Everything you'd want to know before you install.
A detailed look at how AI Semantic Search — PrestaShop works, why we built it the way we did, and the thinking behind the features above.
§ 01
Internal search is the most overlooked conversion lever
A visitor who uses the search bar has strong purchase intent. Yet PrestaShop's native search relies on keyword matching: it fails as soon as the customer uses a synonym, makes a typo or describes what they want in their own words. Every “no results” is a lost sale and a customer who leaves for a competitor.
§ 02
How semantic search works
The module turns each product (name, brand, categories, features, description) into a numerical vector called an embedding, using an AI model. At search time, the customer's query is converted into a vector the same way, then compared against the whole catalog by cosine similarity. The result: products are ranked by closeness of meaning, not by literal presence of words. “Shoes for running in the rain” surfaces your waterproof running shoes, even if that exact phrase appears nowhere.
§ 03
The autocomplete and the results page, together
Most search modules stop at suggestions. This one also takes over the results page through the productSearchProvider hook, PrestaShop's official mechanism used by faceted navigation. Semantic ranking therefore applies to the whole experience: what the customer sees in the dropdown is exactly what they find on the page, with native pagination and sorting. If the embedding API is unavailable, the module silently falls back to native results: search never breaks.
§ 04
Hybrid mode: meaning without ever missing the obvious
Purely semantic search can sometimes under-rank an exact product-name match. Hybrid mode fixes this by placing the semantic ranking first and appending the native keyword results it did not surface. You get the relevance of AI plus the reliability of classic search, and never fewer results than before.
§ 05
A second module for free: semantic cross-sell
The vectors already computed also serve the product page: a “You may also like” block suggests the closest products by meaning, with a bonus for the same category and the same brand. The computation is local, without API calls, cached for 24 hours, and the rendering uses your theme's native miniatures, flags, wishlist and quick view included. The analytics dashboard completes the picture: top queries, zero-result queries (a goldmine to enrich your catalog), daily volume and CSV export.
§ 06
Control your costs and your privacy
Choose your embedding provider: OpenAI, Mistral AI for European GDPR-compliant hosting, or any OpenAI-compatible gateway. The query cache avoids re-billing repeated searches, and checksum-based re-indexing ensures a price or stock update triggers no unnecessary API call. You keep full control of the model, the dimensions and the budget.
§ 07
Built for production
Batch indexing with a progress bar, scheduled indexing via a secured cron, automatic cleanup of disabled products, native theme autocomplete disabled to avoid double dropdowns, query logging with automatic pruning after 365 days. Multistore and multilingual compatible, without Composer or jQuery, following PrestaShop 8 and 9 conventions.
There are no reviews yet.