How to find keyword cannibalization in Google Search Console

Cannibalization is a claim about how search engines respond to a query. Every other diagnostic — content similarity, title overlap, anchor text — is a proxy for that. Search Console is the only free source of the real thing: which of your URLs got impressions for which query, and in what position.

That makes it definitive for confirming a case and weak for discovering one, and most guides conflate the two. The confirmation workflow below takes two minutes. The discovery workflow takes longer and has a ceiling, which is worth knowing before you spend an afternoon on it.

Start here when you already suspect two specific pages compete.

1. Open Performance → Search results and set the date range to the last 3 months.
2. Click + New → Query → Queries containing (or Exact query, if you know the exact term).
3. With that filter applied, switch from the Queries tab to the Pages tab.
4. Read the URL list. One URL means no cannibalization for that query. Two or more with meaningful impressions means you have candidates.

Two URLs appearing for one query is normal. What is not normal is Google being unable to decide between them. That indecision is the cost, and it is visible in Search Console if you compare periods rather than reading one.

1. Keep the query filter applied and open the date picker → Compare.
2. Compare the last 28 days against the previous 28 days.
3. On the Pages tab, read the impression split per URL across the two periods.

  period A        period B        verdict
  -------------   -------------   ------------------------------
  A 95% / B 5%    A 94% / B 6%    stable, one winner - leave it
  A 60% / B 40%   A 35% / B 65%   swapping - real cannibalization
  A 50% / B 50%   A 48% / B 52%   split, both mid-page - costly
  A 100% / B 0%   A 0% / B 100%   a handover, not a competition

The last row catches people out. A clean handover — one URL stops, another starts — usually means you published a replacement and Google switched, which is the system working. Cannibalization is the oscillating middle rows, where neither page can hold the position long enough to accumulate click-through history.

There is a second tell worth checking: average position variance for the query as a whole. A query whose position swings by five or more places month to month while your content has not changed is often being served by different URLs at different times.

The confirmation workflow needs you to already have a hypothesis. To find cannibalization across a whole site you have to invert it: start from queries, not from pages.

Regex query filters

Search Console supports regex in the query filter (+ New → Query → Custom (regex)). This is the most underused feature in the product for this job, because it lets you isolate the query shapes that cannibalize most often.

• chevron_rightbest .* (tool|tools|software) — the classic listicle overlap, where a rewritten post never retired its predecessor.
• chevron_righthow to .* — process content, where a general guide and a specific tutorial drift into the same query.
• chevron_right(vs|versus|alternative) — comparison pages, which multiply quickly and rarely get audited.
• chevron_right^(?!.*brand).* — everything except your brand terms, which otherwise dominate the list and hide the interesting queries.

Apply one, then work down the Queries tab from the top. For each query with real impressions, click it and check the Pages tab. It is manual, but a focused regex turns a list of thousands into a list of dozens.

The export method

For a systematic pass, export rather than click. In Performance, use Export → Google Sheets with no filter applied, which gives you the top queries and top pages as separate sheets — and that is precisely the limitation: the standard export does not give you the query-to-page pairing.

To get the pairing you need the Search Console API (the searchanalytics.query endpoint with dimensions set to both query and page), or a connector that calls it for you, such as Looker Studio's Search Console source with both fields as dimensions. With one row per query-page pair, cannibalization is a group-by: any query with more than one page above an impression floor.

Three blind spots matter enough to change how you audit.

1. Pages with no impressions are invisible. A page that competes but never surfaces contributes nothing to the report, so the most cannibalized pages on a site — the ones so confused that neither ranks — are exactly the ones you cannot see here.
2. It shows the symptom, never the cause. Search Console will tell you two URLs are swapping. It will not tell you they swap because forty internal links use the same anchor text for both, which is often the actual mechanism.
3. It is retrospective. A page published last week has no data. On a growing content library the newest overlaps are the ones you most want to catch before they compound.

All three blind spots have the same shape: they are about pages and links, not queries. Comparing pages against each other — title and heading overlap, body similarity, and whether internal links point at both with the same anchor — finds what the query data structurally cannot. The free Keyword Cannibalization Checker does that pass over a crawl of your site, and the anchor-text angle is the one worth checking first, because anchor text is how you tell search engines what a page is about.

Use both directions. Search Console confirms that a competition is real and costing you something; a structural crawl finds the competitions that have not surfaced yet. Once you have a confirmed case, how to fix keyword cannibalization covers choosing between consolidating, differentiating and leaving it alone — and the internal-link repair that decides whether the fix holds.

1. Run one regex filter per content shape you publish (listicles, how-tos, comparisons). Fifteen minutes each.
2. For every query above your impression floor, check the Pages tab for a second URL.
3. For each candidate pair, run the 28-day comparison. Only oscillating pairs go on the fix list.
4. Separately, crawl the site and check for pages that duplicate each other structurally but have no impressions yet — the pre-symptomatic cases.
5. Re-check fixed pairs one quarter later. The success criterion is that the swapping stopped, which shows up before the position improves.

Quarterly is the right cadence for most sites because cannibalization accumulates with publishing volume rather than with time. If you ship a lot of content in a burst — a migration, a bulk import, an agency handover — audit right after the burst instead of waiting for the calendar.