Keyword cannibalization, measured on 13,001 real searches

The textbook fix for keyword cannibalization — two of your own pages competing for the same search — is to merge them and redirect one into the other. On 26 July I did exactly that here. An old article about hiring a web designer in North York had been collecting impressions for "web design north york" without a single click, and when I launched a proper North York service page in July, the two started splitting the same searches. I moved the useful parts of the article into the service page and pointed the old address at it with a permanent 301 redirect.

Seven weeks later, Google is still showing the old page. Its own inspection tool says the old URL is indexed, is its own canonical, and was last crawled on 18 July — eight days before the redirect existed. The textbook fix has not happened, because Google has not come back to see it.

So I checked the rest of the textbook. Nine Search Console properties I can read — mine, a roofing and restoration company, a stucco contractor, a notary, a paralegal and four more — over 13 March to 9 September 2026, brand searches removed: 13,001 searches and 287,208 impressions. For every phrase with 30+ impressions I pulled every page that ranked, and for the 120 biggest splits I went day by day.

What is keyword cannibalization?

Keyword cannibalization is when two or more pages on the same website compete for the same search, so Google keeps switching between them instead of settling on one. It is not a penalty in any Google document I have read. The searcher simply sees whichever page Google picks that day, and that is not always your best one.

Two pages sharing a search is not automatically a problem: a pricing page and a guide can both deserve "how much does a website cost". It becomes cannibalization when the pages do the same job for the same person, so Google keeps trying both.

How common is keyword cannibalization?

Less common than a crude count suggests. Of 1,581 phrases with at least 30 impressions, 542 had two or more pages ranking — 34.3%. But in 211 of those 542, one page took 90% or more of the impressions and the second was a stray. Real splits, where no page held 80%, came to 256 phrases: 16.2% of the total.

The top two rows are what a lazy report calls cannibalization. Count every phrase that has two pages — the crudest way to build that list — and about four in ten entries are a second page that turned up by accident. Worth remembering before anyone bills you to "fix" them.

Real splits cluster by how a site is built:

Two shapes cover almost all of it. On the stucco site the homepage competes with the city pages: of the eight stucco phrases among my twelve most-switched examples, the homepage was in seven. On the roofing site, articles answer overlapping questions — three took turns on "roof depreciation".

How do you find keyword cannibalization in Search Console?

Open the Performance report, click a search phrase, then switch to the Pages tab. If more than one page is listed, look at how the impressions divide. One page with 90% or more: ignore it. No page above 80%: that is a real split.

1. Performance → Search results, set the range to the last six months so a quiet week does not fool you.
2. In the Queries table, click the phrase you care about. Search Console adds it as a filter.
3. Open the Pages tab. Every page listed there showed for that phrase at least once.
4. Divide each page's impressions by the total. Note the position each one reports.

One detail trips people up. Google counts it as a single impression when "two results from the same site appear for one query", but page rows are counted per page — so on a real split the page rows add up to more than the phrase row. In my data, a median of 18% more: roughly how often two of your pages shared one results page.

Does Google really take turns between your pages?

Yes, constantly. For the 120 biggest real splits I pulled each day and noted which page led. The median phrase appeared on 96 days, and its leading page changed 16 times across them. Four in five of those phrases changed leader at least five times. Only three never changed at all.

The last row is the textbook case: two articles asking almost the same question, swapping places 39 times. The first is the expensive one — 950 impressions over three pages Google never committed to, at 51.5. Sometimes Google shows several at once: on the median split, two or more pages appeared on 44% of the days.

Does Google show the wrong page?

About half the time. In 133 of the 256 real splits — 52% — the page Google showed most often was not the page that ranked best when it was shown. The median gap between the two was 10.7 places — roughly a full page of ten results.

Read the first row slowly. The stucco company's Mississauga page ranks 10.6 when shown; Google shows the homepage more than twice as often, at 34.4. In the second, someone asks about the top of a roof, the article about the peak ranks 7.9, and Google shows the one about the edge just as often, at 73.5. The third row is mine.

This is where a split actually costs something. On a real split, the phrase's reported position sat a median of 5.9 places behind what its best page managed on the days it was shown. For strays, the gap was 0.8. More than a third of real splits — 93 of 256 — sat 10 or more places behind their own best page.

Does keyword cannibalization cost you clicks?

Not at the same position, as far as I can measure. At positions 3.5 to 10.4, where most of this data sits, phrases with a real split were clicked 0.24% of the time against 0.26% for phrases with a single page. If there is damage, it is one step earlier: in the position itself.

Do not read that as "splits get more clicks". All 16 top-3 splits belong to the roofing site, and at 10.5 to 20.4 one phrase supplied 18 of the 32 split clicks — without it, 0.13%. What survives every cut is the middle row, and it says no difference.

So a merge helps, if at all, by stopping Google showing the weaker page — a position change, measurable only before and after. I have exactly one of those.

What happened when I merged two pages with a 301

In Google's index, nothing yet. I redirected the old article on 26 July. Google last crawled it on 18 July, still reports it as indexed and as its own canonical, and gave it 145 impressions at an average position of 13.1 between 9 August and 9 September. The page it redirects to got 57 impressions over the same weeks, at 50.4.

Read only the phrase-level column and this merge is a success: 28.7 down to 15.2, nearly twice the impressions a day. Filter by page and the improvement belongs to the page I retired. Google is still ranking its 18 July copy, which climbed from 31.2 to 13.1 untouched — why, I cannot tell you. Nobody is lost: a click on that old result lands on the service page. What is not happening is the consolidation, which was the entire point.

Google's own documentation is plain about the mechanism: "Googlebot follows the redirect, and the indexing pipeline uses the redirect as a signal that the redirect target should be canonical." Googlebot can only follow a redirect it has actually requested, and it has not requested the old address since 18 July.

Why Google has not been back in 55 days, I cannot prove. What I did: removed every internal link to the old article and took it out of the sitemap, as the usual merge checklist says — which also removed every invitation to return. And by accident, my RSS feed kept listing the old address until this morning, because its build script read the folder I kept for rollback. Fixed.

Two caveats. It will not be a clean test even when Google catches up: that same day I also rebuilt the service page from 1,447 to 3,553 words and added it to a city list on 84 other pages. And the next step is documented: Google's recrawl page names the URL Inspection tool, and warns that "requesting a recrawl multiple times for the same URL won't get it crawled any faster". Nobody has pressed it on this address yet.

How do you fix keyword cannibalization?

First decide whether the pages do the same job. If they do, keep the one whose type and address fit the search, move anything useful into it, and 301-redirect the other. If not, keep both and make each clearly about its own thing. If both sit past position 50, nothing needs merging: no page answers that search.

The fourth row leans on Google's link guidance: anchor text "tells people and Google something about the page you're linking to". A homepage linking to the Mississauga page as "stucco repair in Mississauga" tells Google which one it means. The fifth row is my "orillia web design": 375 impressions split between my homepage at 71.7 and my About page at 66.9. I have no Orillia page, so there is nothing to merge; the only fix would be building one.

When a merge is the answer, this is my order now — including the step I skipped:

1. Confirm it is a real split. Pages tab, last six months, no page above 80%.
2. Read both pages. Same job for the same person? A pricing page and a guide are not.
3. Pick the keeper by type and address, not impressions. My North York article had more impressions; the service page was the right page for someone hiring.
4. Move what is useful, then 301 the other. Google lists a redirect first among its consolidation signals: "a strong signal that the target of the redirect should become canonical".
5. Clear the old address everywhere — internal links, sitemap, RSS feed. Mine survived in the feed for seven weeks.
6. Two weeks later, inspect the OLD address. If its last crawl is older than the redirect, Google has not seen it yet.
7. Judge at eight weeks, filtered by page. The phrase-level number will credit the old page, as it did for me.

I have a live case waiting at step 2. My restoration-company web design page and my construction-and-restoration page both answer "restoration website design", and Google splits it between them and a case study. Nothing gets redirected until I have read both side by side.

Planning a merge or a move? Redirect mapping is part of my website migration service. If an SEO report handed you a long list of "cannibalized keywords", send me the domain and we will sort strays from real fights on a call; costs are on my pricing page. Background: what average position is an average of and how to read Google Search Console.

Sources, read in full on 11 September 2026: Google — Performance report data and aggregation for how two results from one site are counted; Google — Redirects and Google Search for what Googlebot does with a redirect; Google — How to specify a canonical URL for the order of consolidation signals; Google — Ask Google to recrawl your URLs; and Google — Link best practices for links and anchor text. Everything numeric is my own measurement rather than an industry study: Search Console data for nine verified properties covering 13 March to 9 September 2026, pulled read-only on 11 September 2026 — 13,001 non-brand searches and 287,208 impressions. Brand searches were identified per property from its own name and excluded, along with site: and inurl: queries. A "real split" means no page took 80% of a phrase's page-level impressions; a "stray" means one page took 90% or more. Click rates are impression-weighted across each group. The crawl date, index status and canonical for the North York URLs come from Google's URL Inspection API, read on 11 September 2026.