# Search Settings

Tune how your store's product search ranks and matches results without any
code changes. Everything below lives on the dashboard's **Search > Settings**
page and applies to both the `<product-search>` widget on your storefront and
the search playground in the dashboard.

Two kinds of settings:

- **Instant**: field weights and synonyms take effect on the very next
  search.
- **Needs a reindex**: Latin/Cyrillic matching and AI search change how your
  catalog is indexed. After changing them a banner appears; search keeps
  working with your previous settings until you click *Reindex now* (indexing
  time depends on catalog size, and you can keep using the store meanwhile).

## Field weights

Decide how much a match in each field counts: **Title**, **Categories**,
**Description** (each 0 to 127; higher means more important).

With the defaults (title 4, categories 2, description 1) a product whose
*title* contains the search term always ranks above one that only mentions it
in the *description*. Raise or lower the numbers to taste, or set a field to 0
to ignore it entirely. Saved instantly.

## Synonyms

Make different words find the same products. Two types:

- **Two-way**: all terms match each other. Example: `ваза, vase, кана`.
  Searching any of the three finds products containing any other.
- **One-way**: searching the term also finds its synonyms, but not the
  reverse. Example: the search term `съд` also finds `ваза` (but searching
  "ваза" won't return every съд).

Synonyms apply immediately, no reindex needed. Terms are case-insensitive.
Up to 200 rules per store.

## Latin ↔ Cyrillic matching

Shoppers often type Cyrillic words with Latin letters, "vaza" instead of
„ваза". Turn this on and the search index stores a transliterated copy of
every title and description, so Latin-typed queries match your Cyrillic
product names (and matches in the original script still rank first).

Requires a reindex after toggling.

## AI search (semantic)

Matches what shoppers *mean* as well as the words they type: "flower
holder" can find vases, „подарък за готвач" can find kitchen accessories.
Results blend classic keyword matching with AI meaning-matching; the
**AI influence** slider (0 to 1) controls the blend (default 0.3, keyword first).

AI search uses OpenAI embeddings **through your own OpenAI API key**:

1. Create a key at <https://platform.openai.com/api-keys>.
2. Paste it on the Settings page. It is verified with OpenAI on save,
   stored server-side, and never shown again (only its last 4 characters).
3. Enable the toggle and run the reindex.

Embedding costs are billed by OpenAI to *your* OpenAI account, and they are
small: the model is `text-embedding-3-small` (fractions of a cent per thousand
products). A full reindex embeds the whole catalog once; after that only new
or changed products are embedded, and each search embeds just the query, a
few characters of text.

Enabling, disabling, or changing the key requires a reindex. If your key is
ever revoked, keyword search keeps working. Only the AI matching pauses, and
the error is shown on the Search Overview page.

## Search-by-photo

Photo search (the camera button in the widget) is always on and needs no
configuration, because your product images are indexed automatically. See
[widget-embed](widget-embed.md) for enabling the camera button.
