How do I find topic cluster opportunities in my own data?

Topic clusters aren't new. What changed is who reads them first. Google says its AI features may use query fan-out, "issuing multiple related searches across subtopics and data sources", to build one answer (AI features and your website, last updated December 10, 2025). A cluster plan starts from those same subtopics.

That makes it tempting to build a page for every one of those searches. Adam Tanguay's guide to topic clusters on Search Engine Land (October 5, 2026) answers that in one line: "So I focus on depth: covering recurring subtopics instead of building a page for every fan-out." I work the same way, and Google's own line on building a page for every AI search is stricter still.

What is a topic cluster now?

A topic cluster is a hub page and the pages around it, each answering one question people ask about the topic, linked in the order they ask them. It's the content side of my SEO system, and it starts with one question: Which questions deserve their own page, and which belong together?

Three rules carry the answer:

• One page per question, so my own pages never compete with each other
• Hubs that connect the questions people ask in order
• Pages built only when someone is looking for them

Fan-out is also smaller than it sounds. Ahrefs studied 66.6 million ChatGPT answers from August and September 2026: "Only 14.4% of the 66.6 million responses required a web search", and when one did, ChatGPT "ran about 1.6 fan-out queries on average" (Ahrefs, October 12, 2026; a vendor's own study of free, logged-out answers). That's a handful of searches per answer, on the subtopics that keep coming back. Depth covers those.

Which questions deserve their own page?

The ones with no home yet. Before I build anything, I check which page already draws the searches.

On The Quack Report, my Oregon Ducks site, I almost built a separate depth chart page. Those searches were already landing on the roster page, so I kept one page instead of splitting it in two. The Quack Report case study tells the rest.

To find the questions with no home, I read two places. Search Console shows how people phrase things. Bing's new Topics view lists Opportunity Queries, the searches "where other sites appear and yours does not" (Bing Webmaster Blog, October 13, 2026). Bing's own caution goes with it: "Treat the numbers as directional signals for planning, not as a precise ledger." I logged the day it shipped.

The order I run it in hasn't changed: Use Peec for answers, Bing for citations, GSC for phrasing, and GA4 for the money end, and run the loop: listen, mine, answer, verify.

Depth shows up in the data as spread. On WireRef, my electrical reference site, no single page carries the AI citations: the most cited page holds 17% of Bing's page export, and the top ten hold 43%. The table at the end of this post shows where the rest land, folder by folder.

Does every page in a cluster do the same job?

No. Tanguay sorts a cluster's pages by answer behavior: answer-only, still a click, or community-owned. I ran one of my own clusters through it: the book pages on The Crawler's Cookbook, my fan site for the Dungeon Crawler Carl books. The two columns come from different engines, so read each one on its own, never one against the other.

Bing Webmaster Tools, AI Performance page export, and Google Search Console page rows, both Apr 13 to Oct 11, 2026. The label is my rule over each column on its own: "Still clicked" is 1K or more Google clicks; "Cited, rarely clicked" is 1K or more citations with under 100 clicks. 8 more pages in the cluster had under 100 of each.

The Book 8 FAQ is the clearest answer-only page: Bing's AI cited it 7.8K times, and Google sent it 68 clicks over the same days. The Book 9 page sits at the other end, with 14.5K Google clicks.

Those are two different jobs, so I judge them by different numbers. A page AI answers quote needs the answer in its first lines, in words a model can lift, and its measure is citations. Tanguay says to expect it: "Expect answer-only spokes like the grooming guide to lose clicks." A page people still click needs a reason to stay and a way onward, because Content now succeeds when it helps users move from understanding toward application, evaluation, or decision.

What about the questions people take to Reddit?

Tanguay's third kind is community-owned: questions people want answered by other people. My data names them. From Apr 13 to Oct 11, 2026, 21 different searches that showed a Cookbook page had "reddit" in them, 427 impressions in all, and 79% of those impressions were on the books cluster. The biggest was "dungeon crawler carl book 9 release date reddit".

Those people want other fans, not my page, and Tanguay's advice fits: "The job there is to show up as a person in the thread." I read those threads as research, not as targets. They show what fans still ask each other, which is the list of questions the cluster hasn't answered yet.

Where does creativity come in?

Data shows where the gap is. It doesn't show what the page should be. That part starts with the person on the other end of the search: where they are, how much time they have, and what they're holding.

On Title Town PDX, my Portland women's sports site, the gap was the game-day questions the official team sites skip, like the bag policy at each stadium. No keyword list hands you that page. You find it by picturing a fan in the parking lot with the wrong bag. The Title Town PDX case study shows what those guides became.

How does this work on enterprise sites?

The same way, at a bigger size. At my full-time job I work on SEO and AI visibility for enterprise brands across many industries, and the cluster method doesn't change with the logo:

• Split demand into the few themes people actually search, and give each theme its own pages and its own measures.
• Keep branded and non-branded searches apart in every report. Non-branded is where new buyers find you.
• Write for the non-branded questions around each main topic, link the topics to each other, and give every article a natural bridge to the page where the decision happens.
• Find the questions nobody answers well, and build the content only you can write.

The size changes one thing: I work from bulk exports with far more rows than the interface shows. AI helps me sort them into themes, which would take weeks by hand, and I decide which questions deserve their own page. The industry pages show how the method shifts from one industry to the next.

What can't this data tell you?

• It's two engines. The citations are Bing's and the clicks are Google's. Each column is read on its own, never as a ratio.
• The labels are mine. "Cited, rarely clicked" is a line I drew over two columns, not something either engine measures.
• Bing's page export is capped. It lists at most 300 pages. The Cookbook fits under that; WireRef's export is at the cap, so its real spread is wider than the table shows.
• A citation is not a visit. It means an answer used the page, not that anyone clicked through.
• Jump links count toward their page. Search Console reports an address with a #section on its own; the clicks here add those to the page they belong to.

Written for this site and first published October 17, 2026. Every number in it is computed from my own exports when the site builds, with its data-through date beside it.