When keyword cannibalization is fine (and when it isn't)

The two errors are not symmetric. Leaving real cannibalization in place costs you some position stability on one query, and you can fix it next quarter. Consolidating pages that were not competing destroys a page that was earning, loses whatever long-tail it held, and hands you a redirect chain to maintain forever.

That asymmetry should set your evidence bar. You are not looking for a reason two pages might compete — related pages always might. You are looking for evidence they actually are, and that it is costing something measurable.

1. Two URLs, one clear winner

If page A takes 90% of a query's impressions and page B picks up the rest, page B is harvesting long-tail variants that A does not match exactly. Merging them gains you nothing and loses the variants. This is the single most common false positive.

2. Different intent behind one term

Plenty of queries are ambiguous, and Google knows it. Someone searching a product category might want a definition, a comparison, or a buying page. Having one of each is not competition — it is coverage, and it means whichever intent Google decides to serve, you have something for it.

3. Sitelinks and multi-result listings

Two of your URLs appearing in one SERP is frequently a reward, not a symptom. Sitelinks, a parent page with indented children, or two organic results from one domain all mean the engine considers your site the authority on that query. Consolidating to force a single result throws that away.

4. Brand queries

Your homepage, your pricing page and your about page will all show up for your brand name. This is correct and there is nothing to fix. Brand terms dominate most Search Console exports and are the main reason a naive cannibalization audit returns noise — filter them out before you start.

5. Deliberate funnel stages

An informational guide and a commercial page on the same subject serve different stages, and the informational one usually exists to attract readers the commercial one never would. If the guide links well to the page it supports, it is doing its job — that is a content cluster, and clusters are built out of pages that share vocabulary on purpose.

6. Brief overlap after publishing

Google tests new pages. A recently published article may take impressions from an established one for a few weeks and then settle. Diagnosing cannibalization inside that window and consolidating is acting on a transient. Give a new page a quarter before judging it.

Against that, three patterns are worth acting on, and all three are about instability rather than overlap.

1. URL swapping between periods. Compare two 28-day windows in Search Console; if the page Google serves for a query keeps changing while your content has not, the engine cannot settle and neither page accumulates click-through history.
2. Two URLs mid-page, neither breaking through. A 50/50 impression split with both sitting around position 12 is the classic split-signal case — one combined page would likely have cleared the fold.
3. Identical intent, identical audience, identical job. Three “best X tools” posts written eighteen months apart with no editorial difference between them. Nothing is being covered here that one maintained page would not cover better.

  evidence                              act?
  ------------------------------------  -----------------------
  two URLs, stable, one dominant        no
  two URLs, different intents           no
  two URLs in one SERP (sitelinks)      no
  brand query                           no
  new page, under a quarter old         wait
  URL swapping between periods          yes
  50/50 split, both mid-page            yes
  same intent, same audience, no diff   yes

1. Filter Search Console to the exact query and read the Pages tab. Note the impression share per URL.
2. Compare the last 28 days with the previous 28. Look for the serving URL changing, not for two URLs existing.
3. Ask what job each page does. If you cannot state a distinct reader and outcome for each, that is the case for merging — and if you can, that is the case against.
4. Check clicks, not just impressions. A page with impressions and no clicks may be badly matched rather than competing.
5. If you still merge, keep the losing page's unique sections. Most of the loss from consolidation is content quietly deleted in the move.

Step 2 is the load-bearing one — finding cannibalization in Search Console walks through the comparison in detail. And before assuming similar-looking pages are the problem, check whether you actually have duplication rather than competition, because they need different fixes.

Between doing nothing and consolidating there is a middle option that carries almost no risk: fix the internal links.

Borderline cases are often manufactured internally. If forty links across your site use the same anchor text for two different destinations, you are telling search engines both pages are about that term, and the engine's confusion is a faithful reflection of yours. Rewriting anchors so each describes its actual target frequently resolves the oscillation without deleting anything.

That ordering — links first, content second, redirects last — matches the reversibility of each step. Only the last one is permanent, so it should need the most evidence.