What are content pruning and keyword cannibalization?
Pruning answers two questions about an existing site: which pages drift far enough from the main topic to blur its profile, and which are so similar to each other that they compete for the same query.
Cannibalization is that second case: two of your own URLs aiming at the same intent. Pruning responds to both - it is not about deleting everything old, but about deciding which page should carry the topic.
What do you get?
- A table of pages drifting off your site topic, each with a proposed action: remove, redirect or rebuild.
- Groups of competing pages, with a similarity score and a proposal: pick the main page and set up redirects, merge the content, or split it across different queries.
- A topical summary of the site: the main topic and side topics derived from the content of your published pages.
- CSV export. When the analysis finds nothing to fix, it tells you so.
Main site topic: mortgage, creditworthiness, personal loan (161 pages)
Side topics: rental income tax, letting a flat (14) · cost of building a house, emergency fund (10) · overpaying a loan, repayment schedule (7)
These are pruning candidates — pages whose topic drifts away from the main topic of the site. Consider removing them, redirecting with a 301 to a matching article, or rebuilding the content.
Cannibalization
Pages in one group compete for the same intent — a high similarity means the search engine cannot tell which one to show. Pick the strongest, merge the rest, or split them into separate subtopics.
Why tidy up your content?
Sites grow in layers: old blog posts, seasonal campaigns, texts targeting phrases nobody searches for any more. Every such page blurs the picture of what the domain is about - and that picture is what search engines and models use to decide where you are a credible source.
The second problem is worse because it is invisible: two of your own pages aiming at the same query. Instead of one strong position you hold two weaker ones, and in a generative answer the model will pick a single source anyway - possibly somebody else.
Neither can be judged by eye across a few hundred URLs. Every page has to be compared with every other.
How does the analysis work?
- 1
Reading the sitemap
We read the sitemap along with any nested files and skip images and other media. The addresses are checked for availability before the run starts.
- 2
Fetching pages
From each URL we take the title, H1 and meta description - enough for topical comparison and fast to collect.
- 3
Semantic representation
Every page gets a vector describing its content, so we compare meaning rather than word overlap.
- 4
The site main topic
We group the pages and derive the main topic plus side topics from characteristic two-word phrases.
- 5
Pages drifting off-topic
We measure how far each page sits from the main topic and flag the ones above the threshold you set.
- 6
Cannibalization groups
We compare pages pairwise and group those whose similarity crosses the threshold.
- 7
Recommendations
The model describes what was found and proposes an action for each case.
What you set before the run
| Parameter | Range and default |
|---|---|
| Deviation percentile | 50-99 (default 90) - above it a page is listed as drifting off-topic |
| Cannibalization threshold | 0.7-1.0 (default 0.9) - the similarity at which pages count as competing |
What does it cost?
- 1 credit per analysis, regardless of how many pages the sitemap holds.
- Pages that could not be fetched are excluded from the deviation ranking - a network error should not look like off-topic content.
- A job that ends in an error returns the credit automatically.
Frequently asked questions
Will the tool delete my pages?
No. The analysis is advisory only - you get a list of pages and a proposed action, while the decision and the execution stay with you. We do not connect to your CMS.
Large site - one sitemap or several?
For an extensive site it is better to submit section sitemaps separately: blog, categories, products. The results are sharper, because the main topic is then computed within one section instead of being averaged across the whole domain - and it is that topic which decides which pages count as drifting.
How is cannibalization different from ordinary topical overlap?
By threshold and consequence. Two pages on a related topic are normal; cannibalization starts where they are similar enough that the search engine has no reason to prefer one. You set that threshold yourself before the run.