Keyword Cluster Generator

Choose keywords that concern the products, services or questions you can cover well. Keep one language and market together. A neatly organized group around an irrelevant subject is still a poor reason to publish a page.

Put one keyword on each line. Remove accidental clutter before grouping, but keep meaningful modifiers. Words such as free, without, a location or an audience can change what a reader needs. Similar wording does not guarantee the same answer.

Use the SEO tools hub to select a different workflow when your question concerns demand, existing page performance or a technical restriction. This tool compares supplied words or result addresses; it does not create the missing evidence for those other questions.

Keep a copy of the original research. If you change ignored tokens, thresholds or imported results later, you will need that source material to understand why the groups changed.

Word mode is useful when sorting an early research list. Imported-results mode asks a different question: do the same pages appear in the supplied results for these queries? Neither method proves that one page should target every keyword in a resulting group.

If the expected answer is unclear, use the Search Intent Checker to review a supplied search snapshot. Compare page types and what the visitor can accomplish. A guide, category page, comparison and interactive tool can use related words while serving different needs.

Word mode compares distinct tokens, ignoring case and word order. Repeating a word does not increase its importance. The score is the number of shared tokens divided by the number of distinct tokens found across both phrases.

For example, “content planning” and “content planning workflow” share two words from a combined set of three. Their overlap is two-thirds, or about 66.67%. At a 50% threshold, they qualify as a direct match.

The method does not use stemming, a synonym model or an automatic stop-word list. Related meanings can therefore remain separate when their words differ. Conversely, shared words can conceal an important difference in purpose. Read the phrases before treating a mathematical match as an editorial instruction.

Use ignored tokens deliberately. Removing a common brand or other unhelpful token may make a list easier to inspect, but removing negations or audience modifiers can erase the distinction you need to preserve. Do not make groups larger simply by discarding the words that explain the difference.

Imported-results mode needs a complete set of organic page addresses for each query at your selected depth. Choose top five or top ten, then set the minimum number of shared URLs needed for a direct match.

Use CSV or TSV with exactly these three headers:

keyword,url,position

Put one organic result on each row. Every position within the selected depth must be present, and the addresses must be distinct for that keyword. Two rows showing positions one and two explain the format, but they are not enough for a top-five comparison.

Keep advertisements, map results and other search features out of the organic list. Declare one search engine, country or market, language, device and capture date for the batch. Add a source note so another reviewer can understand how the snapshot was collected.

Do not mix locations, dates or devices under a single context label. A keyword can return different pages in a different setting. Matching text labels in the form also cannot establish that an export was collected with those settings; check the source before importing.

The minimum-shared-URLs setting counts matching page addresses. It does not count matching domains. Two different pages on the same website remain different results.

Suppose two complete top-five lists share three page URLs. At a minimum of three, that pair qualifies. This is a statement about the supplied lists, not a 30% ranking probability, a traffic forecast or proof of a common search intent.

Matching URLs receive equal weight within the selected depth. A page does not contribute more to the match merely because it appears first in one list and fifth in another. Position identifies which observations belong in the selected depth; this method does not add a rank-weighting score.

Indirect connections can occur here too. Query A may share enough URLs with B, and B with C, while A and C share too few. Review the direct-pair evidence and the group’s weakest relationship before assigning all three to one destination.

A keyword without a complete result set remains visible as not compared. That differs from a complete keyword with no qualifying match. Missing research does not establish that a term deserves its own page or has a unique reader need.

Repeated rows at the same position collapse when their matching URL is identical. Conflicting URLs at one position, or the same URL at different positions, need correction. Resolve the source evidence rather than selecting an arbitrary winner to fill the list.

The importer rejects unexpected columns. Keep batch context in the dedicated fields instead of adding a country or date column that might otherwise be ignored. Review your export preparation before removing data you may need for a separate analysis.

Importing replaces the result rows and adds their keywords to the current list. Existing keywords remain, so the project can contain terms with no imported results. Check that the intended list and snapshot are still active before generating a new report.

Imported-result matching ignores hostname case and URL fragments. It preserves protocol, path case, query strings and trailing slashes. Those rules concern address matching, not whether the destination content is equivalent.

For example, two fragment links to the same matching address can count as the same result, while two different paths on the same domain cannot. Do not remove every query parameter or trailing slash merely to create more matches.

The tool does not follow redirects or inspect canonical tags. If two addresses genuinely represent the same page, verify that separately and prepare a consistent source file. The Canonical Tag Checker can support a separate investigation of declarations; a declaration is not proof of Google’s selected canonical.

1. Choose the mode. Decide whether the task needs word similarity or a comparison of imported result URLs.
2. Prepare the keywords. Keep relevant phrases, one per line, with their meaningful modifiers.
3. Set the rule. Choose a word-overlap threshold or result depth and minimum shared URLs.
4. Check the evidence. Review ignored tokens in word mode; review completeness and context in imported-results mode.
5. Complete verification. The displayed code expires after ten minutes and works once. Request a replacement when necessary.
6. Generate the review. Inspect groups, individual keywords and any not-compared cases.
7. Open the direct matches. Read the words or URLs that connect the terms, including indirect-connection warnings.
8. Save the evidence. Download the full report or a project draft before leaving the page.

Group labels are representative supplied keywords. They are not automatically chosen primary targets. A single keyword remains useful evidence too: it simply has no qualifying partner in the supplied comparison.

Search and filter the report when working through a larger list. The documented downloads retain the full report, rather than only the current screen. Keep that distinction clear when passing a report to another reviewer.

Export the Mapping CSV and open it in the Keyword Mapping Tool. Each keyword starts with a blank target URL, unknown intent and a review action. Those defaults leave the editorial choice with you.

Choose a primary phrase and destination only after reading the queries, reviewing the relevant results and checking existing pages. Several phrases may belong on one useful resource. Another phrase in the same connected group may need a different page or no new content at all.

Once destinations are agreed, use the Topical Map Generator to arrange the pages and their relationships. A topic structure should express a sensible visitor journey, not just reproduce the largest mathematical cluster.

For an individual page, use the Content Brief Generator to record the audience, purpose, evidence and scope. Explain what the page should help someone accomplish. A list of matching terms is not a complete writing brief.

Save a project JSON to keep inputs and settings. The published workflow also supports restoring inputs from a report JSON. Restored inputs are material for a new review, not evidence that the old result represents a fresh search snapshot.

Projects are not automatically saved in your browser or stored as WordPress content. Download what you need before leaving. Retain the original research and a note about any changes made during preparation so you can explain a later grouping difference.

CSV exports prefix spreadsheet formula-like values with an apostrophe; JSON retains the original text. Preserve the JSON when original wording matters. This documented precaution should not be treated as a guarantee of identical behavior in every spreadsheet application or after repeated saving.

No export changes your website. Assigning a URL in a planning workflow and publishing or consolidating an actual page remain separate actions.