How can you use search results to cluster keywords?

To cluster keywords using search results, choose a primary keyword, then compare its search result URLs with those of each keyword you want to classify. The more the URLs overlap, the more likely it is that one page can cover those keywords. This gives you a basis for grouping keywords beyond similar wording alone. The 2023 clustering demonstration described below compares only the top three results. That is the author’s choice, not a clustering standard set by search engines, and it does not guarantee that a single page can rank for every query.

Find candidate keywords, then collect results for each query

Start with variations of the primary keyword. You can also check which other related keywords the pages in the search results rank for. A 2024 Semrush article uses “credit cards” as an example, noting that ranking pages also cover queries such as “credit card application.” These keywords could therefore be considered when optimizing a page. This is only a way to find candidate keywords, though, not a substitute for comparing search results.

Next, collect search result URLs for the primary keyword and each candidate keyword. The 2023 clustering demonstration takes the top three results for each query because the author considers these pages more likely to satisfy search intent. That explains the author’s choice; it is not an official rule.

Sort candidate keywords by URL overlap

The demonstration compares each candidate keyword with the primary keyword, using these steps:

1. Put the primary keyword’s result URLs into a set, removing duplicates.
2. Do the same for each candidate keyword, then identify the URLs shared by the two sets.
3. Divide the number of shared URLs by the number of URLs for the primary keyword, then multiply by 100%.
4. Sort the candidates from highest to lowest overlap to find keywords that a single page is more likely to cover.

The denominator is the number of URLs for the primary keyword, not the total number of URLs in the combined sets. This method can help assess whether different queries could be covered on the same page, but the demonstration provides no universal threshold for grouping them. You therefore cannot claim that keywords must belong to the same cluster whenever overlap exceeds a particular percentage.

The long winter coat example: wording may not reveal the full search intent

In the results shown in the 2023 article, the following five queries had the same top three results:

• long winter coat
• long winter coat women
• womens long winter coats
• long winter coats for women
• long winter coats

The author therefore concluded that one page could cover these queries. The author also noted that a men’s coat page should not be used to target “long winter coat” and “long winter coats,” even though neither query explicitly mentions women. This example shows why search intent cannot be judged by wording alone. However, the conclusion applies only to the search results shown at the time; it does not establish that these keywords still return the same results today.

How should you present keyword clusters to make comparisons easier?

The demonstration recommends limiting its bar chart to roughly twenty to thirty queries for readability. With more queries, the labels overlap, making a table more suitable. This is a limitation of that chart, not a limit on the number of keyword clusters. A table can list candidate keywords and their overlap percentages, providing a concrete basis for deciding whether one page could cover them.