Faragiri

Learn Cannibalisation Keyword research

A Stockport mower dealer had four pages chasing the same query and swapped between them for a year, until three were merged into one and the position halved.

4Sections
4Min read
4Depth

The tell is a position that swaps between your own pages, not a position that falls. Marple Mowers, which services and sells garden machinery in Stockport, spent 2025 believing it had a ranking problem for lawnmower service stockport. It had a filing problem: four pages appeared for lawnmower service stockport across one year, and the engine chose a different one in nine of the twelve months.

The four were a services page, a blog post about spring servicing, a location page for Stockport and a category page listing mower brands. Nobody at the shop had decided any of them should answer that query. Three of them collected it by accident, which is how this nearly always happens.

Finding it in forty seconds

Check the pages column in Search Console, filtered to one query. Filter the performance report to the query, then open the pages tab, and if more than one address is listed the site is competing with itself. Marple's showed four addresses with 62, 41, 28 and 9 impressions, which is one query's worth of demand cut into four pieces.

Diagnosis

One query's demand cut into four pieces

Filtering Search Console to a single query and opening the pages tab reveals four URLs splitting the same impressions.

Impressions recorded across four competing pages on Marple's site for one query

One queryFour URLs competing
  1. Services pageHeld the largest share of impressions and became the surviving page.62
  2. Spring servicing blog postDrew searches away from the commercial page.41
  3. Stockport location pageMatched the town name and competed on the same term.28
  4. Brand category pageCollected stray impressions without anyone intending it.9

Splitting impressions is a filing problem rather than a penalty; the engine simply picks a different candidate per search.

What the report does not show is a penalty, because there is not one. An engine holding four candidates from one site for one query picks one, and the choice is made per search from signals that move, which is why the position looks unstable rather than bad. Two pages on one query is a filing problem, not a writing problem, and no amount of rewriting any of the four fixes it.

What merging actually did

Marple kept the services page and merged the other three into it in February 2026. Merging three into one moved the position from 11.4 to 5.2 over the following three months, and the total impressions for the query went up rather than down, which is the part people expect to go the other way.

Cannibalisation

Four queries, before the merge and three months after

The query that was split four ways moved furthest, and the one that only ever had one page behind it barely moved at all.

Average position, Marple Mowers, January against May 2026. A lower number is nearer the top.

BeforeAfter

  1. lawnmower service stockport11.45.2Four pages appeared for it across the year, and the engine swapped between them.
  2. mower blade sharpening9.16.4Two pages, both merged into the same survivor.
  3. petrol mower repair near me14.88.9Collected the redirected authority without being written for.
  4. ride on mower parts7.67.3One page all along, and it moved by nothing worth reporting.

Marple kept the services page with eleven internal links and redirected the other three rather than deleting them.

Two other queries moved with it without anybody writing for them: mower blade sharpening went from 9.1 to 6.4 and petrol mower repair near me from 14.8 to 8.9, both of them collected from the merged pages. Ride on mower parts, which only ever had one page behind it, went from 7.6 to 7.3, and that row is the reason the other three can be believed at all.

The surviving page should be the one with the links, not the newest. The blog post was the best written of the four and had two links pointing at it; the services page had eleven, from the menu, the footer and nine other pages. Keeping the blog post would have meant rebuilding eleven links to prove something the site already believed.

Redirect the merged pages, never delete them. Each of the three now returns a permanent redirect to the services page, which passes on what those addresses had earned and keeps anybody who bookmarked them out of an error. Deleting them would have thrown away the 41 and the 28 rather than adding them.

Where this stalls: nobody owns the second page

A category page and a guide can want the same query without either knowing. The blog post was written by somebody in March who had never seen the services page, and the category page was generated by the shop software. Neither author made a mistake, and no meeting could have caught it, because the collision only becomes visible in a report nobody reads by query.

The stall

Why overlapping pages stay invisible and keep getting rewritten

Without looking by query first, unowned pages split demand quietly and get mistaken for weak copy.

Silent collision

  • 01Unowned page appears
  • 02Signals split quietly
  • 03Page-first reports hide it
  • 04More writing is added
  1. 01Unowned page appearsA new post or auto-generated category page targets a query already covered.
  2. 02Signals split quietlyNo error or warning appears; position simply bounces between six and eleven.
  3. 03Page-first reports hide itReading performance by URL makes each page look like ordinary underperformance.
  4. 04More writing is addedTeams rewrite copy or publish again instead of merging the competing URLs.

Rewriting any of the four pages never fixes a collision; only consolidating ownership breaks the loop.

The stall is that cannibalisation is invisible until somebody looks for it on purpose. It produces no error, no warning and no dramatic fall; it produces a page that sits at eleven and occasionally at six, which reads as ordinary underperformance and gets treated with more writing.

Preventing the next one

One query, one page, written down before anybody writes. A sheet with two columns, the query and the page that owns it, checked before a new page is commissioned, is the whole of the prevention, and it is one query, one page, decided deliberately.

Prevention

Three checks that stop the next collision

Preventing overlap takes a shared ownership sheet, a title crawl before launch, and the right filter order in Search Console.

  1. Two-column query sheetCommissioning a new page on a query another URL already ownsUnassigned queries that genuinely need a new page
  2. Screaming Frog title auditDuplicate or near-duplicate titles before impressions startDistinct page titles matching distinct searcher intent
  3. Query-then-pages filterMulti-URL splits hiding inside page-first reportsSingle-URL queries with clean historical attribution
One query, one page

Filtering query first and pages second is free in Search Console; Screaming Frog checks titles before launch.

Search Console is the only free tool that shows this properly, and it replaced the keyword data Google removed from analytics in 2011. The filter combination that matters is query first and then pages, in that order; doing it the other way round shows the queries one page gets and hides the collision entirely.

Screaming Frog SEO Spider finds the other half, which is pages with duplicate or near duplicate titles, free up to a few hundred addresses. It replaced opening every page and reading the source, and two pages titled Mower Servicing and Lawnmower Servicing in Stockport is the same collision visible before any impressions have been recorded.

What each of the four pages should have been aimed at instead is what the searcher wants, which the words hide, and the three that were merged are also pages that dilute the ones that matter seen from a different angle. Both belong to finding what people type, and what they mean by it. Marple changed nothing else in that quarter, which is unusual and is the only reason the 11.4 and the 5.2 can be attributed at all.

Alongside this

Above this