Topic clusters and pillar pages: how do they work now that AI searches for you?

The name sounds like something an agency made up to bill you for, but there’s not much to it, and I think it matters more now than it did before people started asking ChatGPT who to hire. Your pages wind up competing with each other when nobody ever sat down and said which one should rank for what. So a cluster is really just you making that call.

What are topic clusters and pillar pages?

A topic cluster is one pillar page that covers a broad subject plus a set of smaller pages that each answer one question about it, and they link in a set pattern, the small pages up to the pillar and the pillar down to every one of them.

Say you run a dental group and a big piece of your money comes from implants. Then your implants page is the pillar. It covers what implants are and who they’re for, how the process goes, roughly what they cost. But people ask a bunch of narrower stuff before they book, like how long do implants last, should I get a bridge instead, what about bone loss, can I finance it, and each of those gets its own smaller page that links back up to the implants page.

What I like about it is each page gets one job. The pillar owns the big topic and the small pages own the narrow questions, and you stop having three pages fight over the same search with none of them winning.

Do topic clusters still help SEO?

Yes, and it’s cause they’re built on links, which Google still uses. Google’s own link guide says it uses links as a signal when it decides how relevant a page is and to find new pages to crawl, and it says every page you care about should have a link from at least one other page on your site.

So when your supporting pages all point at the pillar with link text that says what it covers, you’re telling Google pretty plainly which page you want showing up for the big topic. I get into the bigger version of this in whether SEO is dead because of AI, and my answer there is no, and site structure is one of the pieces that still pays off on Google and in the AI tools both.

Do topic clusters help you get cited by ChatGPT and other AI tools?

They give you more chances to get picked, and that’s as far as anybody can honestly promise. A University of Toronto study by Mahe Chen and Nick Koudas found ChatGPT splits a question into about four keyword-style searches, and a page that came back in several of those searches had 10.8 times the odds of being cited. A cluster that answers the related questions just has more shots at showing up.

The authors call their results associations and I’m not gonna tell you a cluster guarantees anything, but the logic holds up to me. Say a buyer asks ChatGPT which builder near them does custom homes on sloped lots and it runs a few related searches behind the scenes, maybe one on local builders and one on cost. A site with a strong page for each angle has more ways to come back than a site with one page.

What do AI follow-up searches look like?

Ahrefs looked at 15 million of them, which it calls fan-out queries, pulled from 66.6 million free, logged-out ChatGPT answers in August and September 2026. Only 14.4% of those answers ran fan-out searches at all, about 1.6 each when they did, and the words ChatGPT added most often were “2026,” “official” and “price.”

That tells you buyers and the AI both want current info, the real source and the cost, so I’d want an honest, current pricing page in the cluster, and I think a site that puts its price range out in the open is better off than one that makes people fill out a contact form to find out.

The Toronto paper also found that, comparing pages on the same website, a page whose best paragraph matched the question one standard deviation more closely got 15.5% more citations, and a better title match went with 14.1% more. One focused page per question is the simplest way I know to get that match, and I go further into the ChatGPT side in how to get your business cited by ChatGPT.

How do you build a topic cluster?

Start with the questions buyers ask, not a keyword tool, then decide which page answers each one and link them in a fixed pattern.

1. Pick one topic that makes you money. A service line works, or a practice area or a product category. Then write down every real question buyers ask about it, out of your sales calls and emails, Google Search Console and the “People also ask” box, and if two questions have the same answer they share a page.
2. Give every page you already have one question to own. Two pages answering the same thing get merged. And a page that doesn’t answer anything a buyer actually asks probably doesn’t belong in the cluster at all.
3. Write the pillar, or rework the one you’ve got, so it covers the whole topic and answers the main question right in the first paragraph. Supporting pages open with their answer too.
4. Link each supporting page up to the pillar, and the pillar down to each of them. Sideways links only go where a reader would want them. Keep the whole plan in a sheet with every URL and its question and links next to it.

If I had to pick one step not to skip it’s the second, cause pages nobody gave a job to end up competing, and I think merging two weak pages into one stronger page is often the quickest win in the whole project.

If your topic happens to be AI search, my guide to AI visibility is set up as a pillar, and you can see how it links down to its posts and across to the other pillars.

How should pillar pages and supporting pages link to each other?

Each supporting page links up to the pillar one time, in the body, and the link text should say what the pillar covers. The pillar links back down to all of them. Past that I’d only add a sideways link when somebody reading would actually want that next page.

Google is pretty plain about anchor text. Good link text, it says, is descriptive and reasonably concise and relevant to both the page it sits on and the page it points to. Something generic like “click here” doesn’t tell anybody much, and Google counts keyword stuffing as a spam violation, so don’t cram keywords into your links. And please don’t put forty links on every page. A link is supposed to mean you’ll want this next, and once everything links to everything it stops meaning anything.

How many supporting pages does a cluster need?

As many as there are real buyer questions that need their own answer, and that’s it. There’s no magic number. Google’s AI guide says there’s no ideal page length, and it warns that making separate content for every variation of how people might search, mainly to game rankings or AI answers, violates its scaled content abuse policy.

For most service topics I’d guess somewhere between a handful and a couple dozen supporting pages, but that’s my own working range and not something a study told me. If you catch yourself writing a page that says what another page already says with the words shuffled around, I’d stop and fold it into the one you’ve got.

What mistakes break a topic cluster?

The big one is building pages for keywords instead of questions. Google’s spam policies describe doorway abuse as pages “created to rank for specific, similar search queries,” and one of their examples is substantially similar pages that are closer to search results than a clearly defined, browseable site. A city-by-city or keyword-by-keyword page farm is right in that zone.

The rest are quieter. Two pages on the same question split your signals, a pillar that’s just a list of links gives nobody a reason to land on it, and supporting pages that never link back up leave the pillar sitting there alone.

There’s a trap on the AI side too. TripleDart’s study of 11,002 pages found word count, heading counts, link counts and schema didn’t separate the pages AI Overviews cited from the ones they skipped, which means a cluster that’s big just to be big doesn’t buy you anything. The small edge it did find was in covering the question, with cited pages scoring 0.62 on H2 query coverage against 0.59 for the skipped ones, and I go into that more in how to rank in Google AI Overviews.

Pillar page or supporting page?

The two page types do different jobs, and this is how I tell them apart when I’m planning a site.

What does it cost to wait?

Mostly it costs you the pages you’ve already paid for. While the site stays unorganized your older pages keep competing, the page you want ranking doesn’t get a clear signal, and the AI’s related searches are more likely to land on whoever did answer those questions. AirOps looked at more than 4,000 pages ChatGPT cited and found pages that hadn’t been updated in over a year were more than twice as unlikely to be cited.

Usually nothing crashes and it’s easy to miss, you just rank for less than you should and get quoted less than you could. The fix starts with thinking more than writing. Pick one topic that makes you money, list its questions and match every page you already have to one of them, and you’ll see pretty fast where the gaps are.

What I can show you from my own brand

I don’t have a controlled test that isolates clusters. What I do have is where my own ecommerce brand landed. In July 2026 it ranked #1 on Google for its main “best” product search, and 18 of its 22 target terms were in the top 3. Google Search Console had it on page one for more than 400 terms, averaged over the three months to June 2026, and Bing’s AI Performance report shows its AI citations went from 9 a month in April 2026 to 1,275 a month in August 2026.

I can’t tell you how much of that came from site structure, and the brand runs paid ads too, so I won’t credit it to clusters. Every package I sell includes topic clusters and an internal linking plan, and you can see what each package covers and costs on my homepage.