Keyword cannibalization is rarer than the reports suggest, and two of your pages ranking for one query is usually not it. On our own site that happens to nearly half of them, and almost none of it is worth fixing.
The usual advice says find the overlap and merge. That advice has a poor record, because it treats every split as the same problem.
Here’s what the split looks like when you measure it, and the one test that tells you which ones matter.
What We Measured
We pulled ninety days of Search Console data for this site, cut by query and landing page, then counted the distinct URLs behind each query.
| Queries with | Count | Served by 2+ URLs | Served by 3+ URLs |
|---|---|---|---|
| 50 or more impressions | 600 | 266 (44.3%) | 152 (25.3%) |
| 100 or more impressions | 386 | 173 (44.8%) | 108 (28.0%) |
| 250 or more impressions | 191 | 89 (46.6%) | 58 (30.4%) |
The proportion barely moves as the volume threshold rises. Splitting isn’t something that happens to your weak queries. It happens to all of them, and it’s the normal condition of a site with a few hundred pages.
Why Most of It Doesn’t Matter
It doesn’t matter because a query and an intent aren’t the same thing. Two pages can both surface for one phrase while answering different questions behind it.
Someone searching a tool name might want pricing, a comparison, or instructions. Three pages serving that phrase isn’t a conflict, it’s coverage, and Google picking between them by context is the system working.
The rule that survives contact with a real site
Same topic plus different intent means keep both pages. We learned that from a consolidation that merged on topic alone and cost us the traffic of the page that got absorbed.
The Split That Does Matter
You’re looking for two URLs taking a near-equal share of one query while answering the same question. That’s the case where neither page accumulates the signals it needs, and both sit lower than one would.
On our data that’s 17 queries out of 386, using a simple threshold: the second URL earning at least 80% of what the first one earns.
Seventeen is a working list. A hundred and seventy-three is a research project nobody finishes.
Why evenness is the signal
It matters because a lopsided split tells you Google has already chosen. One URL takes most of the impressions, and the second is picking up scraps from a few oddly-worded searches.
An even split says the choice is unresolved, and that’s the state worth ending.
How to Run This on Your Own Site
You already have the data in Search Console. The report you need is the one most people never open.
- Export queries and pages together. The default view shows them separately, which hides every split by design.
- Group by query and count distinct URLs. Anything with two or more is a candidate, and nothing more than a candidate yet.
- Sort by evenness, not by count. Compare the second URL’s impressions to the first’s, and work the ones above 80%.
This is the same shape of audit we ran on indexation in a real audit of 285 pages, and it takes about the same time.
What to Do With the List
Read both pages before you touch either. The fix depends on what you find, and merging is only one of four options.
- Different intents: leave them alone and link them to each other so a reader can cross over.
- One is clearly better. Point the weaker page’s internal links at the stronger one and let the difference widen.
- Both are half-answers. This is the merge case, and it’s rarer than the advice suggests.
- Neither is aimed at the query. Write the page that is, rather than editing two that aren’t.
The fourth case catches people out. A query landing on a homepage or a category page usually means no page was ever built for it, and no amount of merging creates one.
What This Data Cannot Tell You
It tells you which URL received impressions. It says nothing about why Google chose it, and any explanation of the mechanism is inference.
One property is also not a sample. The proportions here describe a few hundred pages in one subject area, and a larger or more varied site could look nothing like this.
What would make it stronger
The same cut across a dozen properties of different sizes, which would show whether the split rate tracks page count. We haven’t run that, so treat the percentages as one observation rather than a benchmark.
Frequently Asked Questions
What is keyword cannibalization
Keyword cannibalization is when more than one page on your site competes for the same query and the same intent, so neither accumulates the signals it needs. Two pages appearing for one phrase is not by itself cannibalization.
How do I know if I have it
You export Search Console by query and page together, count distinct URLs per query, then compare the second URL’s impressions to the first’s. An even split on one intent is the case worth acting on.
Should I merge pages that overlap
Only when both are partial answers to the same question. Merging pages that share a topic but serve different intents removes coverage, and it’s the most common way this fix backfires.
Is a homepage ranking for my keyword a problem
It usually means no page was built for that query. The answer is to build the page rather than to adjust the homepage.
How often should I check
Once a quarter is enough for most sites. Splits appear when you publish, so check after a run of new pages rather than on a fixed calendar.
Start With the Even Ones
Pull the query and page export this week and sort by how close the second URL is to the first. That list is short, it’s the only part of the problem that pays for the time, and it’s the part almost nobody looks at.