What keyword clustering is
Keyword clustering is the process of grouping related keywords into topical clusters, where each cluster can be answered by one page. The output is a smaller set of mapped URLs, each targeting a primary keyword and a cluster of close variants. Clustering replaces the old one-keyword-per-page approach that dominated SEO in the 2010s.
Example. These five keywords look different at first:
- "how long does SEO take"
- "SEO timeline for small business"
- "how soon will I see SEO results"
- "when does SEO start working"
- "realistic SEO timeline"
They are five keywords but one topic. One page can answer all of them well. Clustering is the step where you recognise that, group them, and treat them as one URL target with five supporting variants.
Why clustering produces topical authority
Topical authority is Google's perception that a site is a credible voice on a subject. The signal builds when many pages on the site rank for many related queries that all sit inside the same topic. A site that ranks for one keyword in a topic looks like a one-off success. A site that ranks for thirty related keywords across the topic looks like an authority.
Keyword clustering is the planning layer that produces that pattern deliberately. Each cluster maps to a page. The pages link together in a hub-and-spoke around a pillar. The cumulative coverage of the topic earns the authority signal that lifts every page in the cluster set. See internal linking strategy for the hub-and-spoke pattern that ties clusters together once they are mapped, and entity SEO for the AI era for how clustering connects to the entity model Google uses now.
Three concrete benefits we see across client retainers:
- Pages within a clustered topic rank faster than isolated pages. Once two or three pages in a topic start ranking, the rest tend to follow inside three to six months as topical signals consolidate.
- Clustered pages are harder to dislodge from rankings. A competitor without topical coverage cannot easily overtake a clustered set with one good article. The SERP rewards depth.
- Clusters earn AI Overview citations together. AI engines that synthesise across multiple cited sources tend to draw from sites with deep topical coverage, not sites with one strong page in the area.
The manual SERP-overlap method
The gold-standard method. Slow but accurate. Use it whenever the keywords genuinely matter.
The principle: if two keywords have the same top three URLs ranking for them, Google has decided they have the same intent and the same page can satisfy both. Cluster them. If two keywords have completely different top tens, Google has decided they need different pages. Do not cluster them.
The steps
- Export your candidate keyword list to a spreadsheet. One column for the keyword, leave the next three columns blank.
- For each keyword, capture the top three ranking URLs. Either by hand-searching or by scraping. Tools like Ahrefs, SEMrush and Surfer can do this in bulk on paid plans.
- Compare URL overlap across keywords. Sort the spreadsheet and look for keywords that share two or three URLs in their top three. Those are clusters.
- Verify each cluster by reading the SERP for the leading keyword. Confirm the dominant intent matches your assumption. The search intent chapter covers the 30-second verification, and the search intent glossary entry gives the short definition.
- Pick the primary keyword per cluster. The one with the highest volume and the cleanest intent match.
- Map each cluster to one URL. Either an existing URL to rewrite or a new URL to create.
The whole process takes one to three hours for a typical small-business keyword list of 80 to 150 keywords. The output is a clean cluster map you can hand directly to a writer.
The AI-assisted method
The fast method. Less accurate but useful for the first pass on a large list.
Modern LLMs (ChatGPT, Claude, Gemini) are reasonably good at grouping keywords by semantic similarity. Paste a list of 200 keywords, ask for them to be grouped into topical clusters with a label per cluster, and the output is usable as a starting point. Most paid clustering tools (Keyword Insights, ClusterAI, Surfer's clustering tool) use a similar approach with extra heuristics.
Where AI clustering shines
- Initial grouping of large lists (500-plus keywords) before manual verification.
- Suggesting cluster labels and themes you might have missed.
- Identifying obvious near-duplicates that can be collapsed.
- Handling categories you do not personally know well: regional industries, technical niches, foreign markets.
Where AI clustering fails
- Intent mismatches. "Plumber Perth" and "best plumber Perth" look similar to an LLM. They have different intents (transactional versus commercial investigation) and need different pages.
- Australian-specific phrasing. LLMs were trained on mostly US data and sometimes mis-cluster AU regional terms.
- Confidence without evidence. The model will happily group keywords that have nothing in common on the live SERP. Always verify against the SERP for the leading keyword in each cluster.
AI clustering is a draft, not a final. Always pass the output back through the SERP-overlap check on the priority keywords.
The hybrid workflow we actually use
Across Perth client retainers we run a four-step hybrid that blends the speed of AI with the accuracy of SERP overlap:
- Pre-cluster with an LLM. Paste the candidate list into ChatGPT or Claude with a prompt that asks for grouping by intent and topic. Output: a draft of 30 to 60 clusters.
- Manually verify the top 20 clusters with SERP overlap. These are the ones the client will write first, so they need to be right.
- For the remaining clusters, do a quick SERP sanity-check on the leading keyword. Five seconds each. Catches obvious mismatches without doing the full overlap workflow.
- Pick primary keywords, map to URLs, record variants. The output goes into the master keyword map spreadsheet.
This hybrid takes two to three hours for a typical small-business list and produces a cluster map that is 90 percent as good as the manual method at a quarter of the time. For larger lists or higher-stakes plans (e.g. a 600-keyword e-commerce migration), we drop the AI layer and run the full manual method, because the edge-case mistakes get expensive at scale.
Mapping clusters to URLs
Once you have clusters, the mapping rules are simple but routinely violated:
- One cluster maps to one URL. Always. Not "this cluster lives across the pillar and three sub-pages". Pick one. Cross-link to the others.
- One URL targets one primary keyword. The primary keyword sets the title tag and H1. Variants from the cluster get woven into H2s, body copy and FAQ questions.
- If two clusters compete for the same URL, you have a clustering error. Re-check. Either combine the clusters or split the URL into two.
- Existing pages get audited as rewrite candidates first. Do not create a new URL when an existing one can be reworked to match the cluster.
Get this mapping right and you avoid the classic keyword cannibalisation problem where two pages on the same site fight each other for the same query. Get it wrong and you spend the next year confused about why nothing ranks.
Common mistakes
- Using SERP overlap as the ground truth for clustering.
- AI for first-pass speed, manual verification for the priority clusters.
- One cluster per URL, one primary keyword per URL.
- Auditing existing pages before creating new ones.
- Building hub-and-spoke internal linking around the cluster map.
- Trusting AI clustering output without SERP verification on the priority clusters.
- Grouping by surface similarity (same word stem) rather than by intent.
- Spreading one cluster across multiple URLs. Hello cannibalisation.
- Stuffing two distinct intents into one URL because it is convenient.
- Skipping the SERP check because the keyword tool's "topic" label looked good enough.
Perth and WA context
Three patterns from running clustering for Perth and WA clients:
Service plus suburb clusters split cleanly. For a Perth tradies business, "plumber Fremantle", "emergency plumber Fremantle" and "after hours plumber Fremantle" cluster into one URL because the SERP overlap is near-total. "Plumber Joondalup" is a different cluster, different URL, because the SERP changes. The suburb modifier is a hard boundary. See SEO Fremantle and SEO Joondalup for the live pattern, and the local keyword research chapter for the modifier framework.
Industry-specific clusters need expert-eye review. Mining, legal and medical clusters often contain technical terms an LLM does not understand well. "FIFO recruitment" and "FIFO accommodation" look similar to a tool. They are completely different services to a WA mining client and need separate pages. See mining SEO, legal SEO and healthcare SEO for category-specific clustering quirks.
Service plus region clusters benefit from manual SERP review. "Mining services Karratha" and "mining services Pilbara" can cluster or not depending on the live SERP. Sometimes Google treats Karratha and the broader Pilbara region as overlapping intent and one page wins both; sometimes it does not. Read the SERP, do not guess. See SEO Karratha and the competitor keyword gap chapter for how regional clusters interact with competitor coverage.
For the wider workflow, see how to do keyword research. For what to do with a clustered map once it exists, see the On-Page SEO pillar and especially internal linking strategy for the hub-and-spoke. For the broader strategic frame on topical authority, the entity SEO chapter and the upcoming Content Strategy pillar both go deeper.