Keyword cannibalisation

Keyword cannibalisation is what happens when two or more pages on your own site are built to win the same search. Nobody plans it. It arrives page by page, as a service page, then an article about that service, then a landing page somebody built for a campaign, until three addresses are all raising a hand for the same question and the search engine has to choose between them.

In short

• Several of your own pages compete for one search.
• There is no penalty. Google just picks a page, and it may pick the wrong one.
• The symptom is pages swapping positions for one query, not a sudden drop.
• Fixing it means deciding what each page is for, then merging the rest.

How a site ends up doing this to itself

Sites are rarely built in one sitting. A service page goes up in the first year. Somebody writes a blog post about the same service two years later, because the subject came up. A campaign needs a landing page, so one gets built quickly and never gets removed. Each of those decisions was reasonable on the day.

The result is three pages that answer the same question. None of them was written with the other two in mind, because the person writing the third had no reason to go looking for the first.

This is why cannibalisation is a symptom of how a site grew rather than a mistake somebody made. It shows up most on sites that have been publishing for a while, which is to say the sites that are otherwise doing the right thing.

What actually goes wrong

The honest version is less dramatic than the name suggests. Google does not spot two competing pages and dock you for it. It reads both, decides which one answers the search, and shows that one. The page it picks is often not the page you would have picked, and it can change its mind, which is where the visible damage comes from. A visitor who wanted to book arrives on an explanatory article instead of the page with the form on it.

There is a second cost, though it is worth being careful about how it gets described. The usual agency version, that your authority is divided three ways like a cake, is not how Google says it handles this, and Google has pushed back on it more than once. What is observably true is duller: three half-finished pages on a subject are three pages nobody has made good, and the effort that produced them would have made one page strong.

The check that settles it

Guessing at cannibalisation from a list of page titles produces false alarms. There is a direct way to look.

In Google Search Console, open the performance report, filter to a single search query, then switch to the pages view. That shows you every address on your site that has appeared for that exact search. Compare a few months rather than a single week.

• One page, steady. Working as intended. Leave it.
• Two pages, one clearly dominant. Also fine. Google has made a choice and stuck to it.
• Two or more pages trading places month to month. This is the real signal. The engine is undecided, and undecided usually means both pages rank lower than one confident page would.

The second most useful check takes ten seconds: search your own site for the phrase, using site:yourdomain.com followed by the search term, and see how many of your own pages come back looking like credible answers to it.

The version that is not a problem

Plenty of what gets called cannibalisation is just a site covering its subject.

Two pages mentioning the same words are not competing. A dental clinic will say “implants” on the implants page, the FAQ, the about page and half the blog, and that is what a site about implants should look like. Overlapping vocabulary is not overlapping purpose.

Competition needs both pages to be built for the same person at the same moment. Somebody comparing options and somebody ready to book are two different readers, and two pages serving those two readers are doing separate jobs even though they share every important noun. Name the reader for each page. If the two names differ, there is nothing to fix, and you have just done the exercise that search intent exists for.

Deciding which page wins

Once the check confirms it, the work is a decision rather than a rewrite. Pick the page that should own the search. Not the one that ranks best today, though that is worth knowing, and not the newest. Pick the one that best matches what the searcher wants to do, because that is the one the engine will keep choosing anyway.

Then handle the others. Anything useful on them moves into the winner. The addresses redirect to it, permanently, so that visits and links follow. If a losing page covers a related but distinct question, it can stay and be reframed around that question instead, with a link to the winner.

What does not work is editing both pages to be slightly more different while leaving both aimed at the same reader. The engine still has to choose, and it will still choose unpredictably.

Preventing it costs nothing

Cannibalisation is cheap to avoid and tedious to repair, which is an argument for deciding before publishing rather than auditing afterwards.

Before a new page is written, name the one search it exists to win, then check whether anything on the site already claims that search. If something does, the new page either replaces it or serves a different reader. Two sentences of thinking, written down somewhere, prevents the whole problem.

A topic cluster is the formal version of that habit: every page in the group has a stated job before anybody writes it, so the overlap gets resolved on paper.

Words you will hear

• Query. The exact phrase somebody typed. Cannibalisation is measured per query, not per subject.
• Consolidation. Merging several overlapping pages into one and redirecting the rest.
• 301 redirect. A permanent forward from an old address to a new one, so visitors and accumulated value arrive at the surviving page.
• Canonical tag. A line of code naming the preferred version of a page. It is for near-identical duplicates and does not fix two genuinely different pages competing.
• Content pruning. Removing or merging pages that earn nothing, which often lifts what remains.
• Site diversity. Since 2019 Google generally shows at most two results from one domain for a search, with subdomains usually counted as part of the parent. It makes exceptions where it judges more to be useful.

If you only do one thing

Open Search Console, sort your queries by impressions, and take the top ten. For each one, switch to the pages view and count how many of your own addresses have appeared for it in the last six months.

Most of them will show one page. The two or three that show several are your list, in priority order, and you now know which pages need a decision made about them before anybody writes anything else. What each page is supposed to be for in the first place sits under on-page SEO.