Playbooks
Passage Ranking: How Google Ranks One Section of Your Page (2026)
Google confirmed in 2020 that it ranks passages, not just pages. In 2026, that same idea is the reason AI Overviews can lift one paragraph out of your content and ignore the rest.
Passage ranking is Google’s ability to rank one specific section of a page for a search query, separately from how the page ranks as a whole. Google confirmed it in October 2020, and the same underlying idea now decides which paragraph of your content an AI Overview lifts out and cites.
Key takeaways
- Google announced passage ranking in October 2020, saying it would improve roughly 7% of search queries in all languages at launch.
- In 2026, Google described a related technique called chunking, splitting a page into passages that AI Overviews and AI Mode evaluate and retrieve semi-independently.
- A single page can win one query on one section and lose the next query on a different section of the same page. Thoroughness alone no longer guarantees a win.
- A rankable passage names its question in the heading, answers it in the first sentence or two, covers one idea, and makes sense with zero context from the rest of the page.
- The fix for a page that loses specific queries despite being comprehensive is almost always tighter section boundaries, not more content.
The 2020 announcement, and what it actually said
Stop thinking of your page as one thing that ranks. Google stopped in 2020.
At its Search On event that October, Google’s Prabhakar Raghavan announced a new system for understanding passages within a page: “By understanding passages in addition to the relevancy of the overall page, we can find that needle-in-the-haystack information you’re looking for.” Google said the change would “improve 7 percent of search queries across all languages” as it rolled out globally, specifically queries where the right answer exists buried inside an otherwise relevant page, three paragraphs deep, not stated anywhere in the page’s title, headings, or opening lines.
That’s a narrow, specific claim, not a vague gesture at “content quality.” It describes a system that can find one good sentence inside a mediocre page and rank it for the exact query it answers, independent of how the surrounding content performs.
Why this became the backbone of AI Overviews
Fast-forward to Google’s July 2026 search talks, where Google spoke openly about chunking: breaking a page into passages and evaluating them semi-independently. Google’s own documentation on AI features in Search confirms the retrieval mechanics underneath AI Overviews and AI Mode work the same way, pulling from passages and sub-topics across many pages rather than scoring one page as a monolithic block.
Google hasn’t officially stated that 2020’s passage ranking and 2026’s chunking are the identical codebase. What’s not in dispute is that they solve the same problem with the same logic: a document is not one unit to be matched against a query, it’s a set of smaller units, and the smallest unit that answers the question best is the one that gets surfaced, whether that surfacing is a blue link or a sentence folded into an AI-generated summary.
Once you see a page that way, a lot of confusing ranking behavior makes sense.
Why a strong page still loses queries it should win
Picture a long, genuinely good guide. It covers a topic thoroughly. By old logic, it should rank for everything in it.
Under passage-level evaluation, it competes passage by passage. If the answer to a specific question is buried in a paragraph that also wanders into three other subtopics, that chunk is muddy. It doesn’t cleanly answer any one query, so it loses to a competitor whose passage answers that exact question and nothing else.
The page is strong. The chunk is weak. And the chunk is what was in the race.
This is why thoroughness alone stopped being enough. A wall of comprehensive prose can lose to a thinner page that happens to have one clean, self-contained passage per question.
How to write a passage that actually ranks
The fix isn’t shorter content. It’s content built so each passage stands on its own.
Name the question in the heading. Give each distinct question its own section, under an H2 or H3 phrased as that question. When Google’s systems chunk the page, your section boundaries become natural chunk boundaries, and each chunk arrives at the query pre-labeled with what it answers.
Answer in the first sentence or two. The direct response goes at the top of the section, then supporting detail. This is the same reason most AI citations come from near the top of a page, applied one level down, inside the section itself.
Keep one idea per section. The moment a passage covers a question and then drifts into two adjacent ones, it stops being a clean answer to any of them. One idea per chunk is the discipline.
Make the section stand on its own. A chunk gets evaluated without the three sections above it, so a passage that depends on earlier context reads as incomplete when lifted out alone. Write it so a reader, or a retrieval system, landing only on that section still gets a whole answer.
None of this fights readability. A page organized into clean, self-contained sections is easier for a human to scan and easier for a machine to chunk. The deeper mechanics of what gets retrieved are in how AI search retrieves content and structuring pages for AI extraction.
How to check which of your passages are winning
There’s no report labeled “passage ranking” in Search Console, so checking means testing indirectly, the same way you’d check query fan-out.
- Pull your query list in Search Console and look for split performance on one page. If a single URL gets impressions for several distinct questions but ranks well on some and poorly on others, that’s often a passage-level problem, not a page-level one. The page is fine overall. Specific sections are losing their specific races.
- Read each H2 section in isolation. Cover everything above it and ask whether it still answers its own heading completely. If it leans on a sentence from two paragraphs up, a retrieval system sees the same gap you just noticed.
- Search your own target question in Google and in AI Mode, and check whether the snippet or citation pulls the sentence you intended, a different sentence from your page, or a competitor’s passage instead. Whichever sentence gets pulled is the one currently winning that passage-level race.
- Watch for a page that ranks on page one for its head term but never gets cited for narrower sub-questions it technically covers. That gap usually means the content exists but isn’t packaged as a standalone passage anywhere on the page.
- Rewrite one weak section at a time and re-check, rather than restructuring the whole page at once. Passage changes are narrow enough that you can isolate what moved.
Before and after: the same fact, rewritten as a passage
Say you run a SaaS product and you want to win the query “does [product] support SSO.”
Before, inside a long “Security and Integrations” section: “Our platform takes security seriously and integrates with a wide range of enterprise tools. We support SSO for Enterprise plans, along with audit logs, SCIM provisioning, and role-based permissions, all backed by SOC 2 Type II compliance.”
That passage answers four different questions in one paragraph: does it support SSO, does it have audit logs, does it support SCIM, is it SOC 2 compliant. Lifted out alone, it’s a run-on list, not a clean answer to any single one of them.
After, as its own section:
Does [Product] support SSO?
Yes. SSO is available on Enterprise plans via SAML 2.0, with Okta, Azure AD, and Google Workspace as supported identity providers.
Same fact. Now it’s a section a chunking system can lift whole, cite, and answer the exact query with, because the heading states the question, the first sentence states the answer, and nothing else is competing for attention inside it.
Passage ranking vs. page ranking, at a glance
| Page-level ranking | Passage ranking / chunking | |
|---|---|---|
| What competes | The whole URL, as one unit | Individual sections within the URL |
| What wins | Overall topical authority and depth | The single cleanest, most self-contained answer to one question |
| Risk for a thorough page | Low, comprehensiveness was the whole strategy | High, if sections blend multiple ideas together |
| Best heading style | Descriptive, keyword-themed | Phrased as the exact question being answered |
| Where the answer goes | Anywhere in the piece | First sentence or two of the relevant section |
| What Google confirmed | Standard page ranking, always | Passage ranking (Oct 2020); chunking for AI features (2026 search talks) |
Common mistakes that keep passages from ranking
Burying the answer under throat-clearing. A section that opens with “there are many factors to consider when thinking about X” before ever answering X is making a retrieval system do extra work to find the actual sentence. Cut the warm-up. Answer first, context after.
Writing one heading and three answers underneath it. A section titled “Pricing and Plans” that covers cost, contract length, and cancellation policy in one unbroken paragraph is really three passages pretending to be one. Split it into three headings, each with its own clean answer.
Reusing the same sentence structure across every section, which makes a page easy to skim as a human but doesn’t change the underlying issue if each section still drifts across multiple ideas. Structure and self-containment are different problems. Fix both.
Assuming an FAQ block fixes this automatically. FAQs help because they’re naturally self-contained, question-and-answer pairs. But burying the real explanation in body prose while leaving a thin, generic answer in the FAQ section underneath undersells the content that’s actually doing the work. If the FAQ answer is good enough to stand alone, the matching body section should be written with the same discipline.
Treating this as a one-time rewrite. Passage-level performance shifts as competitors restructure their own content and as Google’s retrieval systems change. Revisit sections that handle your highest-value questions every few months, not just once at launch.
The bottom line
Passage ranking isn’t new. Google told you about it in 2020. What’s new is how directly it now connects to the AI answers sitting above traditional results, because a system built to find one good sentence in a mediocre page is exactly the system an AI Overview needs to assemble a cited answer out of pieces from many pages at once.
Write pages a machine can take apart, because it already does. Give every distinct question its own labeled section, answer it immediately, keep it to one idea, and make sure it reads as complete without anything above it. That’s the whole discipline, and it was true in 2020 before anyone called it chunking.
Related guides
- 10 SEO Terms Every Marketer Should Know in 2026 (Explained Simply)
- Query Fan-Out: How Google’s AI Mode Actually Searches
- What Is Generative Engine Optimization (GEO) in 2026?
- Entity Salience: What It Means and How to Improve It
