Menu

Cannibalization Analysis

Find queries where two or more of your pages get clicks, and see how evenly the clicks are split.

Updated

On this page

What it shows

Cannibalization Analysis lists queries where two or more of your pages show in Google. For each query, it shows the pages, their clicks, and how evenly the clicks are split.

When two pages compete for one query, Google can show a different page for the query on different days. Then the clicks split between the pages. But two pages for one query are not always a problem.

When to use it

  • You have many pages on similar topics.
  • The ranking page for an important query changes often.

How to use it

  1. Open Optimize → Cannibalization Analysis.
  2. Check the date range and search type. The default is Last 28 days and Web. Discover is not available for this tool. If you select Discover, the page shows “Cannibalization data could not be loaded for the selected filters.” Select a different search type.
  3. Select All, Branded, or Non-Branded queries.
  4. On By Traffic, expand a query to see its pages. The queries with the most clicks are first.
  5. On By Severity, read the queries with the most even click split.
  6. Open the pages. Decide if they answer the same question.

How to read the result

Entropy shows how evenly the clicks are split between the pages. The tool calculates the Shannon entropy of the click shares, then divides it by the highest possible value for that number of pages:

entropy = −Σ (p × log2 p) ÷ log2(number of pages)

Here, p is the share of clicks of each page. The result is from 0 to 1:

  • Near 0: one page gets almost all clicks.
  • Near 1: the clicks are split evenly.
LabelMeaning
Severe (>0.7 entropy)Queries with entropy more than 0.7.
ConsolidateEntropy more than 0.7 and 3 or more pages.
Review intentEntropy more than 0.4.
MonitorOther queries.
Dominant ShareThe click share of the page with the most clicks.
Total Clicks at RiskThe total clicks of the listed queries. It is not a forecast of lost clicks.

Same or different intent

For the top two pages of a query, an AI model can compare the pages:

  • Same intent: the pages answer the same need. Review them.
  • Different intent — likely intentional: the pages answer different needs. This is often correct, for example a product page and a guide.

The model reads the page text with KongMetricsBot. If a site blocks the bot, the model uses the URLs only. Hide likely intentional pairs is on by default. Hidden rows are not counted in the totals.

Example

All data is fictional. For the query “example crm pricing” on example.com:

PageClicksShare
/pricing/6060%
/blog/crm-pricing-guide/4040%

Entropy = −(0.6 × log2 0.6 + 0.4 × log2 0.4) ÷ log2 2 = 0.97. The query is severe. The action is Review intent, not Consolidate, because there are only two pages.

If the AI model shows Different intent, the pricing page and the guide can both stay. If it shows Same intent, compare the two pages and decide which page must rank.

Limits

  • The tool shows a signal for review. It does not prove that the pages harm each other.
  • Do not merge, redirect, or delete pages because of a label only.
  • The AI check uses only the top two pages. Other pages of the query can also compete.
  • The tool uses up to 200 queries and up to 100 pages for each query.

If you see no rows

“No keyword cannibalization detected” means that no query had two or more of your pages in your range.

Choose which optional cookies we may use on kongmetrics.com and in the Kong Metrics App. You can change this at any time from "Cookie preferences" in the footer. Cookie Policy