Passage Ranking

TL;DR — Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. lets Google score one part of a page on its own. So if you’ve got one great answer buried in the middle of a long, sprawling page, Google can now surface that section even when the page as a whole isn’t tightly about that topic. Despite the original name, Google still indexes your whole page — only the ranking changed. There’s nothing special you need to do.

What passage ranking is

Google has described passage-based ranking as helping identify relevant individual passages within pages for particular queries. Evidence for this claim Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Scope: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confidence: high · Verified: 2026-07-14 Google: Search On 2020 It is a ranking capability, not a separate passage indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. or a special markup feature publishers can enable. Evidence for this claim Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Scope: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confidence: high · Verified: 2026-07-14 Google: Ranking systems guide

Normally, Google looks at a whole web page to decide how relevant it is to a search. Passage ranking lets Google zoom in and judge individual sections of a page — what it calls “passages” — separately.

Why does that help? Imagine a long forum thread, or a rambling personal blog post, that happens to contain one perfect answer to a very specific question. The page overall is about ten different things, so it never ranked well. With passage ranking, Google can recognize “this one section here is exactly what this searcher wants” and rank the page for that query on the strength of that passage.

Google announced it in October 2020 and turned it on for US English searches in February 2021. It said it would affect about 7% of all searches.

The naming confusion (it’s “ranking,” not “indexing”)

Google first called this “passage indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.,” and that name caused a lot of confusion — so the company corrected it. Google still stores (indexes) your whole page, exactly like before. What changed is how it ranks: a single page can now get several different relevance scores, one for each passage, depending on the search. As Google’s Danny Sullivan said at the time: “We are not indexing passages. Period.”

What you should do about it

Almost nothing — and that’s the official line too. There’s no setting, no special markup, no trick. The advice is the same boring, good advice it always is:

• Write clear, well-organized content.
• Use real headings so each section stands on its own.
• Answer specific questions thoroughly in self-contained chunks.

Don’t go chopping your content into tiny artificial pieces to “help” passage ranking — that’s not a thing, and Google has explicitly warned against people selling you on it.

Want the deeper version — the multiple-scores mechanism, how it connects to AI Overviews, and what Bing’s been doing — switch to the Advanced tab.

TL;DR — Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. is an AI ranking system (announced Oct 20, 2020; live US English Feb 10, 2021) that scores individual passages of a page so one document can earn multiple relevance scores against different queries. The name was corrected from “passage indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.” because indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. didn’t change — pages are still indexed whole; only the scoring granularity changed. It targets long, messy pages where a strong answer is diluted, affects ~7% of queries, and is a separate system from scroll-to-text fragments and featured snippets. It’s the direct conceptual ancestor of passage-level retrieval in AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and Copilot. SEO action: well-structured, self-contained sections — nothing artificial.

The core innovation: one page, many scores

Public explanations establish passage-level relevance, but not a complete scoring architecture or a guaranteed number of scores per page. Evidence for this claim Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Scope: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confidence: high · Verified: 2026-07-14 Google: Search On 2020 Avoid presenting illustrative mechanics as disclosed source code. Evidence for this claim Official or primary documentation supporting the adjacent article claim, with scope limited to the source's published description. Scope: No ranking guarantee or undisclosed system mechanics are inferred beyond the cited source. Confidence: high · Verified: 2026-07-14 Google: Ranking systems guide

Here’s the mechanism that matters. Before passage ranking, a page got one shot at relevance for a query — Google evaluated the whole document. Passage ranking lets a page receive multiple relevance scores, one per passage, each measured against different queries. Martin Splitt described it cleanly as “one document that can be scored with multiple scores according to the queries.”

That’s the whole idea. It’s not a new index, not a new crawl, not a new storage format. It’s the ranking system getting more granular about which part of a page answers a given query. As Splitt put it, “it’s just us getting better at more granularly understanding the content of a page.”

Why it exists: rescuing buried answers

Google built this for a specific failure mode — long pages where a great answer is diluted by everything around it. Danny Sullivan, October 2020: “Typically, we evaluate all content on a web page to determine if it is relevant to a query. But sometimes web pages can be very long, or on multiple topics, which might dilute how parts of a page are relevant for particular queries… With our new technology, we’ll be able to better identify and understand key passages on a web page.”

Crucially, this is upside-only and narrow. Splitt: “We are only improving rankings for pages that are currently having trouble.” It’s not re-rankingReranking is the second stage of a retrieval pipeline: after a cheap, broad first pass pulls a candidate set of documents or passages, a slower, more precise model re-scores and reorders that shortlist by true relevance before the results are served or handed to an LLM. the whole web; it’s giving a lift to pages whose best content was previously buried. And a passage can be tiny — Splitt: “It can be really short, it can be as short as a few words, it can be as long as a paragraph.” It can even override a title match: he noted a page with a good answer in the body can be more relevant than a page with the query in its title.

”Ranking,” not “indexing” — and why the distinction is load-bearing

Google initially marketed this as “passage indexing,” and both Splitt and Sullivan pushed back hard on the name. Splitt: “While I’m not super excited about us framing it… originally as passage indexing, it is actually a ranking change.” And the flat denial from Sullivan: “But… We are not indexing passages. Period.”

The distinction isn’t pedantry — it changes what you do about it:

• Indexing didn’t change. Google still indexes the whole page as one document. Splitt: “It does not mean that we are storing them differently.”
• Ranking changed. The same stored document now gets scored at passage granularity against queries.

So you can’t “submit passages,” you can’t opt a passage out, and there’s no passage in Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results. you can inspect. Splitt on opting out: “That’s like saying I don’t want to rank for these five paragraphs of my page.” If you find someone selling “passage indexing optimization,” Splitt’s read still applies: “There will probably be some people who will try to capitalize on this. But I wouldn’t fall for it… It’s not for you who know what they are doing.”

Three things it is NOT

These get conflated constantly:

• Not scroll-to-text / text fragments. Passage ranking and scroll-to-text (#:~:text= deep links) are two separate, unrelated systems. Google confirmed this directly. They interact in the AI era — see below — but they aren’t the same feature.
• Not featured snippets. A different system. Passage ranking influences which pages rank; featured snippets are about which answer gets extracted to the top.
• Not artificial chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking.. Splitt was explicit that you don’t need to change your site, pages, articles, or markup. Don’t carve content into fake fragments.

Timeline and scale

• October 20, 2020 — announced at Google’s “Search On” event.
• February 10, 2021 — live for English-language queries in the US (confirmed by SearchLiaison the next afternoon).
• ~7% of search queries across all languages — Google’s own stated impact once fully rolled out globally. Early industry tracking (Search Engine Roundtable) reported minimal observable impact at launch, which fits a narrow, buried-answers-only system.

The AI-era extension: the same idea, everywhere now

This is the part most SEO coverage misses. Passage-level relevance is no longer a niche 7% lever — it’s the default unit of retrieval in AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity.. AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and Copilot don’t rank whole documents to build an answer; they retrieve and score passages.

• AI Overviews use passage-level retrieval to assemble answers, and frequently link to the exact source passage using text-fragment URLs (#:~:text=) rather than the page as a whole. (This is where passage ranking and scroll-to-text — two unrelated systems — finally meet in practice.) Search Engine Land’s AI Overviews guide notes Gemini “can also use passage indexing to incorporate specific sections of a website to fill out its AI Overview response.”
• Copilot / Bing runs a three-step loop: retrieve ~15–20 high-confidence URLs from the Bing index, chunk those pages into passages and score them on relevance/authority/freshness, then validate and cite. The chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking. step is passage ranking by another name. (See chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking., vector searchVector search finds content by comparing the meaning of a query against stored content as numerical vectors, retrieving the closest ones in a high-dimensional embedding space. At scale it uses approximate nearest neighbor (ANN) algorithms — not exact comparison — to search billions of vectors in milliseconds., embeddingsEmbeddings are dense numerical vectors — lists of floating-point numbers — that represent the meaning of text in a high-dimensional space. Semantically similar content lands close together, so search and AI systems can match by meaning, not just keywords., and RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. for the full retrieval stack.)

Bing’s foundational role (the unsung half of this story)

Here’s something almost no SEO article mentions: Microsoft, via Bing, basically funded the modern field of passage ranking. Microsoft Research built the MS MARCO (Microsoft MAchine Reading COmprehension) dataset — 8,841,823 passages extracted from real web documents, with over a million training queries sampled and anonymized from Bing’s query logs. It’s the benchmark for neural passage ranking research. The foundational “Passage Re-ranking with BERTSemantic search is meaning-based retrieval — matching what a user means, not just the words they typed. Search engines detect entities, expand synonyms, infer intent, and rank by conceptual relevance, which is why keyword stuffing lost its power and topical depth gained it.” work that beat prior systems by ~27% relative MRR@10 on that benchmark? Trained and measured on Bing-derived data.

And Microsoft has kept pushing the passage-as-unit idea into the AI era. From the June 2026 Web IQ announcement by Knut Risvik: “Web IQ does not just return documents; it returns passages and structured evidence objects,” and “by operating at the level of passages, we can concentrate useful signal while eliminating irrelevant context.” Microsoft’s broader framing (May 2026): “the unit of value shifts from documents to groundable information.” That is passage ranking, taken to its logical conclusion — the document stops being the unit; the passage is.

What this actually means for your SEO

Nothing exotic, but the through-line is real and worth internalizing:

1. Write self-contained sections. A passage that answers a question completely, on its own, is more retrievable — both for the 7% classic passage-ranking lever and for AI passage retrieval. (AI Overview studies put the sweet spot around 134–167 words per cited passage, with semantic completeness the single strongest predictor of citation.)
2. Use real headings. They give Google and AI retrievers clean passage boundaries.
3. Don’t fake it. No artificial fragmentation, no “passage markup,” no opting passages in or out. There isn’t a lever there.
4. It’s normal SEO. Danny Sullivan’s line generalizes: good SEO is good GEOGenerative Engine Optimization (GEO) is the practice of optimizing content and brand presence so AI-powered search engines and assistants — Google AI Overviews, ChatGPT, Perplexity — cite, recommend, or mention you when generating answers. Google's position is that it's still SEO./AEO. The same well-structured content wins passage ranking, featured snippets, and AI citations.

AI summary

A condensed take on the Advanced version:

• Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. = one page, many scores. A single document can earn multiple relevance scores, one per passage, against different queries. Splitt: “one document that can be scored with multiple scores according to the queries.”
• It’s ranking, not indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.. Google still indexes whole pages; only the scoring granularity changed. Sullivan: “We are not indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. passages. Period.”
• Built to rescue buried answers in long, messy pages — upside-only (“we are only improving rankings for pages that are currently having trouble”).
• Timeline: announced Oct 20, 2020; live US English Feb 10, 2021; ~7% of queries; minimal observed impact at launch.
• Not scroll-to-text fragments, not featured snippets, not a reason to chop up content artificially — three separate, commonly-confused things.
• AI-era extension: AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and Copilot retrieve at the passage level (and AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. link to passages via #:~:text= text fragments). The chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking. step in RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. is passage ranking by another name.
• Bing’s foundational role: MS MARCO (8.8M Bing-derived passages) seeded the whole neural-passage-ranking field; Web IQ now returns “passages and structured evidence objects.”
• SEO action: clear, self-contained sections with real headings. That’s it.

Official documentation

Primary-source documentation from the search engines.

Google

• Google’s Guide to Ranking Systems — the official entry: “Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. is an AI system we use to identify individual sections or ‘passages’ of a web page…”
• In-Depth Guide to How Google Search Works — where passage ranking sits in the crawl → indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. → serve pipeline.

Bing / Microsoft

• Introducing Microsoft Web IQ (June 2026) — passage-and-evidence-object APIs; the clearest Microsoft statement on operating at the passage level.
• Evolving role of the index (May 2026) — “the unit of value shifts from documents to groundable information.”
• Introducing Deep Search (December 2023) — GPT-4 query expansion over ~10x more pages.
• AI Performance in Bing Webmaster Tools (February 2026) — “groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries” and citation data.
• MS MARCO dataset — 8.8M Bing-derived passages; the benchmark for neural passage ranking.

Quotes from the source

On-the-record statements from Google and Microsoft. Each link is a deep link that jumps to the quoted passage on the source page where available.

Google — the official definition

• “Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. 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. Jump to quote
• “This change doesn’t mean we’re indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. individual passages independently of pages. We’re still indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. pages and considering info about entire pages for ranking.” — Google Search Central, Ranking Systems Guide. Jump to quote

Danny Sullivan, Google Public Search Liaison (October 2020 announcement)

• “But… We are not indexing passages. Period.” — the on-the-record correction to the “passage indexing” framing.
• Paraphrasing his announcement thread: long or multi-topic pages can dilute how relevant individual parts are for particular queries; the new technology lets Google better identify and understand key passages — and there’s nothing special creators need to do. Coverage of the announcement (Search Engine Land)

Martin Splitt, Google (via Search Engine Journal, Roger Montti)

• “While I’m not super excited about us framing it… originally as passage indexing, it is actually a ranking change.”
• “It’s just us getting better at more granularly understanding the content of a page.”
• “One document that can be scored with multiple scores according to the queries.”
• “We are only improving rankings for pages that are currently having trouble.”
• “It can be really short, it can be as short as a few words, it can be as long as a paragraph.”
• “It does not mean that we are storing them differently.” Read the Q&A

John Mueller, Google (via Search Engine Roundtable)

• Paraphrased: passage ranking is recognizing that a big page has a part that’s particularly relevant to an incoming query, so Google focuses on that part of the page. Read the coverage

Knut Risvik, Microsoft (Web IQ, June 2026)

• “Web IQ does not just return documents; it returns passages and structured evidence objects.”
• “By operating at the level of passages, we can concentrate useful signal while eliminating irrelevant context.” Jump to quote

Passage ranking — cheat sheet

The one-liner: one page can earn multiple relevance scores, one per passage, for different queries. Pages are still indexedStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. whole; only the ranking got more granular.

IndexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. vs. ranking — what actually changed

Don’t confuse these three separate systems

Fast facts

• Announced Oct 20, 2020; live US English Feb 10, 2021.
• Google’s impact claim: ~7% of queries across all languages.
• Observed launch impact: minimal (fits a narrow, buried-answers system).
• AI sweet spot for a cited passage: ~134–167 words, self-contained.
• Nothing special to do — no markup, no opting in/out, no artificial chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking..

The mental models

1. The document is no longer the unit. Classic ranking scores a whole page. Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. scores sections. AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. takes this all the way — Microsoft: “the unit of value shifts from documents to groundable information.” When you write, think in self-contained answer units, not just “pages.”

2. Ranking ≠ indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.. Storage didn’t change; scoring granularity did. If a claim about passage ranking implies Google stores or submits passages separately, it’s wrong. “We are not indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. passages. Period.”

3. Three separate systems, often conflated. Passage ranking (scoring sections) ≠ scroll-to-text fragments (#:~:text= highlighting) ≠ featured snippets (top-of-results extraction). They can meet — AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. use text fragments to cite the passage they ranked — but they’re distinct mechanisms.

4. Passage ranking → RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights., same idea. The chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking.-and-scoring step inside RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. — and inside Copilot and AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. — is passage ranking generalized. Optimizing for clear passages is one move that pays off across classic search, featured snippets, and AI citationsAn AI citation is the visible source link an AI answer engine shows next to its generated text — the clickable reference that credits the web page it used. A citation's presence is a separate thing from whether the cited page actually supports the statement, and from being retrieved (read behind the scenes) or merely mentioned (named without a link); citation is driven more by brand mentions and being retrievable than by traditional ranking. at once.

5. The action rule. There is no passage lever. The only move is: well-organized, self-contained sections with real headings. If someone’s selling “passage optimization” beyond that, Splitt’s warning applies — “I wouldn’t fall for it.”

Passage-ranking myths to avoid

Creating a separate URL for every paragraph

Passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. scores sections within an indexedStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. page; it does not require a page per answer fragment. Split content only when a topic deserves an independent page for users and intent.

Calling passage ranking passage indexing

Google indexes the page and applies more granular ranking signals to its passages. Keep the distinction clear so stakeholders do not invent a new indexability project.

Writing isolated snippets with no page-level coherence

Self-contained sections still belong in a useful, organized page. Give each section a descriptive heading and enough local context without repeating the entire article.

Test yourself: Passage ranking

Resources worth your time

From around the industry

The announcement and the naming correction

• How Google indexes passages of a page and what it means for SEO — Search Engine Land’s launch explainer.
• Could Google passage indexing be leveraging BERT? — Search Engine Land explores the BERTSemantic search is meaning-based retrieval — matching what a user means, not just the words they typed. Search engines detect entities, expand synonyms, infer intent, and rank by conceptual relevance, which is why keyword stuffing lost its power and topical depth gained it. connection at launch.
• Google Passage Ranking, Not Passage Indexing — Barry Schwartz on the corrected terminology.
• What Is Google Passage Ranking: 16 Key Points — Roger Montti’s Splitt Q&A; the most verbatim source.
• Google Answers Question About Passages SEO — Search Engine Journal on what the feature means for optimization.

Launch and impact

• Google passage ranking now live in US English — the Feb 2021 go-live.
• Google Passage Based Ranking Causing Minimal Impact — early observed effect.
• Google’s John Mueller Explains Passage Ranking — Mueller’s plain-English description.
• Passage Ranking & Scroll To Text Are Unrelated — the confirmation that these are separate systems.
• Reminder: Passage Ranking Helps Google Understand Messy Content — Barry Schwartz revisiting the “messy content” use case.

The technical and AI-era foundations

• Passage Re-ranking with BERT (arXiv) — the foundational cross-encoder rerankingReranking is the second stage of a retrieval pipeline: after a cheap, broad first pass pulls a candidate set of documents or passages, a slower, more precise model re-scores and reorders that shortlist by true relevance before the results are served or handed to an LLM. paper (~27% relative MRR@10 on MS MARCO).
• MS MARCO Datasets — Bing’s 8.8M-passage benchmark that seeded the field.
• AI Overviews optimization guide — Gemini using passage-level selection to fill AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index..
• Google Publishes Guide To Current & Retired Ranking Systems — Search Engine Journal covering the official ranking systems documentation.

My related writing

• The Beginner’s Guide to Technical SEO — where ranking fits in the bigger picture.

My speaking

• How Search Works (SlideShare) — my walkthrough of crawlingCrawling is how search engines use automated bots (like Googlebot and Bingbot) to discover URLs and download pages. A page has to be crawlable to be indexed, but crawling on its own isn't a ranking factor., renderingTurning HTML, CSS, and JavaScript into the final visual page and DOM., indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., and ranking. (Standing disclaimer: “This is my understanding of systems… not going to be 100% complete or accurate.”)

Stats worth citing

• ~7% of search queries across all languages — Google’s own stated impact for passage rankingPassage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed. when fully rolled out globally (October 2020 announcement). Coverage
• 8,841,823 passages in MS MARCO — extracted from real web documents, with 1M+ queries sampled from Bing’s logs; the benchmark behind modern passage ranking. Source
• ~27% relative improvement in MRR@10 — BERTSemantic search is meaning-based retrieval — matching what a user means, not just the words they typed. Search engines detect entities, expand synonyms, infer intent, and rank by conceptual relevance, which is why keyword stuffing lost its power and topical depth gained it.-based passage rerankingVibe ranking is informal practitioner shorthand for LLM-based reranking — using a large language model to reorder retrieved search candidates by holistic, semantic judgment (relevance, clarity, expertise, completeness) rather than keyword overlap or vector similarity alone. It is not an official Google or Bing term. over prior neural systems on MS MARCO, the result that mainstreamed cross-encoder rerankingReranking is the second stage of a retrieval pipeline: after a cheap, broad first pass pulls a candidate set of documents or passages, a slower, more precise model re-scores and reorders that shortlist by true relevance before the results are served or handed to an LLM.. Source
• 134–167 words — the optimal length of a passage cited in AIThree distinct states of AI visibility: retrieved (an AI fetched your page as source material), mentioned (your brand appears in the answer text), and cited (your URL is linked as a source). They don't always happen together, and each is measured with a different tool. Overviews, with 94% of selected content in the 100–300 word range and semantic completeness (r=0.87) the strongest predictor of citation. Source
• Sub-165ms P95 latency — Microsoft Web IQ’s passage-and-evidence retrieval fabric (June 2026), the AI-era realization of passage-level operations. Source

Passage Ranking

Passage ranking is a Google AI system that scores individual sections ('passages') of a page so a single page can earn multiple relevance scores for different queries. Google still indexes whole pages — only the ranking changed.

Related: Chunking, Vector Search, Retrieval-Augmented Generation (RAG)

Passage Ranking

Passage ranking is an AI ranking system Google announced in October 2020 and rolled out for US English in February 2021. Google described the system as a way to understand the relevance of specific passages within a page.

The name matters. Google first called it “passage indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.,” then clarified that it still indexes whole pages; what changed is the ranking system’s ability to evaluate passages. The practical innovation is that a useful section buried deep in a long page can surface even when the page overall isn’t tightly focused on that query.

A few things people confuse:

• Passage ranking ≠ passage indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.. Storage didn’t change; only the granularity of scoring did.
• Passage ranking ≠ scroll-to-text / featured snippets. Google confirmed these are separate, unrelated systems.
• There’s nothing special to do. Google projected that its passage-understanding technology would improve about 7% of queries across languages as it rolled out; the announcement prescribed no special markup for creators.

Passage ranking is the conceptual ancestor of how AI search works today: AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and Copilot retrieve at the passage level too, chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking. pages into passages and scoring those — the same idea that powers RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights., embeddingsEmbeddings are dense numerical vectors — lists of floating-point numbers — that represent the meaning of text in a high-dimensional space. Semantically similar content lands close together, so search and AI systems can match by meaning, not just keywords., and vector searchVector search finds content by comparing the meaning of a query against stored content as numerical vectors, retrieving the closest ones in a high-dimensional embedding space. At scale it uses approximate nearest neighbor (ANN) algorithms — not exact comparison — to search billions of vectors in milliseconds..

Related: Chunking, Vector Search, Retrieval-Augmented Generation (RAG)