What this check inspects
Shipwork reads the title and meta description of a set of sitemap pages, embeds them, and clusters the vectors by cosine similarity. It reports each cluster of pages that are about the same thing, with the peak similarity between them, and says when the site is clear.
What a failure means
A cluster of pages above the similarity threshold is a warning. It does not mean the pages are identical, only that their titles and descriptions describe the same subject, which is enough to set them against each other.
How to fix it
- Decide which page should win for the topic and consolidate the others into it, or differentiate them clearly.
- Give each page a title and description that states what makes it distinct, usually the attribute that differs.
- Link the related pages to each other so a crawler understands the relationship instead of seeing two rivals.
- Watch for template-generated descriptions. If every collection page gets the store tagline, they will all cluster.
A typical failure, worked through
The setupA store has four collection pages for colour variants: red running shoes, blue running shoes, black running shoes and running shoes. All four use the same description template with only the colour substituted.
What the check reportsThe check groups the four pages as near-identical because their embedded text is almost the same, and reports the cluster with the similarity percentage.
The pointA template with one word swapped is not differentiation. Either merge them or write descriptions that describe the actual difference.