Clustering Keywords

Sick of all those keywords? Start looking at clusters!

Keyword Clustering removes the (not provided) in your organic Google analytics reports and brings back the keywords the user searched for at Google. This may lead to a vast amount of keywords to be analyzed and understood. But working with tens of thousands of keywords can be frustrating. We’ve developed a state of the art clustering engine that makes sense of those keywords by grouping them based on the user’s intent. This will totally change how you do SEO and SEA. We’ll show you how to use it and make the most of it!

The clusters of our engine groups keywords in a way that the clusters are the same as those of the world’s top marketers in 94% of all cases!

What does it do?

  • clusters your keywords semantically with 94% intersection with the top human marketers.
  • takes search results from search engines into account.
  • easy processing of the data in Excel or your favorite data tool.

Why you should cluster keywords:

The number of keywords that SEOs and SEAs deal with every day is immense. 20-30k long tail keywords for a single domain are not uncommon and they all need to be monitored and optimized if necessary.
Clustering at that scale can’t be done manually. Although it is important to get an overview of your own page and the performance of individual topics.

Understanding Search Engines:

Since Google’s Hummingbird update, Google understands a user’s search intent much better and can provide her with a lot of suitable search results. Google uses a huge treasure trove of metrics to train their machine learning system. The results of Google’s machine learning efforts are represented by their SERPs that can be used for on an intent basis.
This approach can be used to find out whether a keyword is aimed more at a purchase or purely informational.
The use of n-gram or lemma-based clustering methods does not allow for such conclusions. Using those models you’d put the following search phrases into one bucket.

-Holiday cottage pictures
-Buy a holiday home
-Holiday cottage Spain rent

However, behind each search query, there are extremely different intents. While the first phrase is about images, the second is focused on the purchase and third wants to rent a place in a specific region.
Google knows this and the user intent and the SERPs look extremely different. This should be taken into account both in keyword research and performance monitoring. If you want to sort the keywords by hand, you would not only have to find the keywords and bring them together but also think about them – especially when it comes to issues.

Working with our keyword clustering engine

  1. Log in here, select Clustering
  2. upload CSV with your keywords in any column.
  3. choose the keyword column
  4. select GO and wait
  5. download the results


For sematic clustering of the keywords there is an infinite number of use cases. We show you the most common applications.

Keyword Planning

The Google Keyword Planner is a great tool to generate keyword ideas and to estimate the traffic of a keyword. But after each search there is still a great deal of manual effort involved in selecting and grouping the really relevant keywords and creating a structured text from them.
Our clustering helps here. After working with Keyword Planner, keywords can be thrown into the tool as CSV and the appropriate clusters are created and returned.
You can quickly structure a text, remove irrelevant keywords and see understand the popularity and meaning of the individual keywords.

How it works

Suppose we want to write an article about chia seed and start with a simple keyword search in AdWords:

keyword planner for keyword clustering

AdWords gives us 194 keywords back for our first keyword search.
We download the Keyword Planner’s output as CSV. The file can be uploaded to our clustering engine without any adjustments. You only have to select the column and specify whether you’d like the header row to be grouped as well.

output of keywordplanner and clustering engine

Once the clustering service has completed the calculation, you can download the file. For small files, this is very fast while for large ones (larger than 20,000 keywords) this can take a few hours.

The topic for the individual keywords can be found in the column next to the keyword. In order to get an overview of the keywords, we create a pivot table of topics, keywords, and the search volume. You’ll immediately see that the top topic Chia Pudding has the most search volume (52690).

chia keywords clustered

With a search volume of 35k in third place in our research is the general topic pudding. Although the individual keywords have a lot of traffic strong, they’re not really interesting for us, as we’re interested in Chia:

clustered chia keywords about pudding

By clustering the keywords, we were able to sort out all the keywords that actually don’t fit, which are to 1/3 of the search volume.

Now we can focus on searching for even more keywords for individual promising topics in order to really grasp the complete topic.

keywords about chia clustered and pivoted and excluded

How do I exclude the keywords I don’t want?

We add a column to the pivot where we mark all the topics that should be excluded:

keywords about chia clustered and exluded

Perform a vlookup with the topics to exclude them. The result should look like this:

keywords about chia clustered and exluded


This way we can eliminate the keywords that don’t fit our needs.

Content creation and optimization

How do I structure a text I want to write? Shall I write a large article or rather several smaller ones? Which content form is best to catch those users who are looking for a specific keyword – are they more likely to be looking for a transaction or information retrieval (or even navigation?)?

Even before the content is created, the clustered keywords help you to get an overview of the broader topics.

Let’s take the same keywords that we already received via the Keyword Planner. On the basis of the number of keywords in the topic and the search volume we find six big topics on which we want to take a closer look at first:

keywords about chia now mama chia excluded

Mamma Chia is a brand name with only one keyword, so we put exclude it. Four topics seem to be very recipe-heavy at first glance:

  • chia pudding
  • coconut chia
  • chia bowl
  • chocolate chia

This is also confirmed by a look at the SERPs:

chia SERPs

Hence it makes sense to write about one (or more) recipes. The other two topics deal with chia in more general terms, which is confirmed by a look at the individual keywords:

clustered chia keywords + clusters

We decide to split our article. On the one hand, we write a general article about chia, its effect, etc, and one that contains the individual recipes for chia.

clustered chia keywords, excluding mama chia

The classification is done on a topic level and we again attribute the top to the keyword through a vlookup.

The individual cluster groups already give us a good overview of how we want to structure these two articles. For example, in our generic chia article we dedicate a separate paragraph to topics such as Chia Oil, Chia Diet and “ground Chia” in order to cover them adequately.

Topic ranking distribution of a page

Which topics does my site rank for? It is often difficult to recognize the most important keywords from a whole bunch of them. Our clusters provide a quick overview. This is especially important when you’re an agency, trying to figure out a client’s site.

The historic performance of individual topics can be easily identified and monitored so you can make quick decisions about which topics to approach first. Add new topics or extend old ones that work well but aren’t at their full traffic potential just yet.

Even if a customer project has been newly adopted and Keyword Hero isn’t yet installed, the clustering service helps to get a first overview of the topic. The keywords from the Search Console can be downloaded and clustered to get an overview of the topics (attention: You can only download 1k keywords through SC, through the GA integration up to 5k per file).
Once the keywords are clustered, you can find out the traffic stats of each topic through a pivot table.

This allows for simple and quick overview of the distribution of the topics that pull traffic. At first glance it becomes clear that we should check the topic mouse because the site generates traffic about a lot of topics from this area.

Performance monitoring of individual topics

The differences in the performances between individual topics can be huge. So it may be because the user does not convert because the user’s search intent isn’t fulfilled or because they are generally in an early stage of the search. Clustering makes it possible to analyze keywords with their GA metrics and help monitor particularly important topics across multiple landing pages.

New and old keywords can be analyzed and because of the clustering a longer timeframe is available you can see trends and swings much faster than on the basis of single keywords.
If a topic is particularly doing well, it is easy to find keywords that fit in there and the topic can be extended.

It is almost impossible to monitor individual keywords and estimate their performance due to the extremely high ratio of long tail keywords. Even for small pages,>2000 unique keywords are not uncommon. Only a small part of these keywords get really a lot of traffic, the majority of them are responsible for only a tiny percentage of the sessions. It is impossible to detect performance fluctuations here due to the lack of a proper dataset – also, the effort to monitor these keywords on a regular basis is simply too high.

sessions of individual keywords

Clustering the keywords based on “high-level topics” offers several advantages: on the one hand, the data basis per cluster is large enough to represent fluctuations in frequency and performance and on the other side, clustered keywords are much fewer and can easily be monitored.

To do this, w we export the keywords from the KWH GA-View:

Keyword Hero Google Analytics

Now upload them into our clustering service. Using a pivot table, you can now examine the individual topics for their performance.

pivot table of clustered keywords incl. session count and keyword #

The individual topics let us quickly recognize which topics bring how many keywords and also how much traffic they pull.

This way, conversion rates on a topic basis can also be analyzed:

conversions rates of clustered keywords in a pivot table

The clustering of keywords and the significantly higher volume of data at topic level make it possible to determine performance values faster and more reliably. In our example, we see at a glance, for example, that the themes involving “kot” have the best conversion rates.
It would not have been possible to find that out on the basis of individual keywords since each of them simply contains too little data. However, we can now conclude that users with these or similar keywords are inclined to buy our products – we can now customize our content in these areas – or even create additional AdWords campaigns for these keywords.

Optimizing existing content

Clustered keywords also help you optimize existing content. You can upload your existing keywords from our Keyword Hero GA account into Google’s Keyword Planner to see what is still missing – in which topic area am I doing well and where could I do better?

Now that we know what topics are driving conversions, we want to extend them of course. In the first step, we will take a look at the keywords that already bring us traffic and the current position:

clustered keywords with topic and position

Here we can see that we are ranking with many different keywords for this topic (there are 62 of them) but most are at position > 6.

So here is still potential to generate much more traffic. Also, we use these keywords as a seed set to get additional keyword ideas via from Google Keyword Planner or similar tools for our site.

Analysis of AdWords keyword performance

Clustering isn’t only useful for SEO, but also keywords in AdWords. It can bring new insights into the performance of individual keywords.

Done right, subpar campaigns can be examined and optimized with regard to user intent. In addition, labels can easily be attached to individual campaigns from the cluster to keep track of performance in both AdWords and Analytics at all times.

In the first step we export the keywords from the AdWords account:

adwords account for cluster analysis

Once we have clustered the keywords, we proceed to the analysis of the individual topics. Clustering makes it possible to quickly identify those topics that are cost drivers. Especially with poorly structured accounts, this helps to get an initial overview of how the account is currently performing, which topics are working, and which ads should be deactivated.

makler clustered keywords

The cluster engine is optimized to form clusters that correspond to the size and composition of Ad Groups.

The individual search queries in this example are strongly location-related, the cluster engine recognizes this and clusters the keywords according to the individual locations. In this example, the clusters are extremely helpful, as the entire campaign is structured according to the country and then broad / exact. Due to the clustering, however, we now have a more precise understanding of the performance of individual locations.
In this case, we have decided not to completely rebuild the account, but to tag the keywords with labels, that will help us with future evaluations.

Clustering broad keywords

Clustering broad keywords is a field that can quickly lead to performance gains.
What do the clusters of a broad keyword look like?
Find the keywords that have nothing to do with what you’re offering and stop losing money.

To achieve some quick wins, we want to cluster the search queries we use in the account via our keyword settings and find the ones that make little sense. In the first step, we upload the search query report to the clustering engine.
Thanks to clustering we don’t have to look at the individual keywords of the search report but can quickly analyze the keywords on a topic level and exclude irrelevant topics as a whole with all connected keywords.

excluding clustered keywords

Automatic creation of campaigns and ad groups

In addition to monitoring and re-evaluating existing campaigns, clustering makes it easy to create ad groups and campaigns. These can be imported directly from the Planner via clusters – the topic name will be added automatically. In this way, campaigns can be set up in a short period of time that can be easily monitored and managed.
As an example, we take the Chia keywords described above. We have already carried out a keyword search to create content – but now we also want to launch an AdWords campaign. We have already established that the entire topic is divided into a) generic keywords and b) recipes. We use these two divisions as campaigns. As Adgroups we use the cluster-topics and we even have all keywords already thanks to the research.

Our campaign structure now looks like this:

campaing structures based on keyword clusters

We were able to create the entire structure almost completely automated in just a few minutes.

Clustering keywords from Sisitrix, Searchmetrics, and other rank monitoring tools

As much as we love the rank monitoring tools such as Sistrix, the gigantic amount of keyword data can quickly overwhelm you. Clustering these in a single topic not only allows you to quickly have an overview of the domain, but also to quickly identify potential problems.

Data exported from those tools can be easily clustered and the performance of individual clusters can be analyzed.
We want to get an overview of the Sistrix keywords of one domain. According to Sistrix, in the top 100, there are 1309 keywords that the site ranks for.

sistrix keywords ready to be clustered

The clustering will help us to get an overview of the topics to which this page ranks. The other data provided by Sistrix helps us to better understand how the site is currently doing:

sistrix keywords clustered