Semantic analysis of a website before its migration: the method

A migration is when you decide what the new website will contain. Carrying the old website over page by page repeats its flaws; rewriting everything loses what worked. A semantic analysis of the website, carried out beforehand, settles the matter: it shows how the current content shares out topics and queries, which pages compete with each other, and what structure to give the new website.

This article describes the method. Matching the pages of the old and the new website by their content, once the redesign is built, is covered on our page about semantic similarity between the old and the new website.

The semantic analysis of a page studies its vocabulary and compares it with what the targeted queries expect. At the scale of a website, it answers other questions: which topics the website covers, with which pages, on which queries each one is visible, and how these pages link to each other. Before a migration, this overall view is what matters.

Start from the complete list of pages (a crawl of the old website) and match it with Search Console data over twelve to sixteen months: for each page, the queries it appears on, its impressions, clicks and average position. A page with no impressions at all has no intent recognised by Google; a page visible on queries it does not cover is poorly targeted. To read this data page by page, see analysing content by its rankings.

Put each page in a topic, then at a level within that topic: the page covering the subject as a whole, and those covering one aspect of it. Note the lexical field of each topic: the terms, entities and questions used by the best-ranked pages. This grouping, often called a semantic cocoon, gives the target structure of the new website and reveals the subjects the old one did not cover.

Two pages that appear on the same queries, or whose content is too similar, split their visibility instead of adding it up: this is cannibalisation. In Search Console, filter an important query and see how many pages of the website receive impressions; if several alternate, none of them stands out. Also compare the contents with each other to find near-duplicates; see checking for duplicate content.

• Keep a page that has its own intent and visibility.
• Differentiate two legitimate pages on neighbouring subjects: give each its own angle, title, H1 and link anchors that state the difference.
• Merge a page with no intent of its own and no visibility into the one that covers the subject best, keeping what is unique about it, then redirect it with a 301.
• Delete a page with no value, no traffic and no links, and make it return a 410.

Do not merge as a matter of principle: covering a subject over several pages is legitimate as long as each one answers a distinct question. The case of websites combined into one is detailed in our article on migrating websites that merge.

Link the pages of a topic to each other and to the page heading it, with anchors that name the angle of the target page. Internal linking spreads popularity across pages (see internal PageRank) and tells Google which page is the reference on each subject; when two addresses must coexist for the same content, the canonical tag points to the right one.

Search Console gives the queries of each page; a crawler, the structure and internal links; Semrush or Ahrefs, competitors' keywords and rankings; semantic optimisation tools (1.fr, YourTextGuru, Visiblis), the lexical field expected for a query. To carry out the inventory itself step by step, see our content audit method.

The decisions of the analysis become redirects: a merged page leads to the one absorbing it, a moved page to its new address, a deleted page returns a 410. Once the staging site is built and the content migrated, Swan matches each URL of the old website with the page of the new one closest to it in content: see how Swan calculates semantic similarity between the old and the new website, then our guide to the redirect plan.