Definition
Keyword cannibalisation is when two or more pages on the same site compete for the same search term, so search engines have to choose between them. The usual symptom is that the ranking URL changes over time and neither page performs as well as one page would.

Why it matters
Cannibalisation is a self-inflicted problem and therefore a cheap one to fix. No outreach, no new content, just a decision about which page owns which search.
It accumulates quietly on sites that publish regularly. Three articles written eighteen months apart about overlapping subjects do not announce themselves, and the symptom, a ranking that drifts between URLs, looks like volatility rather than a structural fault.
How it works
Google selects one page per site for most queries, which follows from how its ranking systems choose a single best result. When two pages are both plausible answers, the choice can change between crawls. Neither accumulates the signals a single clear answer would, and the position tends to sit lower than either page could reach alone.
It is worth being precise about what is not cannibalisation. Two pages mentioning the same phrase is normal. Two pages ranking for overlapping long-tail variants is normal and often desirable. The problem exists when two pages target the same search intent, meaning a searcher would be equally satisfied by either, and only one can be shown.
Worked example
Confirm it before fixing it. Search Console's performance report, filtered to one query, shows which pages received impressions for it:
| Page for "seo audit checklist" | Impressions | Position |
|---|---|---|
| /blog/seo-audit-checklist | 420 | 14 |
| /blog/how-to-audit-your-site | 310 | 17 |
| /seo/audit | 90 | 22 |

Three URLs splitting one query is the signature. One page should own it. The others either serve a different intent, or should stop trying.
Four resolutions, in order of how often they are right:
- MergeCombine into the strongest URL, 301 the others to it
- RefocusRewrite the weaker page for a genuinely different intent
- DemoteKeep both, but remove the internal links and title language that made the second one compete
- CanonicalWhere both must exist for users, name the primary one
Common mistakes
- Diagnosing it from a keyword tool.Overlap in a tool is not evidence. Impressions split across URLs in Search Console is.
- Merging pages that serve different intents.A definition page and a service page targeting related words are not competing. Collapsing them loses one of the two jobs.
- Deleting the loser.If it has links, 301 redirect it so its link equity survives. Deleting discards it.
- Fixing it with canonicals by default.A canonical tag is for duplicate content. Two genuinely different pages competing is usually a merge or a refocus.
- Letting it recur.The root cause is publishing without checking what already exists. A pillar page structure makes the overlap visible before the article is written.
How to prevent it
Decide the target search before writing, and check whether a page already serves it. One page per intent, written down somewhere a writer will see it, prevents more cannibalisation than any amount of auditing afterwards, and it is the cheapest habit in on-page SEO.