# AI Semantic Audit — Vector clustering for PrestaShop 8 & 9

> Detect "off-topic" pages that dilute your store's thematic authority. The module turns each product, category and CMS page into a semantic vector, groups them into thematic clusters (k-means cosine), then identifies misplaced content and suggests where to move it to strengthen site coherence.

- Page: <https://www.datafirefly.com/en/product/dfsemanticaudit/>
- Language: en
- Last updated: 2026-06-28
- Platform: PrestaShop
- Category: Artificial Intelligence
- SKU: DF-SEMAUDIT
- Price: 29.00 EUR
- Other languages: [fr](https://www.datafirefly.com/product/dfsemanticaudit/index.md), [es](https://www.datafirefly.com/es/producto/dfsemanticaudit/index.md), [de](https://www.datafirefly.com/de/produkt/dfsemanticaudit/index.md), [it](https://www.datafirefly.com/it/product/dfsemanticaudit/index.md), [pl](https://www.datafirefly.com/pl/produkt/dfsemanticaudit/index.md), [nl](https://www.datafirefly.com/nl/product/dfsemanticaudit/index.md), [pt](https://www.datafirefly.com/pt/produto/auditoria-semantica-ia-prestashop/index.md)
- Index: <https://www.datafirefly.com/en/product/llms.txt>

A high-performing PrestaShop store is a **thematically coherent** store. When your content is scattered across too many topics, Google can't tell what you should rank for and your authority dilutes. This module audits your catalog using the same techniques as search engines: vector embeddings, unsupervised clustering, outlier detection.

The module indexes all your products, categories, CMS pages and manufacturers, turns their titles and descriptions into semantic vectors (1,536 dimensions with OpenAI, or local without any API), then groups them into _thematic clusters_. For each piece of content, it computes the cosine distance to its cluster — above the threshold, the content is flagged as "off-topic". The module then suggests the second-best cluster and the expected coherence gain if you were to move the content.

Three embedding providers to choose from: **OpenAI** (text-embedding-3-small, top quality), **Mistral** (mistral-embed, European alternative), or **local TF-IDF** (no API, no recurring cost). Everything runs headless via a signed cron URL — ideal for an automated weekly audit.

The 2D semantic map shows you at a glance the structure of your catalog: each dot is a piece of content, each color a thematic cluster, red outlines mark off-topic pages. You can literally see which parts of your store are pulling in different directions.