What is keyword grouping?

Keyword grouping is the practice of putting related search queries into one set. Each set should answer one question for the searcher, so the whole set can be served by one page. In this guide, keyword grouping and keyword clustering mean the same thing.

Why group keywords instead of tracking them one by one?

A single query tells you little. A set of queries on the same topic shows whether you own that topic. In our experience, tracking individual keywords stops being useful somewhere past about 50 of them, because most of the movement is noise and the trend disappears into the list. This threshold is our judgment, not a measured number.

Grouping also shows coverage. A topic with ten queries and one page that ranks for two of them has a clear gap. Tracked one at a time, that gap is easy to miss.

How do you group keywords?

There are three common methods. They differ in what decides whether two queries belong together.

Each method has a use. Shared words are quick for a first pass. Shared results check how Google sees two queries. Meaning is the method that gives you groups to plan around, so the rest of this guide uses it.

How do you group keywords semantically?

Semantic grouping sorts queries by the question the searcher is asking. These are the five steps we use, with an example from our own export.

1. Export your queries. Pull the queries report from Google Search Console for at least three months, with impressions, clicks and average position. Three months smooths out one-off spikes.
2. Read each query as a searcher. Ask what the person wants: a definition, a method, a tool, a comparison or a fix. Write that intent next to the query.
3. Put queries that answer the same question together. Name each group after the question it answers, not after a shared word.
4. Pull out the outliers. Brand terms, typos and one-off queries go into a review pile. Do not force them into a group.
5. Check for mixed intent. Section 5 below covers this step.

Here are nine queries from our Search Console export for the three months to 30 October 2026, sorted into three groups:

• Definitions and how-to: how to group keywords (172 impressions), what is keyword grouping (17), keyword grouping guide (11)
• Tracking: keyword groups tracking (158), tracking keyword groups (101), keyword group tracking (20)
• Tools and automation: keyword grouping automation (12), keyword grouping API (7), keyword grouping dashboard (2)

Each group has one intent. The first wants an explanation. The second wants a way to follow groups over time. The third asks whether software can do the work. Those are three different pages, which is why the queries are split.

How do you check that the groups are right?

Two checks catch most mistakes.

The first is intent. Each group should have one search intent. A group that mixes how-to queries with comparison queries needs two pages, not one. Split it.

The second is shared results. This is our method: search each query and note the top-ranking pages. Queries that share a top-ranking page usually belong in the same group. Queries that return different pages usually do not, even when they share words. Google does not publish this as a rule, so treat it as a strong signal and check the results yourself.

What do you do with the groups?

Map each group to one page. For each group, pick one of three options:

• Existing page. A page already ranks for most of the group's queries. Keep it and update it to answer the whole group.
• Expand a page. A page covers part of the group. Add the missing questions to it.
• New page. No page covers the group. Write one, using the group's main question as the title.

If a group maps to two pages, they are probably competing for the same queries. This is cannibalization. Pick one page and merge or sharpen the other. For the wider planning method, see keyword clustering for AEO and how to build topical authority.

Spreadsheet or tool?

A spreadsheet is enough when you group a list once and leave it alone. Sort the queries, add an intent column and filter. Google Sheets and Excel both handle this well.

The case for software is repetition. If you regroup every month, work across several sites, or need the groups to feed a dashboard, a spreadsheet becomes the bottleneck. Enterprise teams, agencies, small businesses and anyone who automates reporting tend to reach that point first. API access matters for the same reason: it lets the groups feed other systems without a manual export. The deciding question is how often you regroup, not how long your list is.

Where TrackClusters fits (and where it doesn't)

TrackClusters connects to Google Search Console through read-only OAuth, groups your real queries into clusters and syncs daily. For each cluster it tracks impressions, clicks and average position over time, and it leaves position 0 out of the averages. You can export to CSV, and it flags cannibalization across clusters.

It does not replace keyword research. It has no search volume data and does not suggest new keywords. It works only with queries your site already ranks for. If you need new topics, find them with a keyword research tool first, then use TrackClusters to see how the queries you already rank for group together.

You can start a 7-day free trial at trackclusters.com/#pricing.

Frequently asked questions

What is keyword grouping?

Keyword grouping sorts search queries into sets that share a topic and an intent, so each set can be tracked and served by one page.

Is keyword grouping the same as keyword clustering?

In most SEO writing, yes. Both describe putting related queries together. This guide uses the two terms interchangeably.

How many keywords should go in one group?

There is no fixed number. A group should be as large as one intent and one page can serve. Three queries with one intent are better than thirty that mix two.

How do you group keywords in Excel or Google Sheets?

Export the queries to a sheet and add an intent column. Read each query and fill in the intent, then sort by it. Filter for rows that do not fit their block, and re-check the groups when you add a new batch of queries.

Can AI group keywords for me?

For a small list, yes, but check the output. Look for groups that mix intents before you use them, the same as with a manual grouping.

How often should you regroup keywords?

Regroup when your queries change in a way that matters, such as after a content push, a site change or a new product area. Otherwise a quarterly review is a sensible default. That is our suggestion, not a rule.

Does grouping keywords improve rankings?

Not on its own. Grouping helps you plan coverage: which topics need a page, which pages compete, and which groups are losing ground. Rankings depend on the content and the rest of the site, so treat grouping as a planning step, not a guarantee.

Here's how AEO keyword research differs from the SEO kind, what tools exist, and a method that uses your own data.

How is AEO keyword research different from SEO keyword research?

SEO keyword research hunts for terms with search volume. AEO keyword research hunts for questions: the full, conversational phrasing people use when they expect an answer. Google's own AI features break a query into subtopics and issue related searches (Google Search Central), so the questions around a topic matter as much as the head term.

What tools can you use for AEO keyword research?

Common options include Google's People Also Ask boxes, AlsoAsked and AnswerThePublic for question discovery, and AI chatbots for brainstorming. Fan-out simulators such as Qforia can show how an AI system might split a prompt (see what query fan-out is).

All of them share one weakness. They show questions that exist in general, not the ones that already reach your site. And chatbot-generated question lists are guesses, not data.

How do you find the questions your audience asks AI tools?

Nobody can show you the exact prompts typed into ChatGPT or AI Mode. The closest real proxy you have is conversational queries in your Search Console data, because people phrase searches more like prompts when they expect an AI answer.

Here's the method.

1. Open the Performance report. Set the date range to the last 6 or 12 months.

2. Add a Query filter. Choose Custom (regex) and filter for long queries:

^(\S+\s+){7,}\S+$

That matches eight or more words. Then run a second filter for question starters:

^(what|how|why|when|can|does|do|is|are|should|which)\s

Many people type queries without a question mark, so the question-word filter catches more than a question-mark filter would.

3. Export the results. Keep the impressions, clicks and position columns.

4. Cluster the queries by topic. Group questions that belong on the same page. This step is where most of the value sits, because a raw list of 400 questions isn't a plan.

5. Map each cluster to a page. Existing page, thin page, or no page.

Our own export is a small example. It includes question queries such as "what is a clustering search engine" and "what is keyword clustering in seo." They tell you the real phrasing people use for the topic, which is more useful than any generated list.

Do prompts and fan-out queries have search volume?

No. Prompts typed into AI tools don't have a public volume figure, and fan-out sub-queries are generated by the system, not typed by people. The nearest data you have is impressions for conversational queries in Search Console. They're real counts, and they only cover queries where your site already appeared.

That's a limit worth stating plainly. Search Console can't tell you about questions where you didn't show up at all. For those, use a research tool, and treat its volume numbers as estimates.

How do you turn a question list into a content plan?

Sort questions by the pattern they follow, because each pattern has a natural format:

• Definition ("what is...") → short explainer with a direct answer first
• Procedure ("how to...") → numbered steps
• Comparison ("X vs Y") → table
• List ("best...", "examples of...") → list with criteria
• Threshold ("how long...", "how many...") → a short, specific answer

These categories are a working taxonomy, not a standard. Then:

1. Cluster questions that share intent. See keyword clustering for AEO for turning clusters into page plans.
2. Give each cluster one page and one H2 per sub-question.
3. Compare the plan against what you've published. Gaps are your next articles. Cluster coverage gaps shows how to prioritize them.

Where TrackClusters fits

The filter and export steps above work in plain Search Console. TrackClusters handles the grouping step: it clusters your real queries by topic and tracks how each cluster performs over time, so a list of hundreds of questions becomes a smaller set of topics. It doesn't generate questions or show search volume. The pricing page has the 7-day free trial.

FAQ

What is AEO keyword research?

Finding the questions people ask around your topic, in full conversational form, and planning content that answers them directly.

Can I see what people ask ChatGPT in Search Console?

No. Search Console shows queries typed into Google. Conversational queries there are a proxy for how people phrase prompts.

How do I find question keywords in Search Console?

Use a Query filter with custom regex, matching long queries or queries that start with a question word.

Do I still need keyword research tools?

For questions where your site doesn't appear yet, yes. Search Console only covers queries where you've had impressions.

Why not just ask ChatGPT for a list of questions?

It's fine for brainstorming. The list is a model's guess, not a record of what people searched, so check it against real data before you plan around it.

How many questions should I target per page?

As many as the cluster contains, each as its own H2. If a page needs to answer unrelated questions, split it.

How often should I repeat this?

Quarterly is a reasonable rhythm. That's a judgment call, not a rule.

What is the difference between SEO, AEO and GEO?

Common usage runs like this:

• SEO (search engine optimization): getting your pages ranked in search results.
• AEO (answer engine optimization): getting your content chosen as the direct answer, as in featured snippets, voice answers and AI answers.
• GEO (generative engine optimization): getting your content cited inside responses generated by AI systems.

These definitions vary between sources, and the boundaries blur. One caution on searching: "AEO" alone also surfaces a clothing retailer's stock ticker, and "GEO" is sometimes used for geotargeting. Use the full phrase when you look this up.

Are AEO and GEO the same thing?

Close enough that treating them separately isn't useful. Both describe making content easy for AI systems to retrieve and cite. The difference in practice is mostly emphasis: AEO leans toward direct answers, GEO toward appearing in generated responses. If a vendor sells them as different services at different prices, ask what exactly is different in the work.

Does AEO or GEO replace SEO?

No. And from Google's side, the question barely exists. Google's guidance, published 15 May 2026, states: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO" (Practical Ecommerce, quoting Google). Its documentation adds there are no additional requirements to appear in AI Overviews or AI Mode, and no special schema or AI text files needed (Google Search Central).

That's Google's position on Google's features. Other AI systems, such as ChatGPT, use their own retrieval, which Google's documentation doesn't cover.

What changes in keyword strategy?

The comparisons usually stop before this. Here's the table they skip:

Why the shift? Because AI features may issue several related searches across subtopics to build one answer (Google). Those sub-queries change between runs, so the stable thing to plan around is the topic. That's the argument in what query fan-out is.

Concretely:

• Research: add conversational questions from your own Search Console data. See AEO keyword research.
• Planning: group queries into clusters and give each a page. See keyword clustering for AEO and how to build topical authority.
• Measuring: track clusters, not only keywords, and add AI impressions per topic. See how to track AI Overview visibility.
• Finding gaps: see cluster coverage gaps.

Which should a business prioritize first?

SEO fundamentals. Crawlable pages, useful content, sensible internal links. Everything in AEO and GEO sits on top of those, and Google says its AI features draw on the same foundations. After that, the cheapest improvement is usually restructuring what you have: questions as headings, direct answers first, one page per topic cluster.

Skip paid tools until you've used the free data. Your Search Console already shows impressions, clicks and the queries people use.

How do you measure AEO and GEO results?

Imperfectly. Be wary of anyone promising precise "AI visibility scores."

• Google's AI features: the Generative AI report in Search Console shows impressions by page, country, device and date, with no queries or clicks (chudi.dev).
• Other AI systems: no first-party data exists. Third-party tools sample prompts, which gives you an estimate, not a count.
• Topic coverage: cluster-level impressions, clicks and position from Search Console. That's real data and the most dependable signal.

Where TrackClusters fits

TrackClusters sits in the measuring and planning rows of the table: it clusters your real Search Console queries and tracks each cluster's impressions, clicks and position over time. It doesn't track AI citations or show search volume. There's a 7-day free trial on the pricing page.

FAQ

What is the difference between SEO and AEO?

SEO aims to rank pages. AEO aims to have your content chosen as the direct answer. In practice, AEO is a layer of formatting and coverage on top of SEO.

What is the difference between AEO and GEO?

Little, in practice. AEO emphasizes direct answers, GEO emphasizes being cited in AI-generated responses. Definitions vary by source.

Does Google say AEO and GEO are different from SEO?

Google says optimizing for generative AI search is "still SEO" from its perspective, with no special requirements for AI Overviews or AI Mode.

Do I need llms.txt or special schema?

Not for Google's AI features, according to Google's documentation.

Which should I do first?

SEO fundamentals, then restructuring content around questions and clusters.

How do I measure AI visibility?

Use the Generative AI report in Search Console for Google's features, and cluster-level performance data for topic coverage. Treat third-party AI visibility scores as estimates.

Is "AEO" or "GEO" the better term to use?

Use whichever your audience searches. Both are in use, and neither is an official Google term.

Clustering works as a process, not a one-off task. Here's the process, step by step, and how to tell whether it's working.

What is keyword clustering for SEO?

Keyword clustering for SEO groups the search queries your site already appears for into topics, then gives each topic one page and tracks that topic as a unit. The goal is coverage. A pillar page, its supporting pages, and the queries they should rank for all belong to one cluster.

Why does clustering matter more than single keywords?

Individual keyword tracking gets noisy once a site ranks for more than about 50 keywords. Rankings fluctuate week to week, and most of that movement is noise. Clusters smooth it out, and they show real trends: a content pillar rising or sliding as a group.

The strongest signal is at the topic level. Which pillars are winning and which are losing is more useful for planning than the position of any single keyword.

How do you run keyword clustering for SEO?

Six steps, in order:

1. Gather the queries. Export at least three months of queries from Search Console. Keep impressions, clicks and average position. Six to twelve months is better for seasonal sites.
2. Clean the list. Remove duplicates and typos you can clearly identify. Keep position 0 rows for now, but exclude them from any average later.
3. Group by intent and topic. Put queries that want the same answer into the same group. A cluster is a set of queries one page should answer, not a set of keywords with the same word in them.
4. Check each cluster for mixed intent. If a group holds both a how-to query and a product comparison, split it. One intent per cluster.
5. Map each cluster to a page. Existing page, a page that needs expanding, or a new page. Each cluster gets exactly one primary page.
6. Measure each cluster monthly. Impressions, clicks and average position per cluster, with position 0 excluded from the average.

Step 3 is where most of the work is. Step 6 is where the process pays off, because it's what catches a cluster that's slipping before the damage shows in your traffic.

How do you decide when two keywords belong in one cluster?

Use your own data as the test. For each query in your export, find which page ranked for it. Queries that share the same top page usually belong in one cluster, because one page is already serving them both.

Then sanity-check with your eyes. A shared top page is strong evidence, but two queries can share a page and still need different answers. If a comparison query and a definition query both land on one long guide, that page may be doing two jobs, and the cluster needs a closer look.

What do you do with clusters that have no page?

Those are coverage gaps. A cluster with high impressions and no dedicated page is the clearest case: demand is visible, and the site hasn't built anything for it. The fix is a new page, not more keyword tweaks on an existing one. Cluster coverage gaps walks through how to find them.

How do you know the process is working?

Watch the cluster-level numbers over several months:

• Impressions should rise as you cover the topic more fully.
• Average position should improve, measured without position 0.
• Clicks should follow impressions. If impressions rise and clicks don't, check whether the page's title and opening answer match the queries.

A cluster that stops improving after a few months is a signal to revisit the page or the intent, not necessarily a reason to start a new page.

Where TrackClusters fits

TrackClusters runs the clustering and monthly measurement from steps 3 and 6 for you. It connects to Search Console through OAuth, groups your real queries into clusters, syncs daily, and tracks impressions, clicks and average position per cluster over time. Position 0 is excluded from averages.

Steps 1, 2, 4 and 5 still need your judgment. The tool groups queries, but you decide which page each cluster belongs to. There's a 7-day free trial and no credit card required on the pricing page.

For the basics, see what keyword clustering is. For real cluster examples, see keyword clustering examples.

FAQ

What is keyword clustering in SEO?

Grouping the queries your site ranks for into topics, with one primary page per topic, then tracking each topic as a unit.

How many keywords should be in one cluster?

There's no fixed number. A cluster is defined by intent: the queries one page should answer. A cluster can hold a few queries or dozens.

How often should I re-cluster?

Re-check the clusters monthly for performance. Re-run the full grouping quarterly, or when your queries change a lot.

Does keyword clustering help with rankings?

It helps you plan coverage and spot gaps, which supports rankings. It doesn't guarantee any position.

What's the difference between keyword clustering and topic clustering?

In practice, the terms overlap. Keyword clustering starts from queries, and topic clustering often starts from a subject. The process here starts from queries, so each group is tied to real search data.

Here are three free methods, how to run each one, and where each breaks.

What counts as free keyword clustering?

Anything that groups keywords without a subscription. Three options cover most cases:

• Spreadsheet: your queries in Google Sheets or Excel, grouped by hand and with filters.
• Open-source script: a short Python program that groups queries by word overlap or by meaning.
• Hybrid: a script does a first pass, and you check and fix the groups by hand.

Each one is free in money. Each one costs time, and the time grows with your list.

How do you cluster keywords in a spreadsheet?

Start with your own data. Search Console is free and already has the queries that bring people to your site.

1. Export the queries. In the Performance report, set the date range to the last three months, switch to the Queries tab, and export. Keep the query, clicks, impressions and position columns.
2. Add a Topic column. This is where your groups go.
3. Filter by a shared term. Use the filter on the query column to show every query containing a term, such as "keyword clustering". Label that group, then move on.
4. Catch variations. "Keyword clustering" and "clustering keywords" share words but not the same order. Run a second filter for each variant.
5. Check the leftovers. Sort the rest by impressions. Most of the high-impression queries you haven't grouped are where the next cluster lives.
6. Calculate per cluster. Use SUMIF for impressions and clicks. For average position, exclude any row with position 0 before you average.

The position 0 step matters. In the exports we've worked with, position 0 means the query didn't rank in the top 100. It isn't a real ranking, and averaging it in makes a cluster look worse than it performs.

This method works well for a few hundred queries. Past that, the filters become slow and easy to get wrong.

Can you cluster keywords with Python for free?

Yes. Two free approaches do the job:

• Word overlap: remove stop words, convert each query to a vector of word counts or TF-IDF weights, and group similar vectors with k-means or hierarchical clustering. This is quick, and it groups by shared words.
• Meaning: turn each query into an embedding with an open-source sentence-embedding model, then cluster the embeddings. This groups by meaning, so "best running shoes" and "running sneakers for beginners" can land together.

Both need a Python environment and some setup. Expect to spend time reviewing the output. Automatic groups almost always need a human pass to remove mixed intents, where one cluster holds a comparison query and a how-to query that deserve separate pages.

What are the limits of free methods?

Free methods break on three things:

• Time. Every re-run takes as long as the first one, unless you script the whole pipeline.
• Staleness. A spreadsheet is out of date the day you finish it. You can't see how a cluster moved last month unless you saved last month's export.
• History. Search Console only keeps 16 months of data, so a spreadsheet can't show trends past that window unless you've saved it yourself.

None of these matter for a one-off audit. They matter a lot if you plan to track clusters over time.

When is it worth paying?

When you're re-doing the work every month, or when your site has enough ranking keywords that tracking them one by one turns into noise. As a rough threshold, once you have more than about 50 ranking keywords, looking at individual keywords stops being useful. Clusters are the more reliable unit.

Where TrackClusters fits

TrackClusters automates the spreadsheet workflow above. It connects to Search Console through OAuth, reads your query data, groups it into clusters, syncs daily, and tracks impressions, clicks and average position per cluster over time. Position 0 is excluded from averages. You can export the result to CSV.

It isn't free. There's a 7-day free trial, no credit card required, on the pricing page, so you can compare its groups with your own spreadsheet before you pay anything.

For the method behind the groups, see what keyword clustering is. For tool types, see keyword clustering tools.

FAQ

Is there a completely free keyword clustering tool?

Google Sheets and Search Console are free. If you're comfortable with Python, open-source libraries are free too. There isn't one free tool that handles everything, so each option trades cost for time.

Can Google Sheets cluster keywords?

Yes, using filters and formulas. It works well for a few hundred queries. It doesn't scale, and it doesn't track changes over time on its own.

What is the best free method for beginners?

The spreadsheet. It teaches you what the groups should look like, and you'll know when a tool's output is wrong.

Why should I exclude position 0?

Position 0 usually means the query didn't rank in the top 100. Averaging it with real positions drags the cluster's average down.

Are free clusters as accurate as paid ones?

The grouping logic can be the same. The difference is the time you spend checking, re-running and tracking the results.

Here are the four types, what each one is good at, and how to choose.

What is a keyword clustering tool?

A keyword clustering tool groups keywords or search queries into sets that share a topic or intent, so each set can be planned as one piece of content. The tools differ in three things: where the keywords come from, how the tool decides two keywords belong together, and whether it shows the groups over time.

What types of keyword clustering tools are there?

The first two work from keyword lists. The last two can tell you something about your own site.

Should a clustering tool use search volume?

Volume is an estimate, and most tools present it with more confidence than the data supports. A keyword showing 1,000 searches a month might get 200 or 2,000. For sites that already rank, your own impressions from Search Console are a more honest signal, because they count what actually happened on your site.

Volume still has a use for planning new topics you don't rank for yet. Just don't treat it as a precise number when you group keywords.

How do you choose a keyword clustering tool?

Run each candidate through the same five checks:

1. Input source. Does it cluster keywords you pasted in, or queries from your own Search Console?
2. Missing data. What does it do with queries that don't rank in the top 100? In GSC-style exports, those show as position 0. Averaging them in makes clusters look worse than they perform, so the tool should exclude them.
3. Time. Can you see how a cluster moves week to week, or only a snapshot?
4. Export. Can you get the groups out as CSV, so they can go into a content plan?
5. Cost per use. Will you re-run it monthly? A tool you use once a year is priced differently from one you use every week.

If you only need a one-off grouping of a few hundred keywords, a spreadsheet is enough. If you need to track clusters across months, you need a tool that runs on a schedule.

Where TrackClusters fits

TrackClusters is a Search Console-based tool. It connects to your GSC property through OAuth, reads your real query data, groups it into clusters, and syncs daily. It tracks impressions, clicks and average position per cluster over time, and exports to CSV. Average position excludes position 0.

It's not a keyword research tool. It has no search volume data, and it won't show you topics you don't already get impressions for. If you need those, use a research tool and treat its numbers as estimates. There's a 7-day free trial and no credit card required on the pricing page.

If you're new to the idea, what keyword clustering is covers the basics. For real examples built from GSC data, see keyword clustering examples.

FAQ

What is the best keyword clustering tool?

It depends on your input. For your own site's performance, use a Search Console-based tool. For planning new topics from a keyword list, a SERP-based or semantic tool fits better.

Are keyword clustering tools accurate?

They group keywords by the method they use, so accuracy depends on the input and the method. Check a few clusters by hand before you plan around them.

Can I cluster keywords for free?

Yes. A spreadsheet works for small lists, and free open-source libraries can group larger ones. The trade-off is your time and the lack of tracking over time. Free keyword clustering covers the options in more detail.

Do clustering tools replace keyword research?

No. Clustering organizes keywords you already have. Keyword research finds the ones you don't.

How often should I re-cluster?

It depends on how quickly your queries change. Monthly is a reasonable starting point for most sites, and quarterly is enough for stable ones.

What is a content gap in AI search?

A content gap in AI search is usually described as a prompt where an AI answer cites your competitors and not you. That definition is hard to act on, because you can't see which prompts were run or which sources were cited without a third-party tool.

A more measurable definition: a cluster coverage gap is a topic where your site earns many impressions across many queries, but you have only one page, or no page, built to cover it. It shows up in your own Search Console data.

How is an AI content gap different from a keyword gap?

A keyword gap compares keyword lists between you and a competitor. A cluster coverage gap compares what searchers ask around a topic with what you've published about it.

Why that matters for AI: AI Overviews and AI Mode may issue several related searches across subtopics to build an answer (Google Search Central). A topic where you only cover one slice gives that process less of yours to use. Fan-out is explained in what query fan-out is.

How do you find the topics your site is missing?

Use the clusters you already have. Take your queries, group them by topic (see what keyword clustering is), and for each cluster record two numbers: impressions and pages built.

Then plot them on a 2x2:

The top-left box is your gap list: topics where demand is visible and your coverage is thin.

A real example from our own site

Our own export gave a clean case. The /blog/ index page picked up 3,722 impressions for "keyword grouping" searches. Our best read is that no dedicated page existed for the topic, so Google showed the closest page it had. The index page isn't written to answer that query.

High impressions, no purpose-built page. That's a coverage gap, and it's the clearest argument for writing the page.

Your version will look similar: an index page, a category page or a loosely related post collecting impressions for a topic it was never written for.

Why does AI cite competitors instead of my page?

Nobody outside Google can say for sure, and sources that claim to are guessing. What you can check are the likely causes:

• Topic gap: you don't cover the subject at all.
• Depth gap: you cover it with one page where the subtopics need several.
• Intent gap: your page answers a different intent than the question asked.
• Format gap: the answer is buried, not stated plainly under a clear heading.
• Structure gap: the page isn't linked from anywhere, so it's hard to discover.

The last one is easy to check and often overlooked. Look for orphan pages: pages with no internal links pointing at them. Add links from related pages, and map each page to its cluster.

Google says there are no special requirements for AI features (Google Search Central). Closing a gap is ordinary content work, not an AI-specific trick.

What to do with the gap list

1. Rank gaps by impressions. Most impressions first.
2. Check the existing page for each. Is it thin, off-intent, or missing?
3. Decide: new page or expand. If the queries share one intent, expand. If they split into several intents, write several pages.
4. Write each page with direct answers. See keyword clustering for AEO.
5. Link it in. Connect the new page to its pillar and sibling pages.
6. Re-measure. Come back after the next Search Console cycle and see whether the cluster's impressions and position moved.

How often should you run a content gap analysis?

Quarterly is enough for most sites. That's a judgment call, not a standard. Sites that publish weekly may want it monthly. Search Console data lags by a couple of days, so checking daily doesn't buy you anything.

Where TrackClusters fits

TrackClusters shows impressions, clicks and average position per cluster, from your own Search Console data. That's the impressions side of the 2x2 above. You count the pages you've built for each cluster yourself. It doesn't find topics you have zero impressions for, because it has no keyword research data. The pricing page has the plans and the 7-day free trial.

FAQ

What is a content gap in AI search?

Commonly, a prompt where AI cites competitors and not you. A more measurable version is a topic with high impressions in Search Console and few pages built for it.

How is it different from a keyword gap?

A keyword gap compares your keywords with a competitor's. A cluster coverage gap compares demand around a topic with the pages you've published for it.

How do I find content gaps without a paid tool?

Cluster your Search Console queries, then compare each cluster's impressions with the number of pages you have for it.

Why would an index page rank for a topic?

When no page covers the topic well, Google shows the closest match. It's a sign you need a dedicated page.

What is an orphan page?

A page with no internal links pointing to it. It's harder for crawlers and readers to find.

How often should I run this analysis?

Quarterly for most sites.

Can closing a gap guarantee AI citations?

No. Nobody can promise that. It improves your coverage of the topic, which is the part you control.

Do topic clusters help with answer engine optimization?

They help in the way covering a topic properly always has. Google says there are no special requirements for AI Overviews or AI Mode (Google Search Central), and describes optimizing for generative AI as "optimizing for the search experience, and thus still SEO" (Practical Ecommerce, reporting on Google's May 2026 guidance).

Clusters matter because AI features may issue several related searches across subtopics to build one answer, a process explained in what query fan-out is. If your page covers a cluster's sub-questions clearly, it has more to offer that process. Nobody has published proof that clustering raises AI citations, so treat it as sound practice, not a guaranteed lever. Be skeptical of any source quoting a precise "AI visibility uplift" from clustering without naming where the number came from.

How are AEO clusters different from traditional SEO clusters?

The clusters are the same. What changes is the output. A traditional cluster ends with a page targeting a head keyword. An AEO cluster ends with a page where each sub-question has its own heading and a direct answer.

How do you turn a keyword cluster into content AI can cite?

Follow four steps. The example below is hypothetical, written to show the mechanic.

1. Start with a raw list.

Say these are queries from Search Console:

what is email deliverability
why do emails go to spam
how to improve sender reputation
email bounce rate causes
spf dkim dmarc setup
email warmup how long

2. Cluster by intent.

If you're new to the idea, start with what keyword clustering is. Group queries that belong on one page. For the full process, see how to group keywords.

Go through semantic keyword clustering for why grouping by shared words fails here: "spam," "bounce rate" and "sender reputation" share no vocabulary but the same intent.

3. Assign a page type.

Definition, how-to, comparison, list, or troubleshooting.

4. Write the sub-questions as H2s.

Under each, a 40-60 word direct answer, then detail.

The result looks like this:

Four clusters, four pages, each section answerable on its own. For what real clusters look like in Search Console data, see these keyword clustering examples.

How many pages should an AEO topic cluster have?

One page per cluster. A topic made of four clusters needs four pages, plus a pillar page if you want a hub linking them (see how to build topical authority). The count comes out of the clustering. If a page has to answer three clusters, split it. If two clusters share most of their queries, merge them.

The rule that makes this work: one cluster, one page. One sub-question, one H2.

How do you avoid keyword cannibalization inside a cluster?

Check for overlap before you write. Cannibalization happens when two pages target the same intent and split the impressions. The test is simple: if two clusters, or two existing pages, receive impressions for most of the same queries, they're competing. Merge them into one page, or sharpen each page so it answers a different question.

Search Console shows this directly. In the Performance report, filter by a query and look at the Pages tab. If several URLs get impressions for the same query, you have an overlap. TrackClusters flags cannibalization across clusters automatically, but the manual check above works at any size.

The part to resist

It's tempting to write a clustered page for every possible question. Don't. Cluster the queries you actually rank for or want to rank for, and ignore the rest. Thin pages written to cover a question list read as filler to people and AI systems alike.

Where TrackClusters fits

TrackClusters groups the queries your site already earns impressions for into clusters, using your own Search Console data. That gives you step 2 for the keywords you already have. It won't suggest new questions to target, because it doesn't use search volume or keyword research data. The page types and the writing are yours. The pricing page has the plans and the 7-day free trial.

Related reading

• How to track AI Overview visibility by topic cluster: measuring AI features in Search Console

FAQ

Do topic clusters help with answer engine optimization?

They help by making a topic's sub-questions explicit and answerable. Google says no special AEO optimization is required, so the benefit is better coverage, not a trick.

What is the difference between AEO clusters and SEO clusters?

The grouping is the same. AEO adds an output format: each sub-question gets its own heading and a direct answer.

How long should the direct answer under each heading be?

Aim for 40-60 words. That's a working guideline, not a Google rule.

How many pages does a cluster need?

One. A topic made of several clusters needs one page per cluster.

What is keyword cannibalization?

Two pages competing for the same query intent, splitting impressions. Merge them or differentiate them.

Can I do this without a keyword research tool?

Yes, for queries you already rank for: your Search Console data is enough. A research tool only adds when you're looking for new queries.

Should every cluster get its own page?

Only if the cluster has enough queries and enough distinct intent to justify one. Thin clusters can fold into a related page.

Every SEO guide tells you to build topical authority. Almost none tell you how to know when you have it. The standard measurement advice is "check your rankings," which is how you measure everything and therefore tells you nothing about this.

Here's what topical authority is, what Google has and hasn't said about it, how to build a topical map from keyword clusters, and a scorecard you can use to measure it per cluster.

What is topical authority, and is it a Google ranking factor?

Topical authority is the idea that a site covering a subject thoroughly is more likely to rank for it than a site covering it thinly. It is not a Google ranking factor you can point to. Surfer's study of the topic puts it directly: "Although the term 'topical authority' is often discussed in the SEO industry, it's not a term that Google uses." The closest Google reference Surfer found relates to news topic authority (Surfer).

So treat it as a working theory, not a setting. The theory has practical support: a site that answers every question around a topic gives searchers, and AI systems that expand a question into sub-queries, more of what they need in one place. That's good content strategy whether or not Google has a name for it.

How do you create a topical map?

A topical map is a plan of the pages you need to cover a subject, grouped by subtopic. Build it from keyword clusters instead of guessing:

1. Pick the core topic. One subject, such as "keyword clustering."

2. Collect every query around it. Start with your own Search Console queries, then add keyword research for what you don't rank for yet.

3. Cluster by intent. Group queries that belong on the same page. Semantic keyword clustering catches queries that share no words but share an intent.

4. Assign each cluster a page and a page type. Definition, how-to, comparison, list.

5. Mark what exists. For each cluster, note whether the page exists, is thin, or is missing.

6. Add internal links. Pillar to clusters and back, and between related clusters.

The finished map is a table, not a mind map. Each row is a cluster, a target page, and a status.

How many articles does a topic cluster need?

There's no fixed number. The number is however many distinct clusters the topic breaks into, with one page per cluster. A narrow topic might split into four. A broad one might split into twenty.

Be suspicious of any article that gives you a magic figure. The right count comes out of the clustering, not out of a rule. If two clusters share most of their queries, they're one page. If one page is trying to answer three clusters, it's three pages. Keyword clustering for AEO covers that overlap check.

How do you measure topical authority?

Measure it per cluster, not per page. A single page's rank says little about whether you cover a topic. A cluster's coverage says a lot.

Here's a scorecard. For each cluster in your map, track:

Two cautions. First, some tools and exports record a query with no ranking in a period as position 0. If yours does, exclude those rows from the average, or the cluster's average position comes out distorted. Second, "top 20" is a threshold we picked for the scorecard, not an industry standard. Pick one and keep it consistent.

A hypothetical comparison shows why this matters. A cluster with 2 of 8 planned pages and one with 7 of 8 can have similar average rankings today. The scorecard shows the first has a gap that caps how far it can go.

Does topical authority help you get cited in AI Overviews and ChatGPT?

Plausibly, but there's no proof that topical authority as such causes it, and Google says AI features need no special optimization (Google Search Central). The reasoning is mechanical. AI Overviews and AI Mode may issue several related searches across subtopics to build an answer. A site with pages for several of those subtopics gives the system more of its sub-queries to match.

Call that a reasonable hypothesis. It's the same advice as always, with a clearer reason behind it.

Why people stall halfway

The most common failure isn't a bad topical map. It's a half-finished one. A pillar page, two cluster pages, and six planned articles that never ship. The scorecard's "pillar completion" row exists for this reason: it makes the unfinished part visible.

Surfer's survey found that 80% of SEOs say a tool that clusters keywords based on search results would be very valuable (Surfer). The study doesn't state its sample size, so read that as a signal of interest, not a precise statistic.

Where TrackClusters fits

TrackClusters builds clusters from the queries your site already ranks for and tracks their impressions, clicks and average position over time, with position-0 queries excluded from averages. That covers the measurement half of the scorecard from your own Search Console data. The map, the pages and the finishing are still your job. There's a 7-day free trial on the pricing page.

Related reading

• Keyword clustering examples: what real clusters look like in GSC data
• How to track AI Overview visibility by topic cluster: measuring AI features in Search Console

FAQ

Is topical authority a Google ranking factor? Google doesn't use the term. It's an SEO industry concept, and the closest Google reference concerns news topic authority.

What is a topical map? A plan of the pages you need to cover a subject, grouped by subtopic. Each row is a cluster with a target page and a status.

How many articles should a topic cluster have? As many as the clustering produces, with one page per cluster. There's no fixed number.

How do I measure topical authority? Per cluster: pages covering it, share of its queries where you rank in the top 20, average position, impressions, clicks, and how much of the planned content is published.

Does topical authority help with AI search? It plausibly helps, because AI features may search several subtopics to build an answer. It isn't proven, and Google says no special optimization is needed.

How is topical authority different from keyword clustering? Clustering is a method: grouping queries by intent. Topical authority is a goal: covering a subject well. Clustering is how you plan and measure the coverage.

How long does it take to build topical authority? No honest number exists. It depends on your starting point, the competition, and how fast you publish. Be wary of anyone who gives you a timeline.

Here's what the report shows, what it leaves out, and a method for tracking AI Overview visibility by topic cluster instead of page by page.

Can you see AI Overview data in Google Search Console?

Yes. Google announced Search Generative AI performance reports in Search Console on 3 June 2026, covering AI Overviews and AI Mode (Google Search Central). It started with a subset of sites. Google's help page now says that as of 31 August 2026 the report has rolled out to all websites worldwide (Search Console Help).

If you still don't see it, Google's help page gives the likely reason: your site hasn't had enough impressions in generative AI features. That can also happen if the site has been excluded from Search generative AI features in Search Console's settings.

AI feature impressions are also still counted in your regular Performance report under the Web search type (Google Search Central). The new report gives them their own view.

What does the Search Console Generative AI report show, and what does it leave out?

It shows impressions: how many times links to your site were shown in a generative AI feature. You can split them by page, country, device and date. Impressions are the only metric Google's help page lists for the report (Search Console Help).

What it leaves out is most of what you'd want to act on:

• No queries. You can see which page appeared, not what anyone asked.

• No clicks, CTR or position. You can't tell whether an appearance sent anyone to your site.

• The usual limits. Google notes the 1,000-row limit and time-period limits for the Performance report apply here too.

Does the Generative AI report show clicks or queries?

Neither. Google's help page lists impressions only. A third-party test also found the report isn't available through the Search Console API, so it can only be exported by hand (chudi.dev). Google's help page mentions an export button but doesn't mention API access either way, so treat that as one author's finding and check it against the current API reference.

That makes the report a good answer to "which of my pages show up in AI answers" and a poor answer to everything after it.

How do you find which queries trigger AI Overviews for your site?

You can't, directly. No Google report ties a query to an AI Overview appearance. What you can do is isolate the queries that look like prompts and track them separately.

AI-style searches tend to be long and conversational. One practical proxy is to filter Performance report queries to those of eight or more words, or those phrased as questions (chudi.dev uses this approach). It's a proxy. Plenty of short queries trigger AI Overviews and plenty of long ones don't.

In the Performance report, add a Query filter, choose Custom (regex), and paste:

^(\S+\s+){7,}\S+$

That matches queries of eight or more words. For question-style queries, use:

^(what|how|why|when|can|does|do|is|are|should|which)\s

Search Console often normalises punctuation, so don't rely on a trailing question mark. Starting words are more reliable.

How do you measure AI visibility for a whole topic, not just one page?

Roll the page-level numbers up to the cluster. The Generative AI report is page-based, and pages belong to topics, so you can aggregate in a spreadsheet.

1. Export the report's Pages table. Do it manually. Set the longest date range available.

2. Build your clusters. Group your Performance-report queries by topic, and note which page each cluster lands on. If you haven't done this, what keyword clustering is covers how.

3. Map each exported URL to its cluster. One cluster can have several URLs.

4. Sum AI impressions per cluster. Add organic impressions and clicks from the standard report next to it.

5. Add a prompt-shaped column. Count the cluster's queries that match the regex above.

The output looks like this. The numbers are invented to show the layout, not real data:

Read it by cluster, not by row. Deliverability earns AI impressions at a high rate relative to its organic impressions and has many prompt-shaped queries, which suggests the topic is the kind AI answers cover. Email tools comparison gets clicks but little AI exposure. That's a topic where AI answers may not be citing you. Page-level data can't show you either pattern, because it never groups the pages.

What to do with the numbers

Three things are worth acting on:

• High organic, low AI impressions: you rank, but AI answers aren't using you. Check whether the page answers the sub-questions around the topic. Query fan-out explains why that matters.

• High AI, low clicks: you're being shown without being visited. That's a content and CTR problem to look at, not a win.

• Clusters with no page in the AI report at all: these may be coverage gaps.

Google says there are no special requirements to appear in AI features (Google), so the fix is almost never an AI trick. It's the ordinary work of covering a topic properly.

Where TrackClusters fits

TrackClusters groups your real Search Console queries into clusters and tracks impressions, clicks and average position for each over time. That covers the organic side of the table above without a spreadsheet. The Generative AI report is a separate export that you'd add yourself. There's a 7-day free trial on the pricing page.

Related reading

• Keyword clustering for AEO: how to turn clusters into answer-ready pages
• How to build topical authority with keyword clusters: a step-by-step cluster plan

FAQ

Is there an AI Overviews report in Google Search Console? Yes. Google announced Generative AI performance reports on 3 June 2026, covering AI Overviews and AI Mode, and says they reached all websites by 31 August 2026.

Does the report show which queries triggered an AI Overview? No. It shows impressions by page, country, device and date. It doesn't list queries.

Can I get the Generative AI report through the Search Console API? One third-party test found no, and that the report can only be exported by hand. Google's help page doesn't mention API access, so check the current API reference.

Are AI Overview impressions included in my normal Performance report? Yes. Google says AI feature appearances are included in overall search traffic, under the Web search type.

How do I find AI-style queries in Search Console? Use a custom regex filter on queries: eight or more words, or starting with a question word. It's a proxy, not a label from Google.

Why track AI visibility by cluster instead of by page? AI answers draw on several sub-questions around a topic, so a topic is the unit that matters. Page numbers don't show whether a topic as a whole is covered.

How often should I export the report? Monthly is enough for most sites. The 1,000-row limit means busy sites may want to export in shorter date ranges so rows aren't cut off.

Here's what query fan-out is, how it differs between Google and ChatGPT, what we know (and don't) about the number of sub-queries, and what to do about it.

What is query fan-out?

Query fan-out is a retrieval technique where an AI search system breaks one question into several related sub-queries, runs them at the same time, and combines the results into a single answer. Google's documentation says it plainly: "Both AI Overviews and AI Mode may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources — to develop a response" (Google Search Central).

Take a search like "best CRM for a small agency." The system doesn't just look that phrase up. It might also look for pricing comparisons, integrations with common agency tools, reviews from agencies of that size, and how a few named products differ. Each of those is a separate search. The answer you read is stitched from all of them.

For you, the consequence is simple: a page can be cited for a question it never targeted, because it answered one of the sub-questions well. And a page that ranks first for the head keyword can be skipped if it only covers one slice of the topic.

How does query fan-out work in Google AI Mode vs ChatGPT?

Both expand one prompt into many searches. The difference is what they search. Google AI Mode runs its sub-queries against Google's own index and, per NoGood's breakdown, specialized data sources such as the Shopping Graph and Knowledge Graph. ChatGPT, per the same source, sends its sub-queries to Bing.

Google describes AI Mode as "breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." It describes Deep Search as the same technique "taken to the next level," capable of issuing "hundreds of searches" (Google).

Practical difference for you: if you want to be cited in Google's AI features, standard Google SEO fundamentals still apply, and Google says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." ChatGPT's retrieval layer is a separate system you can't inspect from Search Console.

How many sub-queries does AI Mode generate per prompt?

Google doesn't publish a number. It says "multiple" for AI Overviews and AI Mode, and "hundreds" for Deep Search. Third-party estimates exist (NoGood describes ChatGPT generating 8 to 15 sub-queries for a typical complex prompt, and 5 to 20 in general), but treat any specific count as an estimate, not a spec.

The better detail is that the count varies. Surfer's research, as cited by NoGood, found that only 27% of fan-out queries stayed the same across repeated runs of the same prompt. Don't build a strategy on the exact percentage. The direction is the useful part: the sub-queries move around between runs.

That has a direct consequence. If the individual sub-queries are unstable, optimizing for any one of them is chasing something that may not be there next time.

How do you see the fan-out queries for your own keywords?

You can simulate them, and in ChatGPT you can read them. You can't pull them from Google.

• ChatGPT: open Chrome DevTools, go to the Network tab, and search the requests for search_model_queries. The JSON lists the queries it sent (NoGood).

• Google AI Mode: there's no official view of the sub-queries. Qforia, a free tool from iPullRank, uses Gemini to simulate the fan-out for AI Overviews and AI Mode. It simulates. It doesn't show you what Google actually ran.

• Search Console: AI feature traffic is counted in your normal Performance report under the Web search type (Google). You won't see the sub-queries there either. What you do see is the real queries people typed, with impressions and positions. That's your ground truth for what the topic looks like from your side.

What a fan-out set looks like once you cluster it

This is the part most explainers skip. They define fan-out, then say "build topic clusters" and stop. Here's the mechanic.

The sub-queries below are a hand-written illustration, not a captured Qforia or Google output. They show the shape of the data. Run your own prompt through a simulator to get the real list.

Prompt: best CRM for a small agency

Ten sub-queries, five subtopics. Run the same prompt again and the exact phrasing changes. "cheapest crm for agencies" might become "affordable crm for agency owners." But it still lands in Pricing. The subtopics survive even when the wording doesn't.

That's the point of the whole article: a fan-out set is a keyword cluster. The sub-queries are unstable. The topics they group into are the stable unit. You can't reliably target the query. You can reliably cover the cluster.

If you've read what keyword clustering is, this is the same operation on a different input. And as the semantic keyword clustering post explains, grouping by meaning catches queries that share no vocabulary, which is exactly what sub-queries do: "crm free plan limits" and "cheapest crm for agencies" share nothing but intent.

How do you optimize content for query fan-out?

Cover the full set of subtopics around a question, and make each one easy to extract. There is no fan-out-specific markup or file. Google is explicit that you don't need new machine-readable files, AI text files, or special schema.

In practice:

1. List the subtopics first. Simulate the fan-out for your prompt, cluster the results, and treat each cluster as a section you need to answer.

2. Give each subtopic its own heading and a direct answer. A question as the H2, a 40-60 word answer under it, then the depth. This helps readers and makes the section easy to lift.

3. Check what you already rank for. Your Search Console data already shows which sub-topic queries you get impressions for. Gaps in that list are the subtopics you haven't answered.

4. Map one cluster to one page. Two pages covering the same cluster compete with each other.

5. Don't chase the individual sub-query. It may not appear next time.

None of this is new advice. It's topic coverage, which SEOs were already doing under another name. Fan-out just makes the cost of skipping it more visible.

Where TrackClusters fits (and where it doesn't)

TrackClusters doesn't see fan-out. Nothing that reads Search Console does, because Google doesn't expose the sub-queries. What it does is group the real queries your site already earns impressions for into topic clusters and track how each cluster moves, using your own GSC data. That tells you which of your subtopics are strong and which are thin. It's the "check what you already rank for" step from the list above, done without the spreadsheet.

If you want to try it, there's a 7-day free trial on the pricing page. More on the clustering side lives in the rest of the blog.

Related reading

• Keyword clustering for AEO: how to turn clusters into answer-ready pages
• How to build topical authority with keyword clusters: a step-by-step cluster plan
• How to track AI Overview visibility by topic cluster: measuring AI features in Search Console

FAQ

What is query fan-out in simple terms?

One search becomes many. An AI system splits your question into related sub-questions, searches for each, and merges the results into one answer.

Is query fan-out a Google ranking factor?

No. It's a retrieval technique AI Overviews and AI Mode may use to build a response, not a ranking signal you can optimize directly. Google says there are no special optimizations required for those features.

Do I need schema or an llms.txt file to benefit from query fan-out? Not according to Google. Its documentation says you don't need new machine-readable files, AI text files, markup, or special schema.org structured data to appear in AI Overviews or AI Mode.

Can I see the exact sub-queries Google runs for my keyword?

No. Google doesn't publish them. You can simulate them with a tool like Qforia, and in ChatGPT you can read them in DevTools, but for Google AI Mode any list is a simulation.

Why not just optimize for each sub-query?

They change between runs. Surfer's research found only 27% of fan-out queries stayed consistent across repeat runs of the same prompt. Topics are more stable than phrasings, which is why clusters are the better unit.

How is query fan-out different from long-tail keywords?

Long-tail keywords are queries people type. Fan-out queries are generated by the AI system on the person's behalf. One typed prompt can produce many of them.

Does AI Mode traffic show up in Search Console?

Yes. Google says sites appearing in AI features are included in overall search traffic in Search Console, under the Web search type. Search Console also has a separate Generative AI report that shows impressions in AI Overviews and AI Mode by page, but it has no queries or clicks, and it only appears for sites with enough AI impressions.

Here's what clustering looks like when it's built from a site's real query data instead of a keyword research tool's guesswork — three examples, the mechanics behind each, and where the "estimated volume" version usually breaks down.

Example 1: The content pillar cluster

Say a site has a guide on email marketing tools. Pull its GSC queries and you'll typically find a spread like this sitting under one topic:

• "best email marketing tools"

• "email marketing software comparison"

• "email marketing tools for small business"

• "email automation software"

• "email marketing platforms"

None of these are the same keyword. All of them are the same intent, hitting the same page, and moving together when that page gets updated or loses relevance. Track them individually and you get five noisy lines on five separate charts. Group them and you get one signal: is this pillar gaining or losing ground.

This is the simplest cluster shape and the one most manual spreadsheet clustering catches, because the keywords look similar on the surface.

Example 2: The question cluster

Question clusters are where manual clustering usually falls apart, because the keywords don't share obvious words:

• "how does email deliverability work"

• "why do emails go to spam"

• "what affects sender reputation"

• "email bounce rate causes"

Four different phrasings, one underlying topic — deliverability — and often one FAQ section or one page answering all of them. A keyword-matching approach (grouping by shared words) misses this cluster entirely, because "deliverability," "spam," "sender reputation," and "bounce rate" share almost no vocabulary. Grouping by which page actually earns the impressions catches it immediately, because GSC already tells you which query hit which URL.

Example 3: The commercial cluster

Commercial-intent queries cluster differently again — usually around a comparison or pricing page rather than a guide:

• "[tool] vs [competitor]"

• "[tool] pricing"

• "[tool] alternatives"

• "is [tool] worth it"

These queries often have lower individual impression counts than the informational cluster above, but they convert at a different rate, so mixing them into the same average as top-of-funnel queries hides the thing that actually matters for revenue: whether the comparison page itself is holding position.

Why individual keyword tracking misses all three

Once a site has more than about 50 ranking queries, tracking them one at a time stops being analysis and starts being scrolling. The signal was never in keyword #37 moving from position 8 to position 11. It's in whether the whole deliverability cluster is trending down while the pillar cluster holds steady — a pattern invisible until the keywords are grouped by topic and tracked as a group over time, not eyeballed individually once a month.

The example nobody's spreadsheet includes: position 0

Here's a detail that skews every one of the examples above if you're building clusters manually in a spreadsheet: GSC exports report position 0 for any query that ranked outside the top 100 during that period. If you average position across a cluster and include those zeros, the cluster looks like it's cratering when it's actually fine — a handful of long-tail queries that briefly showed up and dropped back out are dragging the average toward zero.

This is a specific, checkable data quality issue, not a modeling assumption. TrackClusters excludes position-0 queries from cluster average calculations for exactly this reason. It's a small detail, and it's the difference between a cluster average that reflects reality and one that panics you over noise.

Building these clusters from your own data

You don't need a keyword database to build any of the three examples above — you need your own GSC Performance report exported by query and page, then grouped by which page each query actually lands on. That's the whole mechanic. The manual version works fine at 50 keywords and turns into a part-time job past 200.

If you want that grouping done automatically from your own GSC data, without volume estimates or scraped rankings, that's the entire premise of TrackClusters — connect GSC, get clusters, watch them move. Pricing and the free trial are on the pricing page; more on the mechanics behind clustering is in the rest of the blog, including what keyword clustering actually is if you want the concept explained from scratch first.

Related reading

• Semantic keyword clustering: how grouping by meaning works
• Keyword clustering for AEO: how to turn clusters into answer-ready pages

FAQ

What's the difference between keyword clustering and keyword grouping? In practice, not much — "grouping" is often used for manual, rule-based bucketing (shared words, shared modifiers) while "clustering" implies grouping by intent or by which page the queries actually land on. The GSC-based examples above use the second approach.

Do I need search volume data to build keyword clusters? No. All three examples above were built from impressions, clicks, and position — data GSC already gives you for free. Volume estimates help with new-keyword research; they add nothing to clustering keywords you already rank for.

How many keywords need to be in a cluster before it's useful? There's no fixed minimum, but clusters under 3-4 queries usually don't tell you more than tracking those queries individually would. The value shows up once a cluster has enough queries that an average is more informative than any single line.

How often should clusters be recalculated? As often as your GSC data updates meaningfully — for most sites, weekly is enough to catch a trend without reacting to daily noise. GSC data itself lags by a few days, so recalculating hourly doesn't buy you anything.

Can I build these clusters manually in a spreadsheet? Yes, and Example 1 is genuinely easy to do by hand. Example 2 gets tedious past a dozen or so question variants, and remembering to strip position-0 rows before averaging is the step people skip.

Why does GSC show position 0 for some of my queries? Position 0 in a GSC export means the query didn't rank in the top 100 for that period — it's not a real ranking position, it's GSC's placeholder for "not found." Averaging it in with real positions understates how a cluster is actually performing.

Does clustering work for a small site with fewer than 50 ranking keywords? It still works, but the payoff is smaller. Individual tracking is manageable at that scale. Clustering starts earning its keep once scrolling through individual keywords stops being a five-minute task.

Here's what it actually means, how it works mechanically, and where it breaks down.

How Semantic Clustering Works

Semantic clustering runs on embeddings — numerical representations of a keyword's meaning, generated by an NLP model. Two keywords that mean similar things end up with embeddings that sit close together in vector space. The clustering step is just math from there: measure the distance (usually cosine similarity) between every pair of keyword embeddings, and group keywords whose distance falls under some threshold.

This is why semantic clustering catches things exact-match or string-matching methods miss. "Cluster keywords" and "group search terms" share zero words but mean close to the same thing, and a decent embedding model will place them near each other. That's the entire value proposition in one sentence: meaning over string similarity.

Semantic Clustering vs. SERP Overlap Clustering

The other common clustering method works differently: instead of comparing what keywords mean, it compares what Google does with them — specifically, how much the top-10 results overlap for each query. If two keywords return mostly the same URLs in position 1-10, Google's own ranking system has effectively already decided they're the same topic, and the cluster follows that signal.

Neither method is strictly better. Semantic clustering is faster to compute and doesn't require live SERP data, which makes it the default in most tools. SERP overlap clustering is slower and API-hungry but occasionally catches something semantic similarity misses — cases where Google treats two oddly-worded queries as the same intent for reasons that aren't obvious from the text alone.

Where Semantic Clustering Gets It Wrong

Embeddings measure similarity of meaning, not similarity of intent, and those aren't the same thing. "Apple pricing" and "orange pricing" are semantically close — both are fruit-pricing queries — but if one searcher means fruit and the other means the phone company, no embedding model catches that without more context than a bare keyword provides. Ambiguous, short-tail terms are where semantic clustering produces confident-looking clusters that are quietly wrong.

This is worth knowing before you trust a tool's clustering output blindly. A clustering algorithm has no idea it made a mistake — it just returns groupings with the same confidence whether the underlying similarity score was 0.95 or 0.61. Spot-checking clusters, especially ones built on short or ambiguous queries, catches this before it skews your reporting.

Semantic Clustering on Research Data vs. Performance Data

Most explanations of semantic clustering assume you're starting from a keyword research export — a list of terms with estimated search volume attached, usually pulled from a tool like Ahrefs or SEMrush. That's a fine starting point for planning new content. It's a weaker one for measuring what's already working, because volume estimates are modeled numbers, not observed ones. A keyword showing "1,000/mo" might get 200 searches or 2,000 — there's real variance hiding behind that single figure.

The alternative is clustering your own Search Console performance data — the actual queries your site is already getting impressions, clicks, and position data for. Nothing here is estimated. It's the record of what real searches actually returned your pages. Semantic clustering works exactly the same way on this data — group by meaning, track the group — but the numbers you're clustering are observed rather than modeled.

A Data Quality Detail Semantic Clustering Doesn't Fix on Its Own

Clustering method has nothing to do with a separate problem that shows up specifically in GSC exports: position 0. Any query ranking outside the top 100 shows up in Search Console with a position value of 0, not blank. Semantic clustering will happily group those position-0 queries into a cluster along with everything else, and if you average position across the cluster without excluding them, the average collapses toward zero for reasons that have nothing to do with real performance decline.

It's a small, specific bug in how most people build clustering spreadsheets, and it's easy to miss because the math still runs without an error — it just returns a number that's wrong. TrackClusters excludes position-0 rows from cluster average calculations by default, which is a narrower and more mechanical fix than "semantic clustering" but matters more for whether your reported numbers reflect reality.

What This Looks Like in Practice

TrackClusters connects to Google Search Console over OAuth — read-only, so it can query your data but can't touch the property itself — and clusters your existing ranking queries automatically, then tracks impressions, average position, and clicks per cluster over time. No search volume estimates, no scraped SERP positions: the data is pulled directly from your own GSC property. Plans start with a 7-day free trial, no credit card required, if you want to see your own queries clustered instead of a demo dataset.

How to Try Semantic Clustering on Your Own Data

1. Export 3-6 months of query data from GSC's Performance report, or start from a research tool export if you're planning new content rather than measuring existing rankings.
2. Group by meaning, not exact string. "Semantic keyword clustering" and "clustering keywords by meaning" belong together even though the strings don't match.
3. Spot-check ambiguous clusters. Short or multi-meaning terms are where semantic similarity most often produces a confident but wrong grouping.
4. Exclude position-0 rows from any average position math, independent of which clustering method you used to build the groups.

That's the practical version: cluster by meaning, verify the ambiguous cases, and keep the position-0 problem separate from the clustering method entirely — they're different problems that happen to show up in the same spreadsheet. More breakdowns like this live on the TrackClusters blog, including a longer walkthrough of what keyword clustering is if you're starting from scratch. For the full process with a worked example, see how to group keywords.

Related reading

• Keyword clustering examples: what real clusters look like in GSC data
• Keyword clustering for AEO: how to turn clusters into answer-ready pages

FAQ

What is semantic keyword clustering? Semantic keyword clustering groups keywords by meaning rather than exact wording, typically using embeddings or NLP similarity scoring to measure how close two keywords are in intent.

How is semantic clustering different from SERP overlap clustering? Semantic clustering compares the meaning of keywords directly. SERP overlap clustering compares how much the actual Google top-10 results overlap between two queries. Both group by topic; they use different signals to get there.

Does semantic clustering need search volume data? No. Semantic clustering only needs the keyword text itself to generate embeddings and compare similarity. It works the same whether the keyword list comes from a research tool with volume attached or from your own GSC performance data with none.

Why does semantic clustering sometimes group unrelated keywords together? Embeddings measure similarity of meaning, not intent. Short or ambiguous terms — like a word with two unrelated meanings — can end up in the wrong cluster because the model has no extra context to disambiguate them.

What is position 0 and does semantic clustering fix it? Position 0 is what Google Search Console shows for a query ranking outside the top 100. Semantic clustering doesn't address this at all — it groups keywords by meaning regardless of position. Excluding position-0 rows from average position math is a separate step a clustering tool needs to handle on its own.

Can I do semantic keyword clustering without code? Yes, if you're using a tool that runs the embedding and similarity comparison for you. Doing it manually without code means falling back to eyeballing a spreadsheet and grouping by judgment, which isn't really semantic clustering — it's manual topical grouping.

Is semantic clustering more accurate than manual grouping? It's more consistent and faster at scale, but not infallible — see the ambiguous-term problem above. A hybrid approach, semantic clustering with a spot-check on questionable groupings, tends to outperform either method used blindly.

It's a strange gap for an industry this obsessed with measurement. SEO teams track average position to the decimal point, then still report performance keyword by keyword — 500 rows in a spreadsheet, no pattern visible anywhere. Group those same 500 keywords into 20 topic clusters and you get 20 signals you can actually act on.

How Keyword Clustering Works

Clustering takes a list of keywords and groups the ones that are close enough in meaning, intent, or topic to be treated as one unit. Instead of asking "how does keyword clustering tools rank this week," you ask "how is the keyword clustering tools content pillar trending" — the individual keyword becomes one data point inside a bigger signal.

There are two common starting points for the keyword list itself:

• Keyword research data — exported from a tool like Ahrefs or SEMrush, usually with estimated search volume attached.
• Search performance data — pulled from Google Search Console, showing the queries your site is already getting impressions and clicks for.

Both are valid starting points. They answer different questions. Research data tells you what to target. Performance data tells you what's already working and where it's sliding.

Manual vs. Automatic Clustering

You can group keywords manually — export a list, eyeball it, group by hand in a spreadsheet. It works, and plenty of SEO teams still do it. It also takes hours per pass and goes stale the moment new query data comes in. Automatic clustering tools do the same grouping using semantic similarity or SERP overlap, and update as new data arrives.

Why Cluster Instead of Tracking Keywords Individually

Once a site has more than about 50 ranking keywords, tracking each one separately produces noise, not insight. You can watch a single keyword drop three positions and have no idea whether that's a real problem or normal variance — until you see the other 14 keywords in its cluster move the same direction that same week. That's a content pillar shifting, not a fluke.

This is also where Google Search Console gets underused. It hands you real impression, click, position, and CTR data for every query your site has shown up for, for free. Most people check it once a month, glance at the top 10 rows, and close the tab. There's a lot more signal sitting in there than the top 10 rows show.

Keyword Clustering Methods

A few approaches show up repeatedly when people talk about clustering methods:

• Semantic clustering — grouping keywords that mean roughly the same thing ("keyword clustering tool" and "tool for clustering keywords") using embeddings or NLP similarity scoring.
• SERP overlap clustering — grouping keywords whose top-10 Google results overlap significantly, on the logic that if Google thinks two queries deserve the same pages, they belong together.
• Manual topical grouping — a human decides which keywords belong to which content pillar, based on domain knowledge rather than an algorithm.

None of these methods require search volume to work. That's worth saying plainly, because most clustering guides assume you're starting from a keyword research export. If you're clustering your own ranking queries from GSC, you're clustering based on real impressions and clicks — data with less guesswork than a modeled volume number.

A Data Quality Pitfall: Position 0

Here's a detail most clustering approaches miss entirely. When you export query data from Google Search Console, keywords that rank outside the top 100 show up with a position of 0 — not blank, not null, literally 0. If you average position across a cluster without excluding those rows, your "average position 8" cluster can quietly become "average position 34" the moment a handful of position-0 keywords get pulled into the math. The cluster looks like it's collapsing when it isn't.

It's a small, specific, entirely fixable data quality problem, and it's the kind of thing you only notice after you've built a clustering workflow and watched the numbers do something that doesn't match reality. Excluding position-0 rows from average calculations is one of the more concrete things a good clustering tool should do by default.

Keyword Clustering Tools: What to Look For

Clustering tools generally fall into two camps. Volume-based tools cluster a keyword research export and estimate opportunity — useful for planning new content, less useful for measuring what you already rank for. GSC-native tools cluster your actual search performance data — real impressions, real clicks, real position, pulled directly from your own property rather than modeled or scraped.

TrackClusters is built for the second case: it connects to Search Console over OAuth, groups your existing ranking queries into clusters automatically, and tracks how those clusters move over time — impressions, average position, and clicks, with position-0 keywords excluded from the averages. The OAuth connection is read-only — TrackClusters can query your Search Console data but can't change anything in the property itself, in line with its privacy practices. Plans start with a 7-day free trial, no credit card required, if you want to see your own query data clustered rather than a demo dataset.

How to Start Clustering Your Own Keywords

If you're doing this for the first time:

1. Export your query data. Pull the last 3-6 months from GSC's Performance report, or export a keyword list from your research tool of choice.
2. Group by topic, not by exact match. "Keyword clustering methods" and "how to cluster keywords" belong in the same group even though they're not the same string.
3. Exclude position-0 rows from any average position math, for the reason above.
4. Review clusters every 2-4 weeks, not daily. Cluster-level movement is a slower signal than individual keyword rank, and checking too often just reintroduces the noise you were trying to get away from.

If you've ever built a manual GSC clustering spreadsheet, you already know step 4 is where it usually falls apart — the spreadsheet is accurate the day you build it and wrong three weeks later. That's the specific gap automatic clustering tools close: the grouping updates as new query data comes in instead of freezing the moment you finish the pivot table.

That's the shape of it: group first, track the group, watch for position-0 noise. More breakdowns like this live on the TrackClusters blog — and if you want to see your own queries clustered instead of a demo dataset, the free trial runs 7 days with no card required.

Related reading

• Semantic keyword clustering: how grouping by meaning works
• Keyword clustering examples: what real clusters look like in GSC data

FAQ

What is keyword clustering?

Keyword clustering is the process of grouping related search queries into topic-based clusters, so performance can be measured at the topic level instead of one keyword at a time. For a step-by-step process for grouping your own queries, read our guide to keyword grouping.

How is keyword clustering different from keyword grouping?

They're generally used interchangeably. "Clustering" more often implies an automated or semantic method; "grouping" is sometimes used for manual, human-judgment groupings. Neither term has a strict industry-wide definition.

Do I need search volume data to cluster keywords?

No. Clustering can run on search volume data, but it can also run on real performance data — impressions, clicks, and position — pulled directly from Google Search Console. Volume isn't required.

How does GSC data help with keyword clustering?

Search Console shows the queries your site is already getting impressions and clicks for, with real position and CTR data. Clustering that data shows which topic areas are actually working, not which keywords a research tool estimates you could target.

What is position 0 in Google Search Console and why does it matter for clusters?

Position 0 is what GSC shows for a query ranking outside the top 100. Included in a cluster's average position calculation, it drags the average down artificially and makes a cluster look worse than it performs.

Can I cluster keywords manually in a spreadsheet?

Yes. It works for a one-time pass but takes real time and goes stale as soon as new query data comes in, since nothing updates it automatically.

How often should keyword clusters be updated?

Every 2-4 weeks is a reasonable cadence for most sites. Cluster-level movement is a slower, steadier signal than individual keyword rank.

Is keyword clustering the same as topic clusters in content strategy? Related but not identical. Content topic clusters describe how you structure content (a pillar page plus supporting posts). Keyword clustering describes how you group search queries for tracking and analysis. The two often align, but one is a content structure and the other is a measurement method.