How does AI Commerce search work?
You will learn search logic and weighting to make products and categories easier to find. Includes Algolia and alternate search descriptions.
Written by Petro Mäntylä
Updated at October 19th, 2026
How does AI Commerce 's search work with two search engines?
AI Commerce 's search works with two search engines: it uses either Algolia integration or a built-in database search. Algolia is selected when both ALGOLIA_APP_ID and ALGOLIA_API_KEY are defined for the store. If either ID is missing, the search automatically uses a database search. In both ways, results are generated based on the current store language.
The selection is made before the search. If the specified Algolia search fails at runtime, the system will not switch to a database search in the same request, but will return an empty result and an error indicating that the search was temporarily interrupted. This distinction is important for troubleshooting: the database search is an alternative to missing Algolia tokens, not a hidden fallback for an Algolia runtime error.
In Algolia search, a maximum of 10 products and a maximum of 10 categories are requested; a maximum of 5 categories are hydrated from the local database. The database search limits the visible product list to the top 10 and the category matches to the top 5.
Improve product search discoverability
To improve discoverability in AI Commerce product search in the Products view, fill in the customer-visible product name in each language. Keep the name short and informative, and place the most important words, such as brand and model, at the beginning. Also fill in the product SKU/model, as the database search heavily emphasizes the name and model code.
Add model codes and barcodes for variations so that direct code searches can find the correct product version. Fill in the Short Description and Full Description in each language; they help with indirect searches with less weight. Also, ensure the manufacturer name and maintain language-specific tags for topics, synonyms, campaigns, and seasons in the Product Tags view. Field changes in the dashboard are saved immediately.
Improve category search discoverability
AI Commerce category search discoverability is improved in the Categories view by filling in a clear category name and URL part in each language used. The database search places a strong emphasis on the category name and URL part, so they should describe the concept used by the customer without unnecessary internal vocabulary.
Also fill in the category top description, i.e. the description that appears above the product listing. The weight of the description text is smaller than the name, but it helps with broader and indirect search terms. Keep the names, URL parts and descriptions of different language versions consistent in terms of content. Field changes in the control panel are saved immediately.
Manage Algolia settings
AI Commerce -Algolia search is enabled in the Integrations section of the Configurations view by specifying the Algolia application ID and API key. The application ID's display name is Algolia application ID , and the API key is the secret value. When both IDs are enabled, the store's customer view search will select Algolia instead of the database search.
Algolia indexes must be up to date with respect to the store's products, categories, languages, tags, manufacturers, and variation codes. If the IDs are defined but the search returns a temporary error, check the Algolia connection, indexes, and IDs; a runtime error does not trigger an automatic SQL fallback search.
Database search product weights
Product hits in AI Commerce database search are scored based on field and search method. LIKE stands for partial string match, FULLTEXT BOOLEAN for term-based boolean search, and FULLTEXT NATURAL LANGUAGE for natural language fulltext search. Hits for the same product in different fields are accumulated as relevance points.
| Field | Search method | Weight | Use |
|---|---|---|---|
Product name ( pd.products_name ) |
LIKE |
400 | precise and partial language-specific name search |
Product model/SKU ( p.products_model ) |
LIKE |
400 | model numbers and internal codes |
Variation model code ( products_stock_model ) |
LIKE |
400 | variation-specific codes |
Variation barcode ( products_stock_barcode ) |
LIKE |
400 | barcode and UPC search |
Product tags ( products_tags.tag ) |
LIKE |
50 | language-specific topics and synonyms |
Manufacturer name ( m.manufacturers_name ) |
FULLTEXT BOOLEAN |
100 | brand search |
Product name ( pd.products_name ) |
FULLTEXT BOOLEAN |
70 | name search based on term forms |
| Long and short description of the product | FULLTEXT NATURAL LANGUAGE |
10 | descriptive and indirect terms |
Database search category weights
Category hits in AI Commerce database search are scored from three groups of fields. The category name and URL part receive a weight of 200 in a boolean full-text search, the category name also receives a weight of 200 in a LIKE search, and the category top description receives a weight of 10 in a natural language search.
| Field | Search method | Weight | Use |
|---|---|---|---|
Category name and URL ( categories_name , categories_url ) |
FULLTEXT BOOLEAN |
200 | titles and URL parts |
Category name ( categories_name ) |
LIKE |
200 | partial name matches |
Category top description ( categories_description_top ) |
FULLTEXT NATURAL LANGUAGE |
10 | broader descriptive terms |
In category search, the language is limited to the current trading language, the hits are arranged by relevance score, and the database search returns a maximum of 5 categories. A clear name is therefore more often decisive than a long description.
Algolia search processing
AI Commerce -Algolia search sends a single multi-query request to the indexes <APP_ALIAS>_products and <APP_ALIAS>_categories . Both queries are filtered with languageId = nykyinen kieli , and only the matching product and category IDs are requested from Algolia. A maximum of 10 products and categories are requested for each.
The visible product and category data is hydrated based on ID lists from the local database in the current language. The maximum visible number of products is 10 and categories is 5. Algolia's own index configuration determines its internal relevance; the SQL weights 400, 200, 100, 70, 50, and 10 described here only apply to the built-in database search.
Handling built-in database search
AI Commerce built-in database search process starts when no Algolia IDs are specified. The search text is parsed, blanks and parts containing only - or + are removed, and up to the first three significant terms are used. The search is performed as quick-search type with the LIKE , FULLTEXT BOOLEAN and FULLTEXT NATURAL LANGUAGE conditions described above.
The database search calculates relevance scores for products and categories as the sum of field-specific weights and arranges the hits in descending order. A product query is prepared for a maximum of 20 candidates, but the product list returned to the customer is truncated to the top 10. A category query returns a maximum of 5 hits.
Test the search with customer terms
Testing AI Commerce search with customer terms covers names, models, SKUs, barcodes, brands, and synonyms in each store language. Specifically test combinations of the top three terms, as the built-in database search uses up to the first three significant words.
If the result does not come up as expected, first check the product name, SKU, variation model code, barcode, manufacturer, tags and language. Check the name, URL part and top description for categories. In the Algolia store, also ensure that the index is up to date. Thus, precise codes and clear names increase direct hits, while tags and descriptions support broader searches.