8 grouping methods: shared words, rare-word weighting, semantics, structure, topic density and spelling variants
💡 Any format supported: with or without frequencies, any separator (tab, semicolon, space, etc.)
Compare all and select best
When unsure which to choose
Less = more keywords per cluster
More = fewer final clusters
Fundamental SEO technique for creating effective site structure
Keyword clustering is the process of automatically grouping semantically similar search queries into logical groups (clusters). Each cluster represents one topic corresponding to one page on your site. This allows you to create content that maximally matches user search intent and Google algorithm requirements.
We combine several independent grouping approaches: rare-word weighting, shared-word similarity, word-fragment comparison, spelling-variant tolerance, semantic context analysis and hierarchical topic building. The system automatically compares the results of all methods and recommends the best one for your data.
8 advanced NLP techniques for the most accurate grouping
Terms are weighted by how rare they are: unique modifiers influence grouping more than generic words.
Queries are compared not only by whole words but by short fragments, so different forms of a word stay together.
Typos, alternative brand spellings and transliteration do not break a cluster apart.
Agglomerative algorithm with average linkage for optimal cluster merging.
Word co-occurrence matrix for detecting semantic relationships between terms.
Every cluster keeps its own meaning centre, and new queries join the closest one by meaning.
50+ stemming rules for Ukrainian and English with morphology support.
Automatic removal of 100+ function words for both languages for cleaner analysis.
Four simple steps to perfect site structure
Copy your keyword list from Excel, Google Sheets or any other source. Each keyword should be on a new line.
Basic method is fast and effective. Advanced creates more precise subgroups using spelling-variant matching.
The system will automatically group keywords by semantic similarity and show relevance level for each word.
Download ready clusters in CSV or JSON format for further work in Excel, Google Sheets or your CMS.
Choose method based on data size and needed precision
Fast grouping with combined similarity metric
Cluster meaning centres with rare-word weighting
Word co-occurrence analysis in context
Bottom-up agglomerative clustering
Free clustering is only the entry point. Four premium areas cover the rest of the semantic workflow: AI-driven grouping, page quality analysis, competitor text analysis of the Google Top 10, and clustering based on the real search results.
A language model reads search intent and niche context where pure math falls short.
20 page quality factors, a score, and an automatic content brief in Markdown.
Full text analysis of every page ranking in the Top 10 and the terms you are missing.
Grouping by real SERP overlap plus Volume, CPC and KD from Ahrefs.
Algorithmic methods group queries by shared words and statistics. That is fast, transparent and works well on large, uniform keyword sets — but it does not understand meaning. "Buy running shoes" and "marathon footwear" look different mathematically, yet Google treats them as one page.
The AI mode sends your keyword set to a language model that evaluates search intent, query type (informational, commercial, navigational, transactional) and topical proximity. The resulting cluster reflects what a person is actually looking for, not how many characters match.
Two specialised profiles are available on top of that: Google SEO for classic informational and commercial projects, and iGaming for a market where brands, providers, bonus mechanics and geo modifiers create semantics that generic models handle poorly.
AI modes are not a magic button. They are a controlled pipeline where the language model handles exactly what algorithms cannot: meaning, intent and niche context.
Your keyword set is de-duplicated, cleaned and normalised while search volumes are preserved. The model receives compact batches instead of a raw file — lower cost, more stable output.
The model judges what a query is actually about: "buy washing machine" and "order washer cheap" belong to one page even with almost no shared words. This solves the core weakness of classic methods — synonyms, slang and different phrasings.
Every query gets an intent type: informational, commercial, transactional, navigational, comparative. Clusters never mix "what is X" with "buy X", so your site structure is correct from the first run.
AI Google SEO follows search-result logic and page types. AI iGaming understands brands, providers, bonuses, geo and local niche specifics that generic algorithms ignore.
For each cluster the model produces a human page title, a primary keyword and secondary ones. You get a ready heading for the brief, not "Cluster #17".
A control pass splits over-broad groups, merges near-identical small ones and isolates outliers. Every keyword ends up in exactly one cluster.
Queries with zero shared words land together when they target the same page.
Informational and commercial queries stay apart — less page cannibalisation.
Each cluster arrives with a title, a primary keyword and brief-ready logic.
A second model pass removes groups that are too broad or too fragmented.
You can see why a keyword landed where it did, and export the full distribution.
Keywords are processed within the session and are not used to train third-party models.
AI modes are available on paid plans due to compute cost. For very large sets, run an algorithmic method first and apply AI to the priority segments.
A high-level view of the process: what happens to your keyword set at every stage and what you get at the end.
The list is cleaned of duplicates, stray spaces and stop words; the language and the volume column are detected. Different forms of the same word are reduced to a common base so that "buy laptop" and "laptops buy" never end up in separate groups.
Every pair of queries gets a closeness score from several independent signals at once — shared words, rare modifiers, spelling and usage context. No single signal decides alone: the final score is their balance.
Queries gather around meaning centres. A cluster that nothing else fits into is closed; one that grows too broad is split into subtopics. Isolated queries are not forced into a group — they are kept aside.
The system cross-checks the groups: merged topics are separated, split ones are reunited, and each cluster gets a head query and total demand. Several methods can be run in parallel so you can compare outcomes.
The output is a cluster table with the head query, demand volume, keyword count and each query's relevance to its group — an immediately exportable blueprint for your site pages.
A fragment of a real 42-query set after clustering
Main commercial page
online casino uk
12 queries · 62,800 searches/mo
Generic commercial phrasings with the same intent — one landing page.
"Best of" roundup
best online casino uk
9 queries · 14,200 searches/mo
Comparison intent — a separate ranking or roundup page.
New casinos
new online casino uk
6 queries · 6,570 searches/mo
Freshness intent — a page you update regularly.
Payouts and bonuses
best payout online casino uk
8 queries · 4,340 searches/mo
Qualifying criteria — dedicated sub-pages.
We deliberately do not publish specific thresholds, signal weights or internal formulas: tuning them is exactly what separates a usable site structure from mechanical word grouping.
The tool is built so that a confidential keyword set can be processed without risk to your client or project.
All algorithmic methods run locally on your device. Keyword lists are never sent to a server and are never stored by us.
Premium-mode requests are processed once on the server and are never used to train models or shared with third parties.
Access to external services lives only in encrypted server-side storage. Nothing is present in page code or the browser — no key can be extracted from the frontend.
Saved projects belong to your account, are visible only to you and are deleted on request at any time.
Every algorithm has its own strength. Below is what type of keyword set each method suits, the volume it handles best, and the result you get.
One run, several algorithms on the same set. Results appear side by side with cluster counts and distribution, so the choice is evidence-based rather than a guess.
The system profiles your set — size, average query length, repetition, presence of volume data — and suggests the method that performs best on that specific data.
A universal parser recognises volume columns in any format and sorts clusters by total demand instead of alphabetically.
The analyzer breaks any URL down into 20 quality factors: heading structure, text volume and density, topical coverage, readability, phrasing originality, presence of facts and numbers, internal links, metadata, semantic completeness and signs of generated filler content.
The output is not an abstract number but a report written in a senior SEO analyst voice: what exactly is weak, why it affects rankings and what to do next. The same analysis produces a copywriter brief in Markdown — H2/H3 structure, required topics, terms and recommended length.
Agencies get a bulk mode: upload an XLSX or CSV list of pages, get a processing queue and one report where every site sits on its own tab, with the same analysis and brief you see inside the tool.
The Force takes your query, pulls the real Google results for the region and language you need, downloads the text of every page in the Top 10 and breaks it down into terms. You see what ranking content is actually made of instead of guessing.
Each term gets a Term Relevance Score based on how many competitors use it and how often. That produces a clear list: what to add to your text, what is underused, and what is over-optimised.
SERP results are cached, so repeating the same topic returns instantly and does not consume your limits. The SERP source, depth and geo are configurable per project.
The most accurate way to know whether two queries need separate pages is to look at the SERP. SERP clustering pulls the top results for every keyword and merges queries whose results overlap. That is the same logic Google itself applies, not a word-based approximation.
To make it work on tens of thousands of keywords rather than hundreds, the engine uses our own comparison optimisation: it never brute-forces every pair, so large sets do not stall processing. Cluster formation streams in real time, and partially collected data can be clustered without waiting for the whole queue.
Market metrics from Ahrefs — search volume, CPC and keyword difficulty — sit next to the clusters, with aggressive caching so you never pay twice for the same lookup. Native exports from the popular SEO services are detected automatically, with no manual file conversion.
The tool should fit your process instead of forcing manual data transfers. Any result exports to CSV, JSON, TSV, Markdown or XLSX — including a grouped view where each cluster is its own block, and multi-tab reports for client deliverables.
Project history depends on your plan: 7 days on Free, 90 days on Pro and unlimited on Agency. That lets you revisit old keyword sets, compare iterations and hand work over inside a team.
Basic clustering runs in your browser — the keyword set never leaves your device. Heavy computation moves to a Web Worker, so the interface stays responsive even on tens of thousands of queries. The interface is translated into 30+ languages, including right-to-left scripts.
Why thousands of SEO specialists use our tool daily
Properly structured content ranks higher in Google. Clustering helps create pages with maximum relevance.
Automatic processing of thousands of keywords in seconds instead of hours of manual work. More time for strategy, less for routine.
Each page focuses on one topic with all related keywords. Google better understands your content.
Clusters naturally form page hierarchy: categories, subcategories, articles. Improves UX and indexation.
One page = one cluster = one search intent. No conflicting pages fighting for the same keywords.
Process up to 50,000 keywords per request. Perfect for large projects and agencies.
Tool for professionals and businesses of any scale
Building semantic core, planning site structure, optimizing content strategy
Example: Clustering 10,000 keywords for an electronics online store
Fast processing of large keyword volumes for client projects
Example: Client site audit and restructuring in 1 day
Planning articles with full topic coverage and all LSI keywords
Example: Creating content plan for 100 blog articles
Optimizing product categories and creating SEO descriptions
Example: Structuring catalog of 5,000 products
Finding article topics and avoiding content duplication
Example: Niche analysis and publication planning for the year
Understanding what customers search for and optimizing site for their needs
Example: Competitor analysis and finding untapped niches
Keyword clustering is free and needs no signup. Paid plans lift the limits and unlock AI semantic methods, Content Analyzer, The Force, SERP clustering and bulk reports.
To get started
$0/mo
Core clustering, no signup required
For SEO specialists
$29/mo
The full toolkit for daily SEO work
For agencies and teams
$89/mo
High volumes and bulk reporting for client projects
For large companies
Custom
Custom limits, SLA and integrations
Daily run limits reset every 24 hours; monthly SERP and Ahrefs volumes reset each month. Change or cancel your plan at any time.
Based on reviews from 1,247 users
🔒 Your data never leaves your browser. All processing happens locally.
Everything you need to know about keyword clustering for SEO
Keyword clustering is the process of automatically grouping semantically similar search queries. One cluster = one relevant page on your site. This is the foundation for building site structure and content strategy.
Proper clustering can increase organic traffic by 30-50% through better content relevance.
Clustering helps: 1) Avoid keyword cannibalization when multiple pages compete for the same queries. 2) Create logical site structure. 3) Optimize internal linking. 4) Increase relevance of each page.
Optimally 5-20 keywords per cluster. Less than 5 — consider merging with another cluster. More than 20 — consider splitting into subclusters or creating a series of pages.
If a cluster contains more than 50 queries — this signals creating a hub page with subpages.
Auto — if unsure, system compares all methods. By Intent — for e-commerce and landing pages. By Rare Words — for niche sites. By Structure — for information portals. By Tail — for local SEO.
Similarity (0-100%) shows how well a keyword matches the cluster topic. >80% — perfect match, content foundation. 60-80% — good, include it. 40-60% — questionable, check manually. <40% — probably belongs to another cluster.
Filter results by similarity to exclude irrelevant queries.
Silhouette Score (-1 to 1) — clustering quality metric. >0.5 — excellent, clear groups. 0.25-0.5 — good, some overlap. 0-0.25 — satisfactory, consider another method. <0 — poor, clusters are mixed.
Available formats: CSV — for Excel, Google Sheets, Power BI. Contains cluster, keyword, similarity. JSON — for developers and CMS integrations. Structured format with metadata.
1) Export to CSV. 2) Each cluster = one article/page. 3) Cluster name — main keyword for Title. 4) Other cluster words — LSI for text. 5) Similarity determines inclusion priority.
Sort clusters by size — larger clusters usually have higher search potential.
No, absolutely not. All processing happens locally in your browser using JavaScript. Your data is not sent to server, not stored, and not analyzed by third parties.
Yes. Since processing is completely local, your business data including competitive analysis and semantic core remain confidential. No data leaves your browser.
Up to 50,000 keywords per request. Speed depends on your device. Approximately: 1,000 words — 1-2 sec, 10,000 — 10-20 sec, 50,000 — 1-3 min.
For large lists we recommend using By Intent method — it's the fastest.
Possible reasons: 1) Too many keywords (>10,000). 2) Using resource-intensive method (Semantic, Hierarchical). 3) Weak device or many open tabs. Try closing other programs.
Any format is supported: TAB-separated (keyword TAB 1000), semicolon (keyword;1000), in brackets (keyword (1000)), or space-separated (keyword 5000). You can also drag-and-drop a CSV/TXT file. The system automatically detects the format.
Numbers below 100 (e.g., 'iphone 15', 'top 10') are NOT parsed as frequency — they remain part of the keyword.
The system automatically sums up the search volume of all keywords in a cluster. This shows the total search potential of each group and helps prioritize content for the highest-volume clusters.
Sort clusters by total frequency to find the most valuable SEO topics.
Any exports from popular SEO tools are supported: Serpstat, Ahrefs, SEMrush, Google Keyword Planner, as well as CSV, TSV, TXT files. The system automatically finds the required columns (keyword, volume, etc.) regardless of their position in the file. Just drag-and-drop the file — everything else is handled automatically.
No need to edit files — the system smartly detects columns even if the file has lots of extra data.
Simply export keywords from any of these services in CSV or Excel format, then drag-and-drop the file into the input field. The system will automatically find columns with keywords and search volume, regardless of file structure and column names.
All report types are supported: positions, competitors, keywords, domain analysis, etc.
Algorithmic methods compare words and statistics, so synonyms and different phrasings of the same intent can land in different clusters. The AI mode evaluates search intent and topical proximity with a language model and adds specialised profiles for Google SEO and iGaming. It is available on Pro (20 runs per day) and Agency (100 runs per day), with up to 500 keywords per run.
Content Analyzer scores a single page of yours across 20 quality factors and generates a content brief. The Force works with the SERP: it pulls the Google Top 10 for your query, analyses the text of every competitor and calculates a Term Relevance Score for each term, showing what your page is missing. In practice they are used together — The Force supplies the terms, Content Analyzer validates the finished text.
Yes. Pro and Agency include a bulk mode: upload an XLSX or CSV list of URLs, the system queues them and produces one XLSX report where each site has its own tab with the analysis and the content brief. Exports to CSV, JSON, TSV and Markdown are also available.
Free gives you up to 500 keywords per request, 5 clustering runs per day, all 8 algorithmic methods, exports and 7 days of history — no signup required. Pro raises the limit to 15,000 keywords and 100 runs per day and unlocks AI methods, Content Analyzer, The Force, SERP clustering, Ahrefs data and bulk reports. Agency is built for teams: 50,000 keywords, 500 runs per day and unlimited project history.
Our algorithm is the grouping logic: how queries are normalised, how their closeness is measured, how clusters are formed and refined, and how the system picks a method for your keyword set. External services and data are sources of facts: search results for SERP methods, volume and difficulty metrics, and language models for semantic modes and brief generation. In short: facts come from outside, but the decision about what belongs together is made by our own logic.
That is why clustering stays reproducible even when external data changes: the inputs change, the grouping principle does not.
Algorithmic clustering runs directly in your browser — keyword lists never have to leave your device. For premium modes (AI, SERP, content analysis) the request is processed on our server and is never used to train models. All credentials for external services are stored server-side in encrypted storage, never appear in page code and are never visible in the browser. You can delete your project history at any time.
If your keyword set is confidential, use the algorithmic methods: the data stays on your device.