Cannibalization Audit With Merge or Differentiate Calls

Two URLs alternating in position for the same query is not automatically a problem, and the reflex to merge on sight destroys pages that were working. The test we apply is behavioural: across the full history Search Console retains, does that query's click-through rate sit below what its average position should produce, and do the two URLs trade places week to week rather than settling. On a typical consultancy archive, that filter reduced a list of roughly two hundred suspected conflicts to between fifteen and thirty real ones. The diagnosis: Query-URL pairs pulled week by week rather than as a sixteen-month total, because the alternation that constitutes the evidence is invisible in an aggregate. For every query where more than one URL earns impressions we look at three things. Whether the two URLs alternate, or one sits stable at the top while the other sits on page four, which is not a conflict at all. Whether the impression share moved on a date matching a publication or a deploy, which tells you what caused it. And whether the two pages answer the same question, or two different questions that happen to share vocabulary. The last of these is where tooling gets it wrong, and it takes a person reading both pages properly. Evidence we require before recommending a merge: Both URLs have to serve the same intent, not merely the same phrase. Traffic to the weaker URL has to be non-converting or genuinely duplicative rather than simply smaller. And the surviving page has to be able to absorb the material without becoming a page that answers two questions badly, which is the usual outcome of enthusiastic merging and is harder to reverse than the original conflict. Where those conditions are not all met, the recommendation is to differentiate instead, and we specify how: which page narrows, what comes out of it, which links change and where the removed material goes. What you receive: A conflict list ordered by lost revenue rather than lost sessions. A decision per conflict, signed. A redirect map for the merges in a format your engineers can import without retyping. And the internal link rewrites each merge forces, because the links still pointing at a URL you are about to retire are the part that gets forgotten and the reason merges look ineffective for months afterwards. What we leave alone: Pages that share a query but sit at different points in the funnel and both convert. Paginated series. Language variants correctly separated by hreflang, unless the hreflang itself is misconfigured, in which case we flag it and hand it to your technical team rather than fixing it inside this engagement. Brand queries where several of your URLs appearing together is normal and desirable. Not included: Executing redirects or editing anything in your CMS. Writing the merged page — that is rewriting work, and we will quote it once you know how many merges you are actually approving. Crawler licences. Not the right engagement if: Your site runs under roughly three hundred URLs. At that size a person can hold the inventory in their head, and this becomes an expensive way to confirm what you already suspect. Skip it if the property has under twelve months of Search Console history, because the week-to-week alternation is the evidence and without a year of weeks there is no pattern to read. And if what you want is a tool export with a conflict column beside each query, any of the standard crawlers will sell you that for a few hundred dollars a year, and you should buy it instead of this.