We ask first.

This website works without cookies. If you agree, we will also use cookies for measuring visits and our advertising. You can change your mind at any time from “Cookie settings” at the bottom of every page. Cookie policy

A limestone counter in daylight, with a shop's products set out like search results.

ServicesOnline shops & B2B portals

Shoppers type the way they talk. Search lifts what they mean.

AI search for online shops

Shoppers type the way they talk. Search lifts what they mean.

Shoppers do not know what you call the product. They type “gift for a godmother under €50”, “something for an evening wedding on the beach”, “olive oil for salad, not bitter”. A shop's classic search looks for letters and often finds nothing. Semantic search looks for meaning. Here is how it works, when it pays off, and why it is decided by your catalogue first.

Tell us about yours
54%
of the large online shops studied struggle when shoppers search with abbreviations or symbols; 43% when they search by use (“for running”), 39% by a feature (“waterproof”).Baymard Institute, 2026
3
fields are all WooCommerce's built-in search reads: the product's title, content and excerpt. Colour, size or use are not “understood”.WordPress, WP_Query
Greek
is among the languages of Shopify's store search and Google's AI Commerce Search. Whether the semantic feature covers your language is something to test in your own store.Shopify, Google Cloud
1
language per catalogue is what Google recommends: a query in another language does not match. A bilingual shop needs two indexes, not one mixed one.Google Cloud

Last updated:

Classic search looks for letters. Semantic search looks for what you mean.

Keyword search compares the letters of the query with the letters of the catalogue. If a shopper types “blow dryer” and you call it a “hair dryer”, nothing comes back, however good the product is.

Semantic search turns every product and every query into a numerical representation of its meaning (embeddings, vectors) and returns what is close in meaning, not in spelling. So “something for an evening wedding on the beach” can bring up light dresses and flat sandals, although none of those words are in their titles.

In practice good systems combine the two (hybrid search): keywords for codes, brands and exact names, meaning for everything else. A spare part's code must be found exactly, not “something similar”.

  • Keywords

    Exact and fast for codes, brands and model names. They fail when the shopper uses a different word from yours.

  • Semantic

    Understands synonyms, uses, descriptions and phrases as people say them. On its own it can return “something similar” where something exact was asked for.

  • Hybrid

    Runs both and merges the results. It is what most search providers recommend today.

Shoppers search by use, not by the name of the shelf.

Baymard Institute's research on large US and European shops shows where search stumbles: less on the product's name (20% of sites have issues there) than on its use, its features, the problem the shopper wants solved, and abbreviations.

Other languages add their own obstacles. In Greek, for example: accents and word endings, words typed in Latin letters (“greeklish”), and catalogues where half the names are English. A search that cannot handle these loses shoppers before it ever reaches meaning.

  • Use

    “Shoes for standing all day”, “a pan for induction hobs”. The shopper's word is rarely in the title.

  • Feature

    “Waterproof”, “gluten-free”, “organic”. It exists in the catalogue only if it is written down as an attribute, not hidden in a photo.

  • Problem

    “For dry skin”, “non-slip”. Shoppers describe what they want solved, not what they want to buy.

  • Spelling

    Accents, plurals, transliteration, typos, English and local names for the same thing.

No search finds what was never written down.

Semantic search understands meaning, but only the meaning that is in the catalogue. A product titled “DRS-2291 BLUE” with no description will not be found by any model for “summer dress for a wedding”.

So the work starts with the data: attributes in fields (colour, material, size, use), descriptions that say who the product is for and what for, and prices and stock that stay current, so that search does not put something sold out first. The same clean catalogue also feeds Google Shopping, marketplaces and AI assistants.

  • Attributes in fields

    Colour, material, size, use, certifications: in their own fields, so they also become filters.

  • Descriptions for people

    Who it suits, when it is used, what sets it apart. This is what the model reads to understand meaning.

  • Live price and stock

    From the ERP or warehouse system, so what is available comes first.

  • One language per index

    A bilingual shop's languages in separate indexes, as Google also recommends.

What each platform offers today.

The choice depends on the shop's platform, the size of the catalogue and how much control you want over the order of results. A new system is not always needed; sometimes it is enough to switch on what exists and fix the catalogue.

  • Shopify

    Store search supports many languages, Greek included (plurals, typos). Semantic Search is switched on in the Search & Discovery app, on the plans above Basic, and in multiple languages since September 2024. Test it in your own store with real phrases in your language.

  • WooCommerce

    WordPress's built-in search looks for words in the title, content and excerpt. Semantic or hybrid search needs a plugin or an external search service.

  • Algolia

    NeuralSearch runs keyword and vector retrieval and merges the results. It learns from clicks and purchases, so it needs measurement from day one, and it is not included in every plan.

  • Elasticsearch and OpenSearch

    Open engines with multilingual models for semantic search and hybrid queries. They give the most control and need a team to set them up and maintain them.

  • Google AI Commerce Search

    Google Cloud's search for shops (formerly Vertex AI Search for commerce) supports Greek and many other languages. Google recommends one catalogue per language; built for large catalogues with heavy traffic.

Searches with no results are the list of what is missing.

Before you change engine, measure the one you have. The most useful report is the simplest: which phrases shoppers searched for and found nothing. It shows which words are missing from the catalogue, which products people ask for that you do not stock, and where search returns the wrong things.

After the change, the same numbers show whether it worked: fewer searches with no results, more clicks on the first results, more purchases from people who searched.

  • No results

    The phrases that returned nothing, with how often they were searched.

  • Exits after search

    How many searched, saw the results and left.

  • Purchases from search

    The share of searchers who bought, compared with visitors who did not search.

Before you change your shop's search.

In the order that matters. The first three steps improve search whatever engine you have.

  1. Export the phrases shoppers searched for in the last three months and pick out those with no results.

  2. Search twenty of them yourself, with and without accents, transliterated and with a typo, and note what comes back.

  3. Fill in product attributes in fields (colour, material, size, use) and write descriptions that say who each product suits.

  4. Make sure price and stock update from the ERP, so sold-out items do not come first.

  5. See what your platform already offers (Shopify, WooCommerce or other) before buying anything new.

  6. If you need a new engine, ask for a trial on your own catalogue and your own phrases, in your language, not a demo on sample data.

  7. Keep keyword search for codes and brands, and semantic search for the rest.

  8. Measure before and after: searches with no results, exits after search, purchases from search.

We build search that understands. And the catalogue behind it.

We start from the phrases your shoppers already search for and from your catalogue, not from a tool. Searching by meaning is the same technique the AI assistants we build use to find answers in your own sources. And the same clean catalogue lifts your products on Google Shopping, marketplaces and in AI assistants.

  • An audit of today's search

    The phrases with no results, tests with accents, transliteration and typos, and a list of what is missing from the catalogue.

  • A catalogue machines can read

    Attributes in fields, descriptions for people, price and stock live from the ERP.

  • The right engine

    Your platform's own search when it is enough, set up properly on Shopify or WooCommerce. When the catalogue needs more, hybrid search on OpenSearch or inside the database itself, with PostgreSQL and pgvector.

  • Measurement

    We set up measurement in GA4 or the tool you already use, with the visitor's consent: what people search for, which phrases return nothing, which end in a purchase.

Start with usPart of Online shops & B2B portals

The questions we hear most.

It is search that compares the meaning of the query with the meaning of the products, not just the words. So it finds products for “something for an evening wedding on the beach” even when none of those words are in their titles.

Keyword search finds only what is written the way you wrote it. Semantic search understands synonyms, uses and descriptions. The best solutions today combine the two, so that codes and brands are still found exactly.

Multilingual models understand many languages, and Google and Shopify list Greek, for example, among their search languages. Whether it works well in your shop shows only in a test on your own catalogue, with real phrases, accents and typos.

Not necessarily. Shopify has semantic search through the Search & Discovery app; WooCommerce gets it through a plugin or an external service; a custom platform connects an engine such as Algolia or Elasticsearch.

No. Search brings the right products into the results list; a chatbot talks, answers questions and passes judgement calls to a person. They can share the same catalogue.

Attributes in fields (colour, material, size, use), descriptions that say who the product suits, and prices and stock that stay current. A product with a code for a title and no description is found by no model.

Yes. With enhanced measurement, GA4 records every site search when the phrase shows in the page address, and reports it as the “search term”. To see searches with no results as well, the shop sends one more parameter.

From searches with no results and from how many people leave right after searching. If the phrases with no results are products you stock, search is losing sales.