Passage indexing and content fragmentation: designing for section based visibility

TL;DR

Difficulty: Beginner

  • It is ranking, not indexing. Google still indexes whole pages. Since February 2021 it can also use individual passages as one signal of how relevant a page is.
  • You cannot optimise for it. Google says messy pages are understood about as well as clean ones, and advised against chasing it.
  • Featured snippets and text fragments are separate systems. A passage that answers the question in its first sentence can be quoted and linked to directly.
  • Keep sections visible. For Read more deep links Google says avoid accordions and tabs, scroll scripts, and dropping the hash fragment.
  • Split a page only for the reader, and measure query groups in Search Console.

The name passage indexing promised that every section of a long page would now be found on its own. Google corrected it within days: it indexes pages and ranks pages, and uses passages as one more signal. In this article I separate what passage ranking does from what featured snippets, text fragments and Read more links do, and restructure one long guide for an invented Galway tile shop, with the figures before and after.

Passage indexing was never indexing, it is a ranking signal

In October 2020 Google announced that it had "made a breakthrough in ranking" and could now "better understand the relevancy of specific passages" on a page. The problem it named is one every long guide has: "the single sentence that answers your question might be buried deep in a web page." Google said the change would improve 7% of search queries across all languages once it was rolled out everywhere.1

The name that went round the industry that week was passage indexing. It stuck, it is in the title of this article, and it is wrong. Within days Google's Search Liaison account corrected it in three sentences: "This change doesn't mean we're indexing individual passages independently of pages. We're still indexing pages and considering info about entire pages for ranking. But now we can also consider passages from pages as an additional ranking factor."2 So the honest name is passage ranking, and that is what Google calls it today. Its own guide to its ranking systems describes it as "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."3

The old version of this article got that backwards. The passage helps Google judge how relevant the page is. The page is still what is indexed and what ranks. Nothing on your site is cut into fragments and filed separately.

The change went live for searches in English in the United States on the afternoon of 10 February 2021, Pacific time, with other countries and languages to follow.4 Danny Sullivan of Google confirmed a week later that it had finished rolling out on 15 February.5

Worked example: the sentence on page nine

Corrib Tiles is an invented tile shop in Galway, at corribtiles.example, run by Aoife. Nobody real, every number made up. Two years ago Aoife wrote one long guide, "Tiling a bathroom, start to finish", about 6,000 words. Near the bottom, in a section headed "Old surfaces", there is a paragraph that answers a question her customers ask every week: can you tile over existing tiles? The answer is yes, if the old tiles are firmly bonded, clean and flat, and the extra thickness does not foul the doors or the bath panel.

Before February 2021, someone in Cork typing "can I tile over old bathroom tiles" was unlikely to see Aoife's guide. The page is about tiling a bathroom, mostly setting out, cutting and grouting, and the one paragraph that answers the question is a small part of it.

After the change, Google's systems can notice that paragraph and treat it as evidence that the whole page is relevant to the question. The result is still Aoife's guide, with its title and URL. The page was always in the index. The passage got it into the ranking.

How Google handles a page: crawl, index, rank, and where passages fit

To see why the difference matters, it helps to have the three stages straight. Google describes them in its own guide to how Search works.6 Crawling is when Google's automated programs, called crawlers, download the text, images and videos from pages they find. Indexing is when Google analyses what is on each page, including the text and key tags such as the <title> element and image alt text, works out which of several near duplicate pages is the main one (the canonical), and stores what it learned in a large database called the index. Serving is when someone searches and Google returns the pages it judges most relevant to that query, in order. Ranking is that ordering.

Passage ranking sits entirely in the last stage. It changes how relevance is judged at the moment of a search, not what is fetched or stored.

Where passage ranking sits in Google SearchA top to bottom flow. A crawler downloads the page. The whole page is analysed and stored in the index as one page. A person searches. Google judges relevance using signals about the whole page and, as one additional signal, the passages within it. The page, not the passage, is ranked and shown, sometimes with a snippet drawn from a passage.A crawler downloads the pageThe whole page is analysed and storedin the index as one pageSomeone types a specific questionGoogle judges relevance from signalsabout the whole pagePassage ranking adds one more signal:how well individual sections match thequestionIs the page among the most relevantresults?noThe page is not shown for this searchyesThe page is ranked and shown with itstitle and URLThe snippet under it may quote apassage, chosen by separate snippetsystems

The diagram runs top to bottom. The page enters the index whole and leaves the process whole. Passage ranking is the highlighted step in the middle, one signal among many, feeding a decision about the page. The last step, the snippet, is where most of the confusion comes from, and it gets its own section below.

Worked example: what is in the index for Aoife's guide

Follow Aoife's guide through the diagram. The crawler fetches corribtiles.example/guides/tiling-a-bathroom. Google analyses the whole page and stores it once, under that one URL. There is no separate entry for "Old surfaces", no URL of its own for the paragraph about tiling over tiles, and nothing Aoife can submit to make one.

When the Cork search arrives, Google's systems consider the page as a whole and the passages inside it. The paragraph that answers the question is strong evidence. The page qualifies and ranks, say, sixth, under its own title. In Search Console Aoife sees the query listed against the page, never against the paragraph, because the page is the only thing Google has on record.

What passage ranking does not do, and what the old advice got wrong

The old version of this article made five claims. Only one survives the sources: that clear headings and well organised sections are worth the effort, though not for the reason it gave.

It said a single paragraph or chart "can be indexed and presented as the best answer". It cannot be indexed on its own, as Google said in October 2020.2 A passage can be quoted in a snippet, but that is a different system, covered in the next section.

It said passage indexing "penalises dense, unorganised content". Google said close to the opposite. John Mueller described the change in a Google video at the start of November 2020 as "more about ranking these passages from existing pages rather than indexing them" individually, said it was not a core update, and added that Google can "understand these messy pages more or less the same as we can understand clean pages."7 In September 2021 he explained that the system helps with "a really long page", usually one that is not especially well optimised for search, which "has like different intents almost on the same page".8 Passage ranking rescues the useful paragraph buried in a messy page. It is a help, not a penalty.

It implied that each section now ranks on its own. Mueller again, in the same explanation: "it's not so much that we would pull out like one sentence of this page and then just rank that higher". The system finds that "this one part is super relevant" and Google "will rank this page because of that one part".8 The page ranks. The part is the reason.

It said FAQPage markup supports this. Structured data plays no documented part in passage ranking. Separately, from August 2023 Google showed FAQ rich results only for well known, authoritative government and health websites, and the HowTo rich result is gone from both mobile and desktop.9 Since 7 May 2026, FAQ rich results no longer show at all.10 The markup questions belong in the article on AEO and GEO, which deals with them properly.

And it described a client whose guide sections "now appear independently in search results". There is no source for it, and it describes the mechanism Google says does not exist. I cut it.

Worked example: Aoife's two wasted afternoons

Aoife read an article much like the old one and set aside two afternoons. On the first she copied each of the guide's eight sections into a short page of its own, to "give every passage its own chance". On the second she collapsed every section of the guide behind a clickable heading, an accordion, and added FAQPage markup to the lot.

Here is what the sources say about each move. The eight pages gained nothing from passage ranking, which exists precisely so that a long page does not have to be split for its parts to count. They were thin copies of the guide, and now there were nine pages to maintain instead of one. The accordions, as the next sections show, made it harder for Google to send readers to a specific section. The FAQPage markup produced no rich result. Google stopped showing FAQ rich results in May 2026,10 and a tile shop in Galway was never a government or health website anyway.9 Two afternoons, and the best page was worse than before. Fair play to her for trying, but the effort went the wrong way.

When people say "Google showed my paragraph", they usually mean a featured snippet: the boxed answer, quoted from one page, that sometimes sits above or among the results. That is not passage ranking. Danny Sullivan said in October 2020 that featured snippets "are already identified using different systems than passages", and a Google spokesperson put the difference this way: ranking systems "determine the relevance of any web document via understanding of passages", while featured snippets identify "the most relevant passage in a document".11

Google's documentation on featured snippets is short and worth reading in full. Can you mark a page as a featured snippet? "You can't." Google's systems "determine whether a page would make a good featured snippet for a user's search request, and if so, elevates it", and "featured snippets will only appear if enough text can be shown to generate a useful featured snippet".12 The controls run the other way: the nosnippet rule blocks all snippets for a page, the data-nosnippet attribute keeps marked text out of them, and a lower max-snippet value makes a featured snippet less likely.12

So there are two questions about Aoife's paragraph. Does it help her page rank? That is passage ranking, and she cannot opt in. Is it quoted in the box? That is featured snippets, and she can only make it easier, with a passage that can be quoted whole.

Worked example: the paragraph that can be quoted, and the one that cannot

Here is Aoife's "Old surfaces" section as she first wrote it.

Old surfaces. A good few customers ask us about this one and honestly it depends. Some fitters will tell you never to do it, and in the past we would have agreed, but adhesives have come on a long way and it is often fine, as long as you check a few things first, which we will come to below after a quick word about primers.

There is no sentence in it that answers the question. A system looking for a passage to quote has nothing to quote. Here is the rewrite.

Can you tile over existing tiles? Yes, if the old tiles are firmly bonded, clean and flat, and the extra 10mm or so of tile and adhesive will not stop doors, the bath panel or the shower tray from fitting. Tap every tile first: a hollow sound means it has come loose and must come off. Then degrease the surface and apply a primer made for tiling onto glazed tiles.

The heading is the question as a person would type it. The first word is the answer, the rest of the sentence the conditions, then the detail. Lift the paragraph out of the page and it still makes sense. That is what a featured snippet needs, and what a reader skimming on a phone needs too.

Google's featured snippet documentation describes one more behaviour that people often mix up with passage ranking. When someone clicks a featured snippet, "scrolling to the position that appeared in the snippet happens automatically, without any additional annotation by the site." If the browser cannot do it, or Google cannot confidently tell where on the page the text is, the click lands at the top of the page as usual.12

The browser feature that makes this possible is called text fragments. A text fragment is a piece added to the end of a URL that tells the browser which words to scroll to and highlight, "without requiring the page author to add an ID".13 It starts with #:~:text= and the text to find. The specification is a draft from the Web Platform Incubator Community Group, edited by two Google engineers, and it is not a W3C standard.14 If a browser does not support it, or the words are not on the page, the fragment is ignored and the reader lands at the top.13

Google has also said this is a separate matter from passage ranking. When people asked in February 2021 whether passage ranking would jump readers to the passage, John Mueller replied that "ranking is generally independent of the snippet or the link".15

Two practical points follow. First, the directive is "stripped from the URL during loading so that author scripts cannot directly interact with" it,13 so your analytics will not see which passage a visitor was sent to. Second, if the words are not on the page as written, the fragment is ignored,13 so the exact visible wording matters.

Aoife wants to send a customer straight to the answer in an email, without touching the page. Step one is the page address, https://corribtiles.example/guides/tiling-a-bathroom. Step two is the fragment directive after it, #:~:text=. Step three is the start of the answer, "Yes, if the old tiles are firmly bonded", with each space written as %20 and the comma as %2C. Step four joins them: https://corribtiles.example/guides/tiling-a-bathroom#:~:text=Yes%2C%20if%20the%20old%20tiles%20are%20firmly%20bonded.

In a supporting browser the customer lands on the paragraph, highlighted. In any other, they land at the top of the guide, which is no worse than before, so no bother either way.

Google does send readers into the middle of a page by design in one place, and it publishes the rules. A "Read more" deep link is "a link within a snippet that leads users to a specific section on that page".16 Google lists three things that make these links more likely to appear. The content should be "immediately visible on the page to a human", and Google names content hidden behind expandable sections or tabs as the thing to avoid. You should "avoid using JavaScript to control the user's scroll position on page load", for example forcing the page back to the top. And if your page changes the URL with script as it loads, "make sure you don't remove the hash fragment from the URL", because that breaks the deep link.16

This is the one concrete instruction Google gives about structuring sections, and it cuts against the old version of this article, which recommended collapsible sections. For anything you would like Google to link into, that is the wrong way round.

Worked example: Aoife takes the accordions off

Aoife's guide, after her second wasted afternoon, had every section collapsed. A script on the page also scrolled to the top on load, a leftover from an old cookie banner, and the page's router rewrote the URL as it loaded and dropped anything after the #.

The fix was three changes, one per rule. She removed the accordions, so every section is open when the page loads and a reader can scroll through it. She gave each H2 a short, stable id attribute, such as id="tile-over-tiles", so the section has an address of its own (corribtiles.example/guides/tiling-a-bathroom#tile-over-tiles) that she can link from emails and from other pages. And she asked her developer to delete the scroll script and to keep the hash fragment when the router updates the URL.

None of this guarantees a "Read more" link. Google decides. It removes the three reasons Google gives for not showing one, and readers gain too: nobody is jumped back to the top any more. She also deleted the eight copied pages and redirected each to its section of the guide.

Structure for readers, because that is what the systems are built to reward

So is there anything to optimise? Google's own people have given the same answer every time they were asked. Martin Splitt said that if you already have good, structured pages, passage based ranking "will be the least interesting thing".5 Mueller cautioned against "trying to jump on the train" of optimising for it, and summed up the only advice anyone could give: "if you want a search engine to recognize a part of your page then you should structure your page properly."7

Google's SEO Starter Guide agrees: "break up long content into paragraphs and sections, and provide headings to help users navigate your pages." It also says that having headings in semantic order "is fantastic for screen readers", but from Google Search's point of view "it doesn't matter if you're using them out of order", and there is "no magical, ideal amount of headings".17

That gives a short list, every item for the reader first. One idea per section, under a heading that says what the section answers. A first sentence that answers it, so the section works for someone who lands in the middle. The detail after that. Sections that stay visible on the page. A stable anchor on each main heading. Headings in order, because screen readers and the people using them rely on it, even though Google does not. And a page that stays one page as long as it serves one reader with one broad need.

Worked example: the outline before and after

Aoife's original outline, in the order the headings appeared:

  1. Introduction
  2. Our story
  3. Getting started
  4. Tools
  5. Old surfaces
  6. The fun part
  7. Finishing off
  8. Final thoughts

Only "Tools" tells a reader what is in it, and "Our story" belongs on the About page. Here is the rewrite, each heading the question a reader would ask:

  1. What you need to tile a bathroom
  2. How to set out tiles so the cuts are even
  3. Can you tile over existing tiles?
  4. How to cut tiles round a pipe
  5. How long to leave adhesive before grouting
  6. How to seal grout in a shower

Six sections instead of eight. Each heading is an H2 with its own id, and each section opens with its answer. A reader skimming the headings now gets the shape of the whole job, and so does any system working out which part of the page matches a question.

When a section deserves its own page, and when it does not

Passage ranking removes the old reason to split a long page, which was the fear that a buried section would never count. It does not mean you should never split. The test is the reader, not the algorithm. If people who want this section mostly do not want the rest of the page, it is a different topic, and it should be its own page. If they usually need the rest too, it belongs where it is.

Should this section become its own page?A top to bottom decision flow. If the section answers a question that people search for on its own, and those people rarely need the rest of the page, and there is enough to say to stand alone, it becomes its own page linked from the guide. If any answer is no, it stays a section with a clear heading and an anchor.Take one section of a long pageDo people search for this questionon its own, often?noKeep it as a section, with a questionheading and an anchoryesDo those people usually need therest of the page too?yesKeep it as a section, and link to itsanchor from elsewherenoIs there enough to say for a pagethat stands alone?noKeep it as a section until there isyesGive it its own page and link to itfrom the guide

The flow runs top to bottom through three questions. A section becomes its own page only if it is searched for alone, its readers do not need the rest, and it has enough substance to stand up. Otherwise it stays a section, and its anchor does the linking a separate page would have done.

Worked example: the section on underfloor heating

Inside "Finishing off" Aoife had a long paragraph about tiling over electric underfloor heating mats. Run it through the three questions.

Do people search for it on its own? Yes. In Search Console the page receives searches such as "can you tile over underfloor heating mats" every month.

Do those people usually need the rest of the guide? Mostly not. Underfloor heating is its own job with its own steps, and many of them are tiling a kitchen, not a bathroom.

Is there enough for a page that stands alone? Yes, once she adds the steps, the waiting times and a photo of the mat test.

All three point the same way. The section becomes its own page, corribtiles.example/guides/tiling-over-underfloor-heating, and the guide keeps a two sentence summary with a link across. The "tile over existing tiles" section fails the second question, because those readers still need the setting out, cutting and grouting, so it stays.

Measuring whether the structure helped

No report shows passage ranking. Search Console does show the queries for each page, so group them by the section that answers them and watch each group's clicks.

The measure I use is a plain share:

sg=cgCs_g = \frac{c_g}{C}

Here sgs_g is the share of the page's clicks that came from query group gg, cgc_g is the number of clicks from searches in that group in the period, and CC is the total number of clicks to the page in the same period. It is a ratio, a part over the whole (Foundations: ratios and rates), and the change from one period to the next is a percentage change (Foundations: percentages). Small counts move a lot by chance, so I compare whole months and several of them, never a single week (Foundations: sampling).

Worked example: Aoife's "tile over tiles" queries

Aoife restructured the guide at the end of April. She grouped every query in Search Console that asks about tiling over old tiles ("tile over tiles", "can you tile on top of tiles", "tiling over old bathroom tiles" and so on) into one group, and counted the clicks each month.

Clicks from tile over tiles searches to the bathroom guideA bar chart of monthly clicks to the Corrib Tiles bathroom guide from searches about tiling over existing tiles, January to August. Between 11 and 15 a month before the restructure at the end of April, then 24, 31, 38 and 41 from May to August.Monthly clicks from the tile over tiles query group012.52537.55012Jan15Feb11Mar14Apr24May31Jun38Jul41AugClicksMonth, guide restructured end of April

Along the bottom are the months, with the restructure between April and May. Up the side are the clicks from that one group of searches: 11 to 15 a month before, then 24 rising to 41. The takeaway is the step and then the climb.

Now the share. From January to April the group brought 52 clicks out of 1,040 clicks to the whole page.

sg=521,040=0.05s_g = \frac{52}{1{,}040} = 0.05

With cg=52c_g = 52 and C=1,040C = 1{,}040, the share is 5% of the page's clicks, from people asking about tiling over tiles. From May to August the group brought 134 clicks out of 1,340.

sg=1341,340=0.10s_g = \frac{134}{1{,}340} = 0.10

With cg=134c_g = 134 and C=1,340C = 1{,}340, the share is 10%. It doubled, from 5% to 10%. The monthly average went from 13 clicks to 33.5:

Δ%=33.5−1313×100=157.7%\Delta\% = \frac{33.5 - 13}{13} \times 100 = 157.7\%

Here Δ%\Delta\% is the percentage change between the monthly average before and after, the same formula as in Foundations: percentages.

Two cautions keep this honest. Aoife changed the heading, the first sentence, the accordions and the anchors at once, so she cannot say which did the work, or how much was passage ranking and how much a featured snippet. And 13 clicks a month is a small number, where a few clicks either way is noise. A move to 33.5, held for four months and timed to the change, is not.

What I do on the pages I build

My own practice is the same for a Solidus shop's buying guides as for a SaaS product's documentation. I write every long page as a set of questions a reader actually has, each under a heading that names it, each opening with its answer. I keep every section open on the page, give every H2 a stable id, keep headings in order for screen readers, and check that no script on the site eats the hash fragment or forces the scroll. I split a section into its own page only when the three questions above say so. And I measure query groups monthly in Search Console, not passage rankings, because there is no such report.

What I do not do is "optimise for passage indexing". There is nothing to switch on, and an audit sold for it is selling a name Google corrected in October 2020.

The same passages are what AI answers lift and cite, and that mechanism is set out in the article on winning visibility with AEO and GEO. The ChatGPT side is in how to rank in ChatGPT. This article is the older, narrower question underneath both: Google indexes the page, ranks the page, and since 2021 has been able to notice the one good paragraph inside it. Write that paragraph so it can stand alone, and leave it where a reader can see it.

Footnotes

  1. Prabhakar Raghavan, How AI is powering a more helpful Google, Google, 15 October 2020, https://blog.google/products/search/search-on/

  2. Barry Schwartz, Google Passage Ranking, It Is Not Passage Indexing, Search Engine Roundtable, 19 October 2020, quoting the Google SearchLiaison account, https://www.seroundtable.com/google-passage-ranking-not-passage-indexing-30287.html

  3. Google, A guide to Google Search ranking systems (passage ranking system), Google Search Central documentation, accessed 7 November 2026, https://developers.google.com/search/docs/appearance/ranking-systems-guide

  4. Barry Schwartz, Google Passage Ranking Is Live In The US English Results, Search Engine Roundtable, 12 February 2021, quoting Danny Sullivan of Google, https://www.seroundtable.com/google-passage-ranking-live-30920.html

  5. Barry Schwartz, Google Passage Based Ranking Causing Minimal Impact So Far, Search Engine Roundtable, 18 February 2021, quoting Danny Sullivan and Martin Splitt of Google, https://www.seroundtable.com/google-passage-based-ranking-small-30952.html

  6. Google, In-Depth Guide to how Google Search works, Google Search Central documentation, accessed 7 November 2026, https://developers.google.com/search/docs/fundamentals/how-search-works

  7. Barry Schwartz, Google On Passage Indexing: It's Not A Core Update, It's Ranking Not Indexing & Don't Optimize For It, Search Engine Roundtable, 2 November 2020, quoting John Mueller of Google in a Search Central video, https://www.seroundtable.com/google-on-passage-indexing-tips-30360.html

  8. Barry Schwartz, Google's John Mueller Explains Passage Ranking In More Detail, Search Engine Roundtable, 14 September 2021, https://www.seroundtable.com/google-john-mueller-explains-passage-ranking-32076.html

  9. Google Search Central, Changes to HowTo and FAQ rich results, 8 August 2023, updated 14 September 2023, https://developers.google.com/search/blog/2023/08/howto-faq-changes

  10. Google Search Central, Latest Google Search documentation updates, entries of 8 May 2026 (Deprecating the FAQ rich result feature) and 15 June 2026 (Removing documentation for the FAQ rich result feature), https://developers.google.com/search/updates#faq-deprecation

  11. Barry Schwartz, Google: Passage Indexing vs Featured Snippets, Search Engine Roundtable, 22 October 2020, quoting a Google spokesperson and Danny Sullivan, https://www.seroundtable.com/google-passage-indexing-vs-featured-snippets-30310.html

  12. Google, Featured snippets and your website, Google Search Central documentation, accessed 7 November 2026, https://developers.google.com/search/docs/appearance/featured-snippets

  13. MDN Web Docs, Text fragments, Mozilla, accessed 7 November 2026, https://developer.mozilla.org/en-US/docs/Web/URI/Reference/Fragment/Text_fragments

  14. Nick Burris and David Bokan (editors), URL Fragment Text Directives, Draft Community Group Report, Web Platform Incubator Community Group, 13 December 2023, https://wicg.github.io/scroll-to-text-fragment/

  15. Barry Schwartz, Google Passage Ranking & Scroll To Text/Snippet Are Unrelated, Search Engine Roundtable, 15 February 2021, quoting John Mueller and Danny Sullivan of Google, https://www.seroundtable.com/google-passage-ranking-scroll-to-text-snippet-unrelated-30928.html

  16. Google, Control your snippets in search results ("Read more" deep links), Google Search Central documentation, accessed 7 November 2026, https://developers.google.com/search/docs/appearance/snippet

  17. Google, SEO Starter Guide: the basics, Google Search Central documentation, accessed 7 November 2026, https://developers.google.com/search/docs/fundamentals/seo-starter-guide

If you have a long guide that answers the right questions somewhere around page nine, I run a free workshop and the first two weeks of work are on me. I will find the paragraphs worth surfacing and the accordions worth opening.