Keyword cannibalisation: stop competing with your own pages
When two of your pages fight over one query, the engine chooses for you — and often chooses wrong. Here is the diagnosis and four remedies.
What counts as cannibalisation
Cannibalisation is not simply a keyword appearing on several pages; that is normal. The problem arises when several pages claim the same intent and the engine keeps switching its choice.
The symptoms: two or more of your URLs appearing for one query in the Performance report; unstable positions swinging widely; the ranking page periodically swapping without any change on your side.
Not cannibalisation: a commercial page and an informational article on adjacent phrasings with different intent; pages from different language versions; a product page and a category when the queries genuinely differ in specificity.
Diagnosis in Search Console
Open the Performance report, go to the Queries tab, select the problem query, then switch to the Pages tab. You will see every URL that appeared for that query with impressions and clicks.
Two URLs with comparable impressions on one query is clear cannibalisation. If one URL takes 95% of impressions and the other appears occasionally, that is noise and needs no action.
For a systematic sweep, export a quarter of data via the API or built-in export, group by query and count unique URLs per query. Queries with two or more URLs form the working backlog.
Where it comes from
Source one: a blog duplicating commercial pages. A 'How to choose PVC windows' article and a 'PVC windows' service page start competing once the article carries sales blocks and commercial anchors.
Source two: regional and filter clones. City pages differing only by toponym compete with each other and with the generic service page, because they are substantively identical.
Source three: history. Over the years three landing pages were built for the same query at different times and nobody removed the old ones. On sites older than five years this is the most common cause.
Four remedies
Merge: when pages are close in content, consolidate the best into one and 301 the others to it. That combines link equity and engagement instead of splitting them.
Refocus: when the pages could serve different intents, separate them. Rewrite the article for informational intent and strip its sales blocks; strengthen the commercial page with pricing, terms and an enquiry form.
Canonicalise: when users need both pages (a filtered view and the parent category, say) but content nearly matches, put rel=canonical from the secondary to the primary. Gentler than a redirect and keeps both accessible.
Delete: when a page serves neither user nor business and has no links or traffic, remove it with 410 and drop it from the sitemap. This is the most underused option — extra pages always feel worth keeping, and they get in the way.
Prevention during planning
Maintain a query-cluster-to-URL map and consult it before creating any new page. If a cluster is already owned by an existing URL, a new page is justified only by a demonstrably separate intent.
Separate intents explicitly: commercial queries get service or category pages; informational queries get articles; navigational queries get contact or brand sections. One page serves one intent.
With multiple authors or contractors the map is mandatory and must be shared. Most cannibalisation happens not by design but because a new copywriter did not know the existing page existed.
Confirming it improved
After merging, look at total impressions and clicks for the query cluster rather than position. The correct picture is cluster traffic rising while the number of URLs shown for it falls to one.
Give the change time: after a 301 the engine must recrawl both pages and re-evaluate the target. On a mid-sized site that takes two to six weeks.
Verify the redirects did not form chains and that internal links now point at the new URL. Links left on the old address keep dragging the crawler through a redirect and slow consolidation.