Clustering search queries
Insert a list of requests and get a breakdown by content and by page: what to combine on one page, what to separate, and where you compete with yourself.
Queries are grouped by expected response rather than by word matches: “buy a rug” and “make your own rug” contain the same word and require different pages. The tool sorts the list by intent, suggests page structure, and shows where two of your pages are targeting the same intent and interfering with each other.
AI answers are free, but limited by the overall site limit. If the limit is reached, come back later: repeated clicks will not speed up the response.
You should group by expected answer, not by words
A classic mistake when assembling semantics is to group by word co-occurrence. “Buy a rug,” “do-it-yourself rug,” and “mat reviews” contain one word and require three different pages, because the person behind each request wants to see something completely different. Put together, they produce a page that doesn't answer any of them.
The correct sign is the expected answer. If a person expects to see the same page for two requests, they are in the same cluster. If they are different, they must be separated, even when the wording is almost the same.
A separate check is cannibalization: when two of your pages target the same intent and interfere with each other. The search engine chooses from them itself and often chooses the wrong one, but in total both receive less than one would receive. On this site, four such pairs were hand-glued in July.
What the tool does not do: it does not invent frequency and does not claim that it is in the search results. He didn't see her. Intent is determined by the wording, and where the wording is ambiguous, it is flagged as requiring eye inspection.
What the tool intentionally doesn't do
Does not insert numbers, deadlines, benchmarks and names of companies that did not exist in your input. If the output requires a number that you did not provide, in the answer will be marked “to clarify” rather than a plausible number. This limitation is worth in the system prompt and works against the main property of language models: It’s difficult to complete what’s missing.
The request text is sent to the AI provider, and the response is returned to the browser. The site does not save these texts in the database. To limit expenses separately the number of requests and the anonymized connection identifier are taken into account. If the free limit is reached, the tool will offer to come back later.