Your store search now understands what customers really mean.
With our plugin, visitors find products faster, even if they make typos or don’t use the exact “right” term. This results in fewer blank search pages, better user experiences, and ultimately more sales.
What customers notice:
- Relevant results despite typos
- Helpful suggestions as they type
- Better results even with “difficult” search terms
- Significantly fewer “No search results found” messages
Why this matters:
Every unsuccessful search costs trust—and often the sale. Our intelligent search picks up where traditional shop searches fail—and turns search queries back into real buying opportunities.
In short:
More relevant results. Less frustration. More conversions.
Instead of rigidly searching for exact matches, it tries to better match the search intent. That means:
- It handles minor typos. For example, “Kleterpflanse” instead of “Kletterpflanze”
- It understands alternative terms better. A search for “Gartenliege” also finds a sun lounger. And a garden chair, too.
- It often shows relevant suggestions as you type.
- It finds results even if the initial search returns nothing, by trying meaningful alternatives. Searching for “Lavendula” also finds lavender.
Why is this better than the standard store search?
The standard search is often too strict: a single different word or a typo can result in “no results.” Our plugin processes product data specifically for search purposes and can therefore respond faster and more intelligently. Customers are more likely to get a relevant result instead of a blank page.
Technical Implementation
The search operates in stages:
Synonyms
Search terms are supplemented with known alternatives (e.g., laptop <-> notebook).
Fuzzy
If nothing is found, terms are checked for similar spellings (Levenshtein-based) to account for typos.
LLM Rewrite (Mistral)
If there are still no hits, the plugin queries the European artificial intelligence Mistral for alternative search terms.
The response is filtered (only useful single terms, no long phrases).
These terms are searched against the store index locally once again.
Successful alternatives are cached so that subsequent identical search queries are faster.
Additionally:
Autocomplete uses common terms from the search database.
The actual set of results is determined against a pre-prepared search database, not through numerous heavy live queries.
FAQ:
Do I need to configure any special settings on my server?
No additional search servers such as Elasticsearch, Meilisearch, or Typesense are required for operation. The plugin runs directly in your modified shop using PHP and MySQL/MariaDB.
Technical Requirements ?
- Modified E-Commerce (current 3.x versions)
- PHP 8.2 or newer
- MySQL/MariaDB with InnoDB and FULLTEXT support
- Support for cron jobs (included with most hosting providers, but it’s best to check)
- Write permissions to the shop’s /log folder
- SSL connection (i.e., shop accessible via https instead of just http)
- For full functionality (integration with the AI “Mistral”): Register for an API key at Mistral.ai
- PHP-cURL enabled on the server for Mistral integration (check with your host)
Important:
Even without Mistral, the plugin already works with intelligent search (synonyms, typo tolerance, suggestions).
With Mistral, the search is further enhanced for difficult terms. This incurs costs, which, based on experience, are in the range of a few cents.
Why don't you just use Elasticsearch?
Because our lawyer says that Elasticsearch illegally transfers data to the U.S.
What are these strange new database tables?
ai_search_index
Central search database per product (for German, language_id = 2).
Contains the processed search text and key filter fields (e.g., category, manufacturer, price, stock status).
The actual product search runs on this database.
ai_search_synonyms
Contains synonym pairs.
Used to expand search terms before the search is performed.
ai_search_suggest
Contains frequent search terms with weighting.
Used for suggestions while typing and as a basis for fuzzy correction.
ai_search_rewrites
Cache for alternative terms from the Mistral fallback.
Prevents repeated API calls for identical queries and speeds up follow-up queries. And saves costs.
Deutsch
English