Rewriting does not fix cannibalisation: one firm, two live domains
The two sites carried different text on the same four topics, so everyone concluded there was no duplicate content problem. They were right. That was never the problem.
Duplicate content and keyword cannibalisation are different failures with different fixes. Duplicate content is text similarity, and rewriting solves it. Cannibalisation is two URLs answering the same query, and rewriting does nothing at all — a search engine still has to choose one. A firm running two live domains with independently rewritten guides had solved the first problem and left the second untouched.
The problem
A law firm operated two live websites: a legacy site on a website builder, and a current site on a modern framework. Both were indexed. Both carried guides on the same four subjects — sponsor licences, refused settlement applications, division of assets on divorce, pension entitlements on divorce — under different titles and different slugs, written at different times, with genuinely different text.
When I raised it, the response was reasonable and, on its own terms, correct:
“Several of those I rewrote completely, and the rest are not identical either — plus they are localised. How is that duplicate content?”
It is not duplicate content. It never was. Understanding why that answer does not resolve the problem is the whole point of this case.
Two failures that look alike
Search engines evaluate two separate things, and the industry habit of calling both “duplicates” hides the distinction.
| Duplicate content | Keyword cannibalisation | |
|---|---|---|
| What matches | The text | The query and the intent behind it |
| Consequence | One version is selected as canonical, the other is filtered from results | Two of your pages compete; ranking signals split between them |
| Does rewriting fix it? | Yes | No — the query is unchanged |
| Actual fix | Canonical tag, or rewrite | Consolidate to one URL and redirect the other |
Rewriting a page changes its words. It does not change what the page is for. If the legacy site’s guide and the current site’s guide both exist to answer what happens when a settlement application is refused for absences, then a search engine facing that query has two candidate pages from the same brand and has to pick one. It will. Which one is not the owner’s decision, and it can change from month to month.
Localisation is not duplication. An English page and its Russian and Lithuanian equivalents target different queries in different languages and are legitimately separate URLs, provided the hreflang cluster is correct. Language versions are an asset. Two English pages on two domains answering one English query are not.
What the pairing looked like
Four topics existed twice in English, once per domain, under different titles and slugs. Because each domain also carried Russian and Lithuanian versions, the practical result was two parallel three-language sets for the same four subjects.
| Topic | Legacy domain | Current domain | Pairs |
|---|---|---|---|
| Sponsor licence and certificate of sponsorship | Market-entry framing, own slug | Employer-obligations framing, own slug | 2 × 3 langs |
| Settlement refused for absences | Appeal-and-review framing | What-to-do-next framing | 2 × 3 langs |
| Division of assets on divorce | 2026 legal guide | Legal rules explained | 2 × 3 langs |
| Pension entitlement on divorce | Question-form headline | Descriptive headline | 2 × 3 langs |
Two of those pairs had diverged far enough that the client remembered the legacy titles as the current ones and the current titles as the legacy ones. That is a useful diagnostic in itself: when the author cannot tell which version is live, no search engine is going to make a confident choice either.
Why a second domain costs more than the pages on it
Cannibalisation across two domains is worse than within one, because the split is not only at query level.
- Links divide. Every citation, directory entry and client link points at one domain or the other. Neither accumulates the full authority the firm has actually earned.
- Brand signals divide. Two domains presenting the same practice look, to an entity-resolution system, like two organisations with similar names.
- Reporting divides. Two Search Console properties, two exports, and no combined view. Nobody can answer “how is the firm doing in search” without manual reconciliation.
- The structured data may point the wrong way. On this site the canonical tag named one domain while the organisation markup named the other — the subject of a separate case study.
How to decide which domain survives
The decision is not aesthetic and should not be made on which site the owner prefers. Score both on the following, then commit — a half-migration is worse than either domain alone.
- Which domain holds the search history
Export both properties and compare impressions and positions per topic. Authority already earned is the most expensive thing to rebuild.
- Which domain the outside world points at
Directories, professional registers, citations, email signatures, printed material. A domain that other people link to is not a domain you can quietly abandon.
- Which domain matches the entity everywhere else
Email addresses, structured data, registration records, social profiles. The domain that the rest of the identity already uses is the cheaper one to standardise on.
- Then map every URL, topic by topic
For each topic, choose the better of the two texts — not the newer one automatically — publish it on the surviving domain, and 301 the losing URL to it. Same-topic, same-language, in both directions across the domains.
- Redirect at page level, never wholesale to the homepage
A blanket redirect to the root discards the topical relevance each page earned. The point of consolidating is to keep it.
Leaving the second domain up “for now” while building on the first is the default and the most expensive option. The split persists for as long as both are indexed, and the longer it runs, the more links and citations accumulate on the domain that is eventually switched off.
Method and limits
Source. A Google Search Console export for one of the two properties covering 5 May 2025 to 31 August 2026, the current sitemap, and direct requests to live pages on both domains on 4 September 2026.
Not established. Search Console data for the second domain was not available at the time of the audit, so the split cannot be quantified from both sides. The pairing itself was confirmed by reading both live sites; the size of the loss was not measured and is not claimed here.
Client identification. The client is described by practice type. Domains, URLs and page titles are paraphrased rather than reproduced.
