Cannibalization Fix: Before and After Data

Keyword cannibalization is easy to explain and strangely hard to believe until you watch a fix land — so here's one watched closely: a site whose three overlapping articles had spent a year trading page-two positions for a commercial query none of them ever won, consolidated per the playbook, with the before-and-after data that makes the case better than theory ever does.

The patient: the flip-flop signature

Three posts, written across two years by different authors, all targeting variants of the same buying-intent query (call it "[category] software comparison"): overlapping coverage, split internal links, and the diagnostic tell in Search Console — the ranking URL for the query changed week to week (Google re-picking its representative among near-duplicates), positions oscillating 8–14, none stabilising, combined CTR miserable because the "winning" page kept changing under searchers. URL-level link equity: split 7/4/3 referring domains across the trio — a divided vote that, combined, would have been the query's third-strongest page. Twelve months of this cost quantified: the query's estimated clicks going almost entirely to competitors while three pages did the work of none.

The fix, step by step

1. Choose the survivor: the URL with the most links, best engagement history and cleanest slug won — the other two graded merge-donors, per the playbook's rule (strongest URL inherits; sentiment doesn't vote).
2. Merge for coverage, not concatenation: the survivor rebuilt to absorb the donors' genuinely unique material (one's comparison table, the other's pricing analysis) into a single definitive piece — the coverage outline run against all three plus the live SERP, ~6 hours.
3. 301 the donors to the survivor — exact-relevance redirects, the textbook case for them — and update every internal link sitewide to point at the survivor directly (14 pages edited; redirects are for the outside world, per the hygiene findings).
4. Consolidate the metadata: survivor's title rewritten for the full query family, canonical self-referencing, donors removed from the sitemap.

The after

Weeks 1–3: the familiar consolidation wobble — the survivor briefly dipped as the flip-flopping stopped and re-evaluation ran (panic window; nothing touched). Weeks 4–8: the survivor stabilised at position 6 — better than any of the trio had ever held — with the URL-switching in Search Console gone completely: one page, one query, one history accruing. Month 4: position 4, and the page's other rankings (the long-tail variants each donor had weakly held) consolidated onto it too — total query-family clicks up roughly 150% versus the trio's best combined month. The donors' redirected equity, engagement history unified, and the ended self-competition are inseparable in the data; the point is they arrive as one intervention.

The generalisations

Cannibalization is a stability tax before it's a position tax — the flip-flop itself suppresses CTR and history accumulation; ending it pays even before equity math kicks in. Merges outperform differentiation when intent is genuinely one — the alternative fix (re-aiming the extra pages at distinct intents) is for pages that have distinct intents available; forcing differentiation on one intent just produces politer cannibals. And prevention costs one row in a spreadsheet: all three posts predated the site's keyword map; nothing written after the map's one-keyword-one-URL rule has cannibalised since — the cheapest lesson in the series.

Frequently asked questions

How do I find my own cannibalization quickly?

Search Console, performance report, filter by query, check the pages tab per money query — multiple URLs sharing impressions on one query with unstable splits is the signature. The audit's stage two systematises the sweep; the inventory catches the title-overlap cases before they mature.

Is it always a merge? I'm afraid of deleting content.

The four resolutions (merge, differentiate, canonicalise, prune) map to diagnosis, per the playbook — merge when intent is one, differentiate when intents genuinely diverge. And a merge deletes nothing readers valued: the donors' substance lives in the survivor, their URLs redirect, their equity transfers, per the redirect findings. What dies is the self-competition.

The dip in weeks 1–3 — how do I not panic?

Annotate the change date, pre-commit to a six-week no-touch window, and watch the URL-switching metric rather than position (its disappearance is the fix working before positions show it). Consolidation wobble resolving upward is the pattern across every merge we've run — and the position it resolves to still obeys the standing arithmetic: coverage sets eligibility, authority sets rank (our department).