AI Semantic Search

Let shoppers describe what they want in plain language — "a warm jacket for winter", "smart casual for the office" — and get products matched by meaning, not just keywords.

What it does

Regular search matches the words a shopper types. Semantic search understands intent: someone searching for "comfy sofa" also sees "comfortable couch", and a query like "warm jacket for winter" surfaces the right coats even when those exact words aren't in the title. It picks up synonyms and context automatically, handles conversational queries, and works across languages (Greek and English in the same catalog).

When AI search is on, shoppers get an Ask AI button next to the search bar that opens a conversational popup with suggestion chips and a "Recommended for you" results carousel.

How it works (in one paragraph)

You bring your own OpenAI API key. SearchX turns each product into an AI embedding — a numeric representation of its meaning — using a short text template you control, so you decide which fields describe each product. Shopper queries are converted the same way and matched by meaning rather than exact words. Your catalog and your OpenAI key stay under your control; SearchX orchestrates the indexing and matching.

What you'll need

• An OpenAI account with an API key (Step 1).
• Products already imported into your SearchX application.
• Access to your storefront integration (theme snippet or plugin) to turn on the AI button (Step 3).

Step 1: Create an OpenAI API key

1. Sign in at platform.openai.com.
2. Make sure billing is set up — embeddings require a paid OpenAI account.
3. Open API keys, create a new secret key, and copy it. You'll paste it into SearchX once; it's stored encrypted and never shown again.

Step 2: Enable AI Semantic Search in the dashboard

Open your application, go to Settings → AI Search, and turn on Enable AI Semantic Search.

Fill in:

• Provider — OpenAI (the only provider right now).
• API Key — paste your OpenAI key. On later edits, leave it blank to keep the saved key.
• Model — text-embedding-3-large is recommended: highest quality and the best multilingual / Greek results. text-embedding-3-small is a cheaper option for simple English-only catalogs; text-embedding-ada-002 is a legacy model.
• Document Template — the text that gets embedded for each product, written in Liquid. This is where you decide what "describes" a product for AI matching.

A solid starting point for a fashion catalog — copy it as a single line:

{{doc.brand}} {{doc.title}}{% if doc.category_display %} in the {{doc.category_display}} category{% endif %}. {% if doc.description %}Product Description: {{doc.description}}{% endif %} Attributes: Brand: {{doc.brand}}{% if doc.mpn %}, MPN: {{doc.mpn}}{% endif %}{% if doc.color %}, Color: {{doc.color}}{% endif %}. {% if doc.sale_price %}Sale Price: {{doc.sale_price}} Euros (originally {{doc.price}}, on discount).{% elsif doc.price %}Price: {{doc.price}} Euros.{% endif %}

Fields you can reference include doc.title, doc.brand, doc.description, doc.category_display, doc.color, doc.sizes, doc.mpn, doc.price, and doc.sale_price. The Vectorization Preview under the template renders this against a real product from your catalog, so you can fine-tune the wording before saving.

When you Save and confirm, SearchX re-indexes your catalog and generates an embedding for every product. This runs in the background and can take a few minutes for large catalogs.

Step 3: Turn on the AI button in your storefront

Enabling embeddings in the dashboard makes semantic search available; the storefront widget still needs to be told to show the Ask AI button. There are two switches, and you need both:

How to set the storefront switch:

• Custom snippet / SDK (HTML, Shopify, Next.js, React, OpenCart): add aiSearchEnabled: true to the settings object in your SearchXSDK.init({ ... }) call.
• WooCommerce plugin: tick AI Search under WooCommerce → Settings → SearchX.

SearchXSDK.init({
  // ... app_id, api_key, components ...
  settings: {
    aiSearchEnabled: true,
    // ... other settings ...
  },
});

If the button doesn't appear, the storefront switch is off. If the button appears but returns nothing, the dashboard feature isn't enabled yet (or the re-index hasn't finished).

AI Search Chips

Chips are the clickable starting points shown when a shopper first opens the AI popup — for example "Comfortable walking shoes" or "Smart casual for the office". They nudge shoppers toward queries that work well.

Manage them in the dashboard under Settings → AI Search → AI Search Chips:

• Add up to 10 chips, each ≤ 50 characters.
• Drag to reorder — the order you set is the order shoppers see.
• Each chip has English and Greek translations (edit both in the chip's translation dialog). The widget shows the chip in the shopper's language automatically and falls back to English when a translation is missing.

Chips are delivered to the widget automatically — you don't need to change your snippet when you add or edit them.

Preview before going live

You can try the AI experience in the Playground (and the in-app preview): open the preview's settings panel and toggle Enable AI Search? to show the Ask AI button and popup without touching your live storefront. To get real semantic results in the preview, the application must already have AI Semantic Search enabled and the re-index finished.

See Playground for how to share a preview link.

Styling the AI UI

Every AI surface — the Ask AI button, the popup, the suggestion chips, the recommended-products carousel — has a stable CSS class you can target from your custom theme. See the AI Search selectors in Widget Styling.

Troubleshooting