wary of tools that claim more. Google: "No third-party tool has access to our internal ranking or AI systems." To test ChatGPT by hand, use the 20-question test.

Example

Illustrative cases with invented numbers.

A furnace repair company in Ohio (US dollars)

A homeowner asks: "My furnace is blowing cold air. Can I fix it, and what does a repair cost?" The AI may search causes, safe checks and repair prices. The company's pricing page ranks locally and has a heading "What does furnace repair cost?" followed by "Most repairs run $150 to $600; a new igniter is about $250." That sentence can be lifted and cited. "Affordable pricing, call today" can't.

A bookkeeper in Ottawa (Canadian dollars)

Prices live only in a PDF and a banner image. Moving one line into page text, "Monthly bookkeeping from C$300 for up to 100 transactions", gives every engine a passage it can read and quote.

A dental clinic missing from ChatGPT

The clinic's hosting firewall blocks OpenAI's crawler. OpenAI says sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers." The clinic allows the crawler in robots.txt and at the firewall; OpenAI says robots.txt changes take about 24 hours to register. See blocking AI crawlers.

FAQ

Does SEO use machine learning?

Yes. Google lists AI systems in its regular ranking, including RankBrain, BERT, neural matching and passage ranking (Google Search Central: ranking systems guide).

Does an AI answer just cite the top Google result?

Not always. It runs several searches, so a page ranking for one of them can be cited. Google says AI shows more diverse links.

Can I pay to be cited in an AI answer?

Not in Google's results. Google says "nobody can pay us" to rank; ads are labeled and separate. See ads in AI Overviews.

Do I need an llms.txt file or special schema for AI?

Not for Google. It needs no "AI text files" or special schema.org markup and ignores llms.txt. See local business schema markup.

Different for you?

Keep learning

NextGenerative engine optimization (GEO): what it is and how it worksWhat GEO means, where the term came from, how ChatGPT and Google's AI choose the pages they cite, and why it now matters to a small business.

Do it yourself

Does ChatGPT recommend your business? A 20-question testA free test any owner can run to see whether ChatGPT names their business, which competitors it names instead, and why.

Ask us one question

One line is enough. We reply by email.

Sources

  1. Google Search Central: Optimizing for generative AI search
  2. Google Search Central: AI features and your website
  3. Google Search Central: A guide to Google Search ranking systems
  4. Google Search: How Search works, ranking results
  5. Google Search Central Blog (June 3, 2026): Introducing Search Generative AI performance reports in Search Console
  6. Microsoft Advertising Blog (October 8, 2025): Optimizing your content for inclusion in AI search answers
  7. Bing Webmaster Blog (February 10, 2026): Introducing AI Performance in Bing Webmaster Tools
  8. OpenAI: Overview of OpenAI crawlers
  9. OpenAI API docs: Embeddings
  10. Perplexity Help Center: How does Perplexity work?
  11. Perplexity: Perplexity crawlers
  12. Karpukhin et al. (2020): Dense Passage Retrieval for Open-Domain Question Answering
  13. Aggarwal et al. (KDD 2024): GEO: Generative Engine Optimization