Keyword clustering groups related search terms by shared intent so one page can target all of them, instead of writing a separate article for every keyword variant. It is the practical answer to a problem that is getting bigger: Google’s own search chief now says people type longer, messier, need-based queries instead of short keyword fragments, and that makes single-keyword planning less reliable than it used to be.
Our beginner’s guide to keyword research for a new website covers how to find your first realistic keywords. This piece picks up where that one leaves off. It walks through a repeatable process for turning a long keyword list into clusters that build real topical authority, and it explains why that process matters more now that AI search breaks one question into several smaller ones behind the scenes.
What Is Keyword Clustering, and Why It Works
Keyword clustering is the practice of grouping keywords that share the same search intent, then mapping that group to a single page instead of many thin ones. HubSpot describes it as an SEO technique that groups related keywords with matching intent and targets them together on the same page, rather than spreading them across separate posts.
The logic is simple once you see it in a SERP. If Google shows the same set of ranking pages for five different search terms, those five terms are not five separate topics. They are five ways of asking for the same thing, and Google has already decided that one page can satisfy all of them.
Ahrefs built its clustering tool around exactly this idea. Instead of grouping keywords by how similar the words look, Ahrefs groups them by how much their top-ranking results overlap, under the assumption that shared rankings signal shared intent better than shared wording does.
Why Keyword Clustering Matters More in the AI Search Era
Search Engine Journal’s Roger Montti reported on a Bloomberg Odd Lots interview where Liz Reid, Google’s VP of Search, described a real shift in how people type queries. Reid said AI Overviews have produced “meaningfully longer queries,” because users no longer compress a specific need into a short keyword. They describe the actual problem and let Google translate it.
Reid gave the example of someone searching for a restaurant. In the old model, that search might have been “restaurants New York.” Under AI-assisted search, the same need can arrive as a single, detailed query covering cuisine, price range, group size, and dietary restrictions at once.
Behind the scenes, Google handles this through what it calls query fan-out: a complex query gets split into several smaller sub-queries, each one runs against the regular search index, and the results get synthesized into one answer, according to Montti’s reporting. That single mechanic is the real reason keyword clustering is worth doing properly now.
A site built around isolated keywords has, at best, one shot per fan-out sub-query. A site built around a tight keyword cluster, where every sub-question already has a page or section that answers it directly, has several shots. Reid’s own advice to publishers, per the same report, was to define their expertise and the unique angle they bring, which is a content-depth problem a cluster structure is built to solve. Our guide on practical strategies for ranking in AI search covers the content-quality side of that same shift.
Step 1: Collect and Enrich Your Keyword List for Clustering
Start wide, not narrow. Pull keyword ideas from your keyword tool of choice, from Google Search Console’s query data for pages you already have, and from the “people also ask” and related-search boxes on the actual SERPs for your main topic.
For each keyword, record four things: estimated monthly search volume, keyword difficulty, which URL currently ranks for it (yours or a competitor’s), and a rough guess at intent. You do not need perfect data at this stage. You need enough to sort the list in the next step. Google’s own Search Console documentation is a good primary source for the query data you already own.
A list of 200 to 2,000 keywords is a reasonable range for a single content area. Below that, you may be defining your topic too narrowly. Far above it, and you likely need semantic grouping tools just to make the list manageable, which the next step covers.
This raw list is the raw material for keyword clustering, not the finished plan. The next two steps turn it into one.
Step 2: Segment by Search Intent Before Clustering
Sort your raw list into informational, commercial, transactional, and navigational intent before you cluster anything. Our breakdown of the four types of search intent and how to target each one is a useful reference if you need a refresher on telling these apart from the SERP itself.
This step matters because clustering across intent types produces pages that satisfy no one well. A keyword cluster that mixes “what is keyword clustering” with “keyword clustering tool pricing” is really two separate pages wearing one topic’s clothing, one informational and one commercial.
Skipping this step is the single most common reason a keyword cluster underperforms after it is built.
Step 3: Choose a Keyword Clustering Method
Once your list is intent-sorted, you need a method for grouping keywords within each intent bucket. Two approaches dominate in practice, and they are not interchangeable.
SERP-Overlap Clustering
This method checks how many of the same URLs appear in the top 10 results for each keyword pair. Ahrefs treats three or more shared top-10 URLs as a reasonable threshold for grouping two keywords into the same cluster, while zero or one shared URL usually means they belong on separate pages.
SERP-overlap clustering is slower to run by hand because it requires pulling live search results, but it reflects what Google has actually decided about intent, not what a word-similarity algorithm guesses. It is the more defensible method when you are planning pillar pages that need to rank.
Semantic Clustering
Semantic clustering groups keywords by meaning rather than by shared rankings, typically using a cosine similarity score between keyword embeddings. HubSpot suggests a similarity range of roughly 0.75 to 0.85 as a workable threshold for grouping, and recommends this method once a keyword set grows past about 10,000 terms, since checking live SERPs for that many keywords becomes impractical.
Semantic clustering is faster at scale and does not require fresh SERP data, but it can miss cases where Google ranks oddly similar-sounding keywords with completely different pages. Treat it as a first pass to make a large list manageable, then spot-check the resulting clusters against the actual SERPs before you commit to a content plan.
Step 4: Map Keyword Clusters to Pillar and Supporting Pages
Once your clusters exist, assign each one a role in a simple two-tier structure: one pillar page per core topic, and several supporting pages beneath it for sub-questions that deserve their own depth rather than a paragraph on the pillar.
HubSpot’s guidance treats 5 to 20 keywords as a typical healthy cluster size, and suggests 6 to 10 supporting pages as a rough minimum before a dedicated pillar page is worth building at all. A cluster with only three closely related keywords probably belongs as a section on an existing page, not a new post.
Link every supporting page back to its pillar, and have the pillar link out to its best supporting pages. Our earlier framework on building a content strategy for AI search covers this same pillar-and-cluster architecture in more detail, including how to prioritize which clusters to build first based on demand and genuine expertise.
Step 5: Decide Which Zero-Volume Keywords Belong in a Cluster
Every keyword list includes terms that show 0 or near-0 estimated monthly searches. The instinct is to delete them, but that is not always the right call inside a cluster.
Ahrefs tested this directly. Its research found that the keyword “hreflang tag seo” showed an estimated 10 monthly U.S. searches, while Google Search Console recorded 20 impressions over 28 days for a page targeting it, roughly double the tool’s estimate. Across a small sample of its own top-five rankings for keywords with a volume of 10 or lower, the average page drew 11.3 impressions, close to the average estimated volume, and fewer than 1% of sampled keywords exceeded 100 impressions.
Ahrefs’ broader conclusion is that apparent traffic from a single zero-volume keyword usually isn’t real. It comes from a page ranking for hundreds of unmeasured long-tail variations of the same underlying question, not from the one keyword you typed into the tool. A page targeting “submit domain to search engines,” cited in Ahrefs’ analysis, shows only 10 monthly searches for that exact phrase but draws an estimated 3.8K monthly visits because it ranks for many similarly worded queries.
The practical rule inside a cluster: keep a zero-volume keyword if it represents a distinct sub-question your cluster would otherwise miss, especially one tied to a trending topic or a clear conversion path. Cut it if it is just an awkward rewording of a keyword you are already targeting elsewhere in the same cluster. Good keyword clustering treats search volume as one input, not the deciding one.
Step 6: Link Cluster Pages Together
A keyword cluster without internal links is just a loosely related pile of pages. The links are what tell both Google’s crawlers and an AI fan-out system that these pages belong to the same body of knowledge.
Link supporting pages to their pillar using descriptive anchor text, not generic phrases like “click here.” Link supporting pages to each other where one sub-question naturally leads to another. Keep your most important cluster pages within two or three clicks of your homepage, which is the depth HubSpot’s guidance recommends for pages you want search engines to prioritize.
Once a cluster is live, Google Search Console becomes your best source for refining it further. Check which queries are sending impressions to each page in the cluster, and use the ones you didn’t originally target to spot a missing supporting page or a pillar section that needs expanding.
Common Keyword Clustering Mistakes to Avoid
A few mistakes show up repeatedly once teams start doing keyword clustering at scale.
- Clustering before segmenting by intent. This produces pages that try to serve a reader researching a topic and a reader ready to buy, and usually satisfy neither.
- Trusting semantic clustering without a spot-check. A high similarity score does not guarantee Google treats two keywords as the same intent. Pull the actual SERP before finalizing a cluster.
- Deleting every low-volume keyword on sight. As the Ahrefs data above shows, volume estimates miss a large share of real demand, particularly for specific, narrow phrasing.
- Building clusters with no internal linking plan. Keyword clustering is an architecture decision, not just a grouping exercise. Without links, you’ve made a spreadsheet, not a content structure.
- Treating cluster size as a vanity metric. A cluster with five tightly related keywords and one excellent page beats a cluster padded out to twenty keywords with four thin articles.
| Clustering Method | How It Groups Keywords | Best For | Typical Threshold |
|---|---|---|---|
| SERP-overlap clustering | Shared top-10 ranking URLs between keyword pairs | Pillar pages that need to match what Google already ranks | 3+ shared top-10 URLs (Ahrefs) |
| Semantic clustering | Cosine similarity between keyword embeddings | Large keyword sets (10,000+) where live SERP checks aren’t practical | ~0.75–0.85 similarity score (HubSpot) |
| Term clustering | Shared words or phrases across the keyword list | Quick trend and niche spotting before deeper clustering | No fixed score; manual review of shared terms (Ahrefs) |
A quick comparison of the three keyword clustering approaches covered above, based on methodology described by Ahrefs and HubSpot.
Frequently Asked Questions
How is keyword clustering different from a regular keyword list?
A keyword list is flat. Keyword clustering groups that list by shared search intent and ranking overlap, then maps each group to a specific page, which is what turns a list into a content plan.
Do I need a paid tool to do keyword clustering?
No. SERP-overlap clustering can be done manually for a smaller list by searching each keyword and comparing the top 10 results yourself. Paid tools mainly save time once your list runs into the thousands.
Should every zero-volume keyword be included in a cluster?
No. Include it only if it covers a distinct sub-question your cluster would otherwise miss. Cut it if it is simply an alternate phrasing of a keyword you already cover in the same cluster.
How does AI search change keyword clustering specifically?
It raises the cost of gaps. Google’s query fan-out process, as described in Search Engine Journal’s reporting on Liz Reid’s comments, splits a complex query into sub-queries behind the scenes, so a cluster with thorough sub-question coverage has more chances to be pulled into an AI-generated answer than a single isolated page does.
What to Cluster First
Pick one topic where you already have two or three published pages and a decent amount of Search Console data. Pull every query sending impressions to those pages, run it through SERP-overlap or semantic clustering depending on list size, and map the result against your existing content.
You will usually find gaps: sub-questions with real search demand that none of your current pages answer well. Those gaps are your next content brief, not a brand new topic pulled from a keyword tool. Running keyword clustering on a topic you already rank for, rather than starting cold, is the fastest way to see whether the process is worth rolling out across the rest of your site.
Leave a response