Cannibalisation is when two pages on the same site target the same query and end up splitting the clicks. This checker takes a list of page-and-keyword pairs and finds the places where that has happened in your plan: the same keyword assigned to more than one URL, and keywords that contain each other on different pages.
It works from the list you paste, not from live rankings. That makes it a planning check you can run before publishing, and a quick audit of a content map you inherited.
The list it expects
Two lists, one pair per line. Put the keyword in the left box and the URL you plan to target with it in the right: line 1 of each forms the first pair, line 2 the second, and so on. The tool pairs them by position, so the two boxes have to stay in step.
A blank line skips that pair instead of shifting everything under it, and any line starting with # is ignored, so you can keep notes beside the data. If one list runs longer than the other, the unpaired lines are dropped and the result says how many. Pasting a two-column CSV export into the URL box, while the keyword box is empty, still works: each line is split on the tab, pipe or comma.
How matching works
With normalisation on, each keyword is lowercased, split into words and stemmed, then joined back into a key. Keyword Density, keyword density and keyword densities all become keyword density, so they count as the same target. Turn normalisation off and only the case and surrounding spaces are ignored.
The stemming is deliberately simple. It strips a final s, turns ies into y, and removes ing or ed from longer words. That catches plurals and regular verb forms, but it also produces keys such as runn from running and boxe from boxes, which will not match run or box. If a pair you expected does not appear, the stem is the first thing to check.
Conflicts and overlaps are different
A conflict is the clear case: the same normalised keyword is assigned to two or more URLs. The table lists each one with its page count and the URLs involved. A conflict is a decision you have not made yet, even if the pages are deliberately different.
An overlap is softer. It appears when every word of one keyword also appears in another, and the two keywords are not on exactly the same set of URLs. keyword density and keyword density checker overlap. That can be a sensible hub and spoke, where the short phrase owns the guide and the long phrase owns the tool page. It only becomes a problem when both pages use the shorter phrase in the title and H1.
A worked example
Seven rows covering six unique keywords produced one conflict and two overlaps. The conflict was keyword research, assigned to both /blog/keyword-research and /guides/keyword-research. The overlaps were keyword density against keyword density guide, and keyword density against keyword density checker.
That is a realistic result: one genuine duplicate, two pairs worth a look, and four rows that were fine. The status line read 1 keyword conflict found across 7 rows, and the CSV contained only the conflicting keyword, ready to paste into a planning sheet.
What to do with a conflict
Pick the page that should own the query, then make the others support it. The usual options, in order of strength, are a permanent redirect from the weaker URL, a rel="canonical" pointing at the owner, or a rewrite that gives the second page a genuinely different angle and keyword.
Google's canonicalisation guidance is specific about the order: redirects are the strongest signal, a canonical link element is strong, and sitemap inclusion is weak. It also warns against using robots.txt or noindex to consolidate duplicates, because both can remove the page from search rather than merge it. A canonical tag on the page itself should point to itself, not to a third URL.
What this check cannot tell you
It cannot see rankings. Two pages can compete for a query without either one naming it in a target keyword, and a listed conflict can be harmless because one page never ranks. Confirming real cannibalisation needs query and landing-page data from Search Console: the same query showing two URLs, with impressions split between them.
It also cannot judge intent. Two pages can target the same words with different jobs, such as a definition page and a calculator, and both can deserve to exist. The tool flags the overlap; you decide whether the overlap is a mistake.