Keyword Cannibalization Checker: Free, Runs in Your Browser

Privacy: nothing you paste or upload leaves your device. The file is read and processed by JavaScript in this browser tab. Nothing is sent to a server or saved.

Most cannibalization checks stall at the data step. Search Console’s Performance export puts queries and pages in separate tabs, so you can see that a query had impressions and that a page had impressions, but not which page earned them for which query. This checker needs the joined version, and the three ways to get it are in the how-to under the tool. Once you’ve got your list of flags, the keyword cannibalization guide covers how to fix each one.

How to Use the Keyword Cannibalization Checker

Six steps. Only the first one is fiddly, because the data you need isn’t in the standard export.

Three ways to get query and page together

The export button in the Performance report gives you a Queries table, a Pages table and a few others, each grouped by one dimension. Google’s help page also notes that exported data is truncated to 1,000 rows. That’s why a Search Console cannibalization report has to come from somewhere else. Any of these works.

How to Read the Results

The summary tells you how many queries were checked, how many were skipped and why, and how many crossed the line. Then come two tables.

The flagged queries table

Impressions at stake is every impression your pages got for that query, added up page by page. Pages are labelled A, B, C in order of impressions, each with its share of impressions, its share of clicks and its average position. Greyed pages showed up, just not enough to count as competing.

Click any column header to sort. Impressions at stake is the default, because a split on a 3,000 impression query matters more than a split on a 40 impression one.

What each suggested action means

The page type comes from the URL path: /blog/, /guides/ or a dated path reads as an article, and /products/, /collections/, /category/ or /services/ reads as commercial. If your URLs don’t say (lots of WordPress sites put everything at the root), the checker falls back to Merge and says it couldn’t tell. That’s your cue to look at the results page yourself.

The page pairs table

This one turns the flags around. Each row is two of your URLs and how many flagged queries they split. A pair that collides on 1 query is a small problem. A pair that collides on 15 is structural: two pages written for the same topic, so every query in that topic gets split. Fix the pair once and all 15 queries move.

Worked example: the sample data

Click Load sample data. It’s a made-up outdoor shop, Acme Trail, on example.com, exported the Looker Studio way: 34 rows, 17 queries, with acme as the brand term. Two brand queries get excluded. Merino hiking socks has 9 impressions, so it falls under the minimum. The size guide’s #boots jump link folds into the main size guide URL, so that query isn’t flagged. That leaves 11 flagged queries: 6 merge, 3 differentiate, 2 ignore.

The biggest flag is waterproof hiking boots, with 3,220 impressions at stake. The hiking boots collection takes 57% at position 14.2, and a blog post on the best waterproof hiking boots takes 41% at 17.8. Article versus collection, so the checker says differentiate. The post is chasing a buying query, and neither page is on page one.

The pairs table shows the bigger story. The boot care guide and the how to waterproof boots post collide on 3 queries (how to waterproof hiking boots, hiking boot care and how to clean hiking boots), and both sit between positions 8.1 and 19.5. That’s one topic written twice. Merge.

Trekking poles is a different kind of merge: the collection and the same collection with ?sort_by=price-ascending split the query. A canonical tag or a 301 sorts that out, and nobody has to rewrite anything. And two flags say leave it: trail running shoes, where the collection holds 89% of impressions, and hiking backpack 30l, where both pages already average position 3 or better.

When Split Impressions Aren’t Cannibalization

A flag is a question. Plenty of splits are fine, and the checker can’t see the results page.

• Sitelinks and double listings. When Google shows two of your pages for one search, each URL logs its own impression when data is grouped by page (Google explains the counting here). Two of your pages in the top 3, or a result with sitelinks under it, is usually a win.
• Two real intents. A query like hiking boots can mean show me boots or help me choose. If Google keeps your category and your guide on page one together, it may want both.
• Brand searches. Your home page, about page and product pages all showing for your name is normal. That’s what the brand filter is for. If you haven’t got a clean brand list, the branded vs non-branded keyword splitter helps you build one.
• Short tests. Google sometimes tries a second URL for a week or two. One export can’t show that, so compare two date ranges before acting.

When the split is real, the pages chase the same searcher and neither climbs. The keyword cannibalization guide has the decision framework for that, and the Search Console walkthrough shows the manual check, click by click, for confirming a single query. If a merge means retiring dozens of URLs at once, read how to redirect thousands of URLs without losing traffic first.

Limits of This Checker

• It only sees your file. Search Console leaves out anonymized queries (ones searched by too few people), and the 1,000 row interface cap and 50,000 row API cap mean a big site’s export is the top of the data. Long-tail splits can hide below that.
• No time dimension. The clearest sign of cannibalization is the ranking URL flipping between periods. One export is a single snapshot, so run two date ranges and compare, or check the top flags by hand.
• Page type is a guess from the URL. It doesn’t open your pages, read titles or compare content. Two pages with different intents can still get Merge if their URLs look alike.
• Impressions at stake runs high. It adds up page-level rows, and a search that showed two of your pages counts for both. Expect it to be bigger than the same query’s figure in the Queries tab.
• The brand filter is a plain contains match. A brand term of art would also catch smart watch. Use your full brand name, or a few spellings of it.
• Thresholds are judgment calls. 10% and 75% are starting points. Google doesn’t publish a threshold, so move them and watch how the list changes.

For a sweep that also covers rankings Search Console hides, the Ahrefs and Semrush method finds candidates from rank data. Confirm them with your own Search Console numbers before you change anything.

Keyword Cannibalization Checker FAQ