Analysis Published 17 November 2026 15 min read

Pillar Pages in SEO: When a Hub Helps and When It’s Just a Long Page

A pillar page works as a hub only when its subtopics have search results of their own. The test for that, what Google's documentation says about long pages, and the failure that matters more than length: a hub whose sections compete with its own cluster pages.

Roman Makuev
Last updated

A pillar page is a broad page that covers one topic at overview depth and links to separate cluster pages for the subtopics that need a URL of their own. Whether it works as a hub or just sits there as a long page depends on something you can check before writing a word: whether those subtopics have search results of their own. When a subtopic brings back the same pages as the head term, it belongs in a section of the pillar page. When it brings back different pages, it needs its own article, and the pillar’s job is to send readers there.

Most advice on pillar page SEO turns that into quotas. 2,000 words, or 3,000, or “well over 5,000”, with ten to thirty cluster articles around the page. Google has never asked for either number.

I’d go further. A long page with no cluster pages at all is a legitimate result of that check, and usually the safer one. The pillar page that does damage is the hub built to a quota. Its sections answer the same questions as the cluster pages it links to, two of your URLs compete for one query, and neither holds it.

Pillar page, hub page, long-form guide: one name, two models

HubSpot built the topic cluster model on its own research in 2015, and in HubSpot’s version the pillar page is the hub: a broad page, cluster pages around it, links in both directions.

Animalz split the words in 2020. A hub, in their usage, is a table-of-contents page whose job is to send the reader on to the spokes. A pillar is one long piece built to keep the reader on the page. Fold a hub and all its spokes into a single URL and you get a pillar. Conductor uses the same split and calls the page in between a hybrid; Siteimprove calls the long page that never had a cluster cornerstone content. Most arguments about pillar pages are two people using different definitions.

PageWhat it doesWhat search gets
Hub (HubSpot’s pillar, Animalz’s hub)Summarises a broad topic and routes readers to cluster pagesOne URL for the head term, separate URLs for subtopics that have their own results
Long-form pillar (Animalz’s pillar, cornerstone content)Covers the topic and its subtopics on one pageOne URL; passage ranking can still match its sections to narrower queries
HybridOverview sections, with a link out wherever a subtopic has outgrown its sectionBoth of the above; where most working pillar pages end up
Cluster pageAnswers one narrow question in depth and links up to the pillar pageIts own URL for a query whose results differ from the head term’s
Landing or service pageConvertsCommercial results of its own; a pillar page can link to it, it can’t stand in for one
Category pageLists productsThe hub for commercial head terms where guides don’t rank at all

“Content pillars” is something else again. In social media planning it means the three to five themes a brand posts about, and it has no bearing on how a site is built.

The test: does the subtopic have search results of its own?

Take each subtopic you’d put on the pillar page, search it, and compare its top ten with the top ten for the head term.

If several of the same URLs come back, search is treating the two queries as one intent. The pages that rank for the head term also rank for the subtopic, usually with a section. That subtopic stays on the pillar page. If the top ten is a different set of pages, the subtopic gets a cluster page.

“Several” needs a number. The default in the SERP clustering tool in Algorithm, our platform, is three shared URLs out of ten, adjustable from two to seven. Three is a practitioner’s threshold. Google doesn’t publish the point at which it treats two queries as one intent.

One or two shared URLs is a grey zone, and I put those subtopics on the pillar page as sections. A section can be promoted later. A thin cluster page that never needed to exist has to be merged and redirected. A sign worth waiting for before promoting a section: in Search Console the pillar page sits somewhere between 8 and 20 for the subtopic’s query, and the pages above it are articles about that subtopic alone.

The keywords for this article went through the same check on 15 November. The check ran in a search tool that isn’t Google, so treat the detail as indicative. “What is a pillar page” brought back the same guides as the head term: Reliablesoft, Semrush, Whitehat. So the definition is in the first paragraph here and has no page of its own. “Pillar page examples” brought back a different top ten, roundups from Bynder, Siege Media, Ahrefs and HubSpot, and this article leaves the gallery to them. “Content pillars” brought back social media planning guides.

HubSpot had a sniff test for this in the original research: if the page you’re working on is meant to rank for a long-tail keyword, it isn’t a pillar page. The shared-URL count is that test with a number.

What Google documents about long pages

Search Central has no page about pillar pages. Four documents cover the decision anyway.

DocumentWhat it saysWhat it means for a pillar page
A guide to Google Search ranking systemsRanking systems are designed to work on the page level. The passage ranking system identifies individual sections of a page to understand how relevant the page is to a searchA section of a long page can answer a narrow query without a URL of its own
Creating helpful, reliable, people-first contentAsks whether you’re writing to a word count because you heard Google prefers one, and answers in brackets: “No, we don’t.”Length is not a property of a pillar page
Optimizing for generative AI features in Search (15 May 2026)No need to split content into small chunks for AI features; no ideal page length; separate content for every variation of a query, fan-out queries included, made primarily to manipulate rankings or AI responses, breaks the scaled content abuse policy of March 2024A cluster page per question variation is the pattern Google named
Snippet documentation: “Read more” deep links (April 2026)A link inside the snippet that takes the user to a specific section of the page. More likely when the content is visible on load, no script forces the scroll position, and the History API leaves the hash fragment in the URLSearch can already send a reader straight to one section of a long page

The May guide answers one piece of advice by name: write every section as a short self-contained chunk so an AI model can lift it. Google says there is no need. The pressure to split comes from fan-out. AI Mode and AI Overviews can run several related searches to build one answer, each of those searches looks like a subtopic someone forgot to cover, and a page for each of them is what the guide tells you not to build.

Text hidden behind expandable sections or tabs lowers the chance of a “Read more” deep link to that section (Google’s snippet documentation, April 2026).

Pillar page templates are full of accordions, because a page folded into twenty collapsed headings looks short in a design review. By the documentation, that is also the version least likely to get its sections linked from the results. A jump-link table of contents shortens the page for the reader and hides nothing.

The ids matter too. If the template generates a section’s id from the heading text, every edit to the heading changes the id, and a deep link that pointed to the old one lands at the top of the page. Fixed ids, set once, belong in the development spec.

When a hub helps

Three conditions. The first is the test above, applied to the whole topic: subtopics with results of their own, each with enough to say to fill a page. Two or three of those can hang off a long page without changing what it is. A hub starts to pay when there are enough cluster pages that the reader needs routing.

The second is the head term’s own results. Search “redirect checker” and the results are tools from top to bottom: Kinsta, WebFX, SE Ranking, a dozen more. A page called “Redirect checking: the complete guide” won’t get into that top ten however good it is. The hub for that topic is the tool, with short explanations that link out to the cluster pages. In ecommerce the hub is usually the category page, with buying guides linking into it.

Our own blog has a topic that passes both conditions. We’ve published three articles on redirects: whether a 302 passes link equity, what redirecting 404s to the homepage does, and why a 301 isn’t penalised. Their intents are different enough that no clustering run is needed to see it, and each is a cluster page already.

The head term doesn’t come from grouping; it comes from keyword research on the topic, which turns up “seo redirects” and “types of redirects” as the broad wordings. Ahrefs’ guide to types of redirects ranks for both, so they look like one group. “Seo redirects” brought back guides in a check outside Google on 15 November: Ahrefs on types of redirects, Screaming Frog’s learning page, Search Engine Journal, urllo. seoClarity keeps its redirects guide at a URL that ends in /pillar/url-redirects. “Redirect checker”, one word away, is all tools. So the topic has room for a pillar page on SEO redirects that says which redirect does what and when, with the three articles under it and a link to a checker instead of an attempt to rank for one. We haven’t written it yet. When we do, the three articles keep the how-to depth and the pillar page gets the what, the why and the which one.

The third applies to sites that sell a service, and on a service site it’s the use of a hub I’d put first. An article about fixing one crawl problem doesn’t link naturally to a service page, because its reader isn’t choosing an agency yet. A pillar page written from the decision side of the topic sits between the two: the approaches, what each costs, when to do it in-house. The cluster pages link up to it, and it links to the service page in the one place where the reader is deciding. Whether that concentrates anything Google measures, I can’t tell you.

WordStream counted three well-known pillar pages in 2022. Localiq’s had 8,068 words and 32 internal links. WordStream’s own had 4,390 words and 69. Shopify’s had 6,759 and 44. That is a link about every 64 words on WordStream’s page and about every 250 on Localiq’s: the first is run as a hub, the second as a long page with references. The pages have changed since, and neither ratio is a target.

When it’s just a long page, and why that’s fine

Zero cluster pages is a legitimate outcome. If every subtopic shares its top ten with the head term, search reads the topic as one intent, and a page that answers it in sections is the kind of page that ranks. Reliablesoft says it in a line: a topic with few subtopics doesn’t need a pillar page, a regular post is enough.

Splitting it anyway looks like progress. Five new URLs in the sitemap. Then the new pages start with no links and less context than the sections had, the pillar page no longer answers those queries, and for a while neither URL holds them.

Search Console can show whether a section already works. In the Performance report, URLs ending in a hash and a section id sometimes appear next to the page’s own URL. John Mueller has explained these as links from Google’s results that lead to a specific section of a page, and the Read more links documented this April are one source of them. Filter the report by the pillar page’s URL and open the Pages tab. A section whose hash URL collects clicks stays where it is.

I don’t have a figure for how often splitting a working section costs its queries. Nobody publishes one, and the hash URLs only show up on sites where Google happens to link into sections.

The broken hub: a long page pretending to be one

The failure worth looking for has nothing to do with length. It is the hub whose sections answer the same questions as its own cluster pages.

Terakeet traces it to writing the pillar page first and the cluster pages after. Writing the cluster pages first and then a pillar page meant to be “complete” gets to the same overlap from the other side. What prevents it is a rule about the level of the answer. The pillar page says what the thing is, why it matters, what it’s made of and how the parts connect. The cluster page says how to do one step or fix one problem. When the pillar page’s section on a step is as detailed as the cluster page on that step, search has two URLs for one question, and that is keyword cannibalization inside one cluster.

SignWhere it showsWhat to do
The pillar page and a cluster page swap places on the same query from week to week; cannibalization in its plain formSearch Console, Performance, filter by the query, Pages tab: both URLs with impressionsCut the pillar page’s section to a summary that links to the cluster page; or, if the subtopic shares its top ten with the head term, merge the cluster page into the section with a 301
The pillar page ranks for its cluster pages’ long tail and not for the head termFilter by the pillar page’s URL, Queries tab: the head term missing, or far below the narrow queriesIt’s a long page in a hub’s position. Check what kind of pages rank for the head term before rewriting anything
Cluster pages are reached only from the table of contents, a sidebar or a related-posts blockA crawl that counts body links apart from navigation and widgetsOne link from the pillar page’s section on that subtopic, anchored on it; one link back from the opening paragraphs of the cluster page
A section and a cluster page say the same things in different wordsNowhere in a list of URLs. Only a section-by-section comparison of the two pages shows itRewrite the section down to what, why and how it connects; the how-to lives on the cluster page

On the merge in the first row: point the 301 at the pillar page’s URL with the section’s id after the hash. The visitor lands on the section. Google, by Mueller’s account, ignores everything after the hash and folds links to a hash URL into the page’s own URL, so the fragment costs nothing.

The third row is the one a template produces on its own. Table-of-contents links point inside the page. A card grid of cluster pages at the top and a related-posts block at the bottom give the cluster pages internal links with no context around them, and Google’s guidance on links asks for anchor text that describes the destination, which a widget title rarely does. We treat a link inside a paragraph as worth more than a widget link. Google hasn’t published a weighting.

Two of these checks run in Algorithm. The internal link checker counts links in body text separately from navigation, footer and sidebar links, and flags pages that only menus reach. The Topical Map compares pages section by section in embeddings, which is how the fourth row gets found. That comparison is our model’s reading of two pages. Google doesn’t publish how it compares them, and the proof that an overlap costs anything is still the Pages tab from the first row.

How to create a pillar page, decision by decision

The usual steps are all here: topic, keywords, outline, write, link. Each one comes with something to check, and the order matters, because the test above can group queries but can’t find them.

  1. Collect the keywords for the topic. Keyword research gives the head term, the broadest query that names the topic, and usually two or three wordings of it. It also gives the subtopics, along with People Also Ask, the H2s of the pages that rank, related searches, and the posts you’ve already published.
  2. Group the whole list with the shared-URL test. The head term’s wordings should land in one group; if they split, there are two topics. Subtopics with three or more shared URLs join the head term’s group and become sections. Subtopics with a different top ten become cluster pages. One or two shared URLs: a section to watch.
  3. Check what ranks for the head term’s group. Guides in the top ten mean a guide-shaped pillar page can compete. Tools, categories or service pages mean the hub is one of those, and the guide becomes a cluster page under it.
  4. Match the list against what’s live. An existing post on a subtopic is its cluster page already, and an existing broad post may be the pillar page, which Reliablesoft suggests reusing before writing a new one. With eight or ten narrow articles already live, the order can run backwards: give them a few weeks of impressions, read which queries Search Console attaches to each, and write the pillar page to the shape the cluster already has.
  5. Write each section at the level the test gave it. Sections that stay on the pillar page get the full answer. Sections with a cluster page get the what and why, and a link.
  6. Internal links from the body, both ways. Down from the section on the subtopic, up from the opening of each cluster page, anchors naming the page they point to. A new cluster page gets its link from the pillar page the day it goes live.

Which subtopics a site should cover in the first place, across all its pillar pages, is a planning question one level up.

How long should a pillar page be?

A pillar page should be as long as the subtopics it keeps need it to be. It runs long when most subtopics share a top ten with the head term and stay on the page as sections. It runs short when most of them have results of their own and move to cluster pages.

SourceLengthCluster pages
Semrush (November 2026)2,000 words or more5 to 15
TerakeetAt least 2,000 words—
SuperblogAbout 2,500 or more, as a pattern, not a rule—
Whitehat (competitive B2B)3,000 to 5,000 words10 to 15
WordStreamOften well over 5,000—
HubSpot—20 to 30 supporting articles
GoogleNo preferred word count; no ideal page lengthA page per variation, made to rank, is scaled content abuse

The recommendations differ by a factor of two and a half on length and six on cluster pages, and none cites a study of pillar pages. The number of cluster pages comes from the same check as the length. On some topics it is zero.

How to tell whether the hub is working

In the Search Console Performance report, a page filter with a regex for the pillar page and its cluster pages returns the whole cluster’s queries in one view. Then look at which URL holds which query. In a working hub the head term and the broad queries sit on the pillar page and the narrow ones on the cluster pages. The pillar page holding the narrow queries while the head term goes to someone else is the second row of the table above.

Brand searches hide whether the cluster reaches anyone new. The branded queries filter, open to all eligible properties since 11 March 2026, takes them out. I give a new cluster four to eight weeks before reading anything into positions.

Does a hub build topical authority?

The case that hubs build topical authority usually rests on vendor correlations: sites with topic clusters cited 3.2 times as often in AI answers (Yext, quoted by Whitehat), 43% more organic traffic (HubSpot). Numbers like these can’t separate the hub from what comes with it: more pages, more internal links, more time spent on one subject. I haven’t found a published study that does.

Google documents something narrower. Its ranking systems guide describes systems that work on the page level, and the one topic-level system it has named is news topic authority, announced in 2023 for news publishers. The 2024 leak of Google’s internal API documentation contains fields with names like siteFocusScore and siteRadius. Names are all anyone outside Google has; what the fields do is unconfirmed.

How much a hub adds over the same pages without one is a number I don’t have.

Pillar page SEO, as this article covers it, is about one page and whether each of its subtopics needs a URL. The plan above it, which subjects a site should cover and how their pillar pages relate, is a topical map question, and both questions sit inside our SEO service, where the shared-URL test runs before a brief gets written.