Keyword cannibalization is when two or more pages on your own website compete for the same search term. Google can't tell which one you want ranking, so it splits its attention, and usually neither page performs as well as one strong page would.
Watch the video
I've just run this check on a website I've been adding pages to since 2003. Twenty-three years of articles, hub pages and course pages stacked on top of each other. I used Claude with the SEMrush connector to find the clashes, and it found more than I expected, including one on the site's biggest keyword.
Here's the process, what it found, and — more usefully — the three places where following the AI's advice would have cost me traffic.
What keyword cannibalization actually is
Two pages, one search term, one website. Your own pages compete, the ranking signals split between them, and both sit lower than a single consolidated page would.
It isn't the same as duplicate content. The pages can be completely different in wording and still cannibalise each other, because what matters is the search intent they target, not the words on the page.
Why old websites accumulate it
On a site that's been running for two decades, this isn't anyone's fault. It's arithmetic.
You write an article. Two years later you build a proper hub page on the same subject. Someone adds a product or course page. Nobody goes back and reviews the older material, because nothing appears to be broken. After twenty years you have three or four pages all telling Google roughly the same thing.
I've worked in SEO for sixteen years and I knew this was happening. I hadn't gone looking, because on a large site finding it by hand means exporting keyword reports, sorting by keyword, and hunting for duplicate rows. It takes days.
How to find keyword cannibalization with Claude and SEMrush
The SEMrush connector plugs into Claude with no code — you sign in and it's available. The process is three steps:
- Pull the ranking data. Ask Claude to fetch every keyword the site ranks for in the top 50, with the URL that ranks for each one.
- Group the clashes. Ask it to list every keyword where two or more different URLs appear.
- Classify each clash. For each group, ask which page should win and what the fix is: merge, redirect, differentiate, noindex, or leave alone.
One warning on cost. Pulling 300 keywords in a single request used 3,000 SEMrush API units. Start with your top 100 to 200 keywords rather than the whole site.
I wrote up the wider version of this workflow in how to use Claude for SEO audits with SEMrush, which covers the connector setup in more detail.
What it found on a 23-year-old site
Six patterns, all of which will look familiar if you run an older site:
- A hub page losing to an old article. The site's biggest keyword, around 12,000 searches a month. The article I wrote years ago ranks 9th. The hub page that's meant to own that term ranks 47th.
- Five URLs splitting one keyword. Including dated booking pages. None of them in the top 10.
- A provider-specific page competing with a general page for the same branded searches, with the weaker page pulling attention from the stronger one.
- A detail page competing with its own category hub.
- An article with last year's date in the URL, still ranking for more than a dozen terms, but ageing badly by definition.
- Two clashes that turned out to be fine, which is the part most articles on this subject skip.
What to do with each clash
There are only five outcomes, and picking the right one matters more than finding the clash in the first place:
- Merge — two thin pages covering the same intent become one stronger page, with the weaker URL redirected.
- 301 redirect — the losing page has nothing unique to contribute, so it points at the winner permanently.
- Differentiate — both pages deserve to exist, so you rewrite headings and internal links to target clearly separate intents.
- Noindex — the page serves a purpose for users but shouldn't be in search results.
- Leave it — the two pages answer genuinely different questions and Google is happy showing both.
Claude drafts the merged outline and the redirect map — old URL, new URL — which is the tedious part, so that's where the time saving really lands.
Where the AI got it wrong
This is the section worth your attention, because it's the difference between using AI as a tool and letting it drive.
It flagged deliberate pages for removal. Some pages on the site are effectively pre-built search results for a service plus a town. They're indexed on purpose, because they capture long-tail location searches nothing else on the site would. Claude looked at the URL pattern and recommended noindexing them. Following that blindly would have wiped out traffic I built deliberately. Once I explained they were intentional, it reassessed and agreed to leave them. It simply had no way of knowing.
It treated different intents as a conflict. An explainer article and a commercial page both ranked for the same term. One answers "what is this", the other sells. Google is content to show both. The fix there is internal linking, not merging.
It misread the data. Sometimes the same URL appears twice for one keyword at different positions. That's one page showing up twice in the results, not two pages competing. Those rows need excluding before you analyse anything, or you'll invent problems you don't have.
Do you need the SEMrush cannibalization tool?
SEMrush has a dedicated Cannibalization report inside Position Tracking, and it's good. The catch is that it's a Guru plan feature — it isn't included on Pro.
If you're on Pro and don't want to upgrade for one report, the Claude approach above gets you most of the way there using data your plan already includes. If you're on Guru, use the built-in report, but note it needs a Position Tracking project set up for the site and a little time to gather data. My longer verdict on the platform is on my SEMrush page.
Check these before you delete anything
Redirects are easy to make and awkward to unpick. Before you act on any recommendation, AI-generated or not:
- Search Console. Which URL does Google currently show for that query, and how many impressions is each page getting?
- Backlinks. A page with external links pointing at it is worth redirecting, never deleting.
- Actual traffic. A page ranking 40th for one term may be ranking 3rd for something you didn't check.
- Intent. Do the two pages genuinely answer the same question, or do they just share words?
FAQ
Is keyword cannibalization always bad?
No. If two pages target the same term with different intent — one informational, one commercial — both can rank and both can be useful. It's a problem when both pages are trying to be the same page.
How do I know which page should win?
Usually the one with the stronger backlink profile, the better conversion role, or the more complete content. If a hub page is losing to an old article, the article's content usually needs folding into the hub.
Does merging pages hurt rankings?
Short term it can wobble. Long term, one strong page generally outperforms two weak ones, provided you 301 redirect properly and keep the content that was earning the rankings.
What happens next
I'm working through the fixes now, starting with the biggest keyword, then the five-page split, then the article with last year's date in it.
If you want the underlying SEO principles in plain English, that's what Website SEO for Beginners covers, and the SEO & Creator guides collection has the rest. If it's the Claude side you want to get comfortable with, the Claude AI for Beginners course starts from zero. Everything I run on day to day is listed in my daily stack.
0 comments