Merchants who want better product discovery on Magento 2 often ask the same thing: do we need an AI sales assistant, or do we just need better site search? It is a fair question, and the honest answer is that they solve different problems. This post explains where each one fits, so you can decide before you spend money on either.
What site search is good at
Magento 2 ships with catalog search built on OpenSearch, and many stores add a third-party search extension on top. Search works well when the shopper already knows what they want and can type it: a product name, a SKU, a brand, a model number. Tuned well, with synonyms, typo tolerance and good result ranking, it gets those shoppers to the right product page fast.
If most of your traffic arrives with a clear product in mind and still struggles to find it, fix search first. An assistant will not make up for missing synonyms or broken relevance.
Where search runs out
Search gets harder when the shopper describes a need instead of a product. Think of queries like “a waterproof jacket for hiking in spring, not too heavy” or “a drill bit for concrete that fits my 10 mm chuck”. The answer is usually spread across a category and several attributes, and a results grid leaves the shopper to work that out alone with filters.
That is the gap a catalog-aware assistant is built for:
- The shopper asks in plain words, and the assistant works out the category first, then narrows by attributes such as size, material or compatibility.
- When the request is too broad, it asks one clarifying question instead of dumping fifty results.
- It answers from your synced Magento catalog, so prices, stock and configurable product options come from your store, not from a crawl of your pages.
- It can answer common pre-sales questions from a curated set of store answers, such as delivery or technical fit, alongside the products.
How Claspwell actually finds products
It helps to be precise about what happens under the hood, because “AI search” is used loosely. Claspwell syncs your categories, products and shopper-facing attributes into its own database. A light language model step reads the shopper's message and turns it into a structured intent: which category, which attributes, what the shopper is asking for. The product lookup itself is then a normal database query over your catalog, and the results are shown as product cards from templates.
So the model helps understand the question, while the products shown always come from your real catalog data. Many simple turns never reach the model at all, thanks to three-tier routing.
Quick way to decide
- Shoppers search by name or SKU and still fail: improve site search.
- Shoppers land on a category and bounce because there are too many similar products: an assistant helps, because it can narrow by attributes in conversation.
- Your catalog is attribute-heavy (parts, technical goods, fashion with many variants) and support keeps answering “which one fits?” questions: an assistant helps.
- Your attributes are sparse or inconsistent: fix the data first. Neither search nor an assistant can filter on attributes that are not there. Our attribute checklist is a good starting point.
Most stores that benefit from an assistant keep their site search as it is. The two sit side by side: search for shoppers who know the product, the assistant for shoppers who know the problem.
Next step
If you run Magento 2 (Hyvä or Luma) and think your shoppers fall into the second group, request an integration. We review your catalog shape first, and a typical install takes 3 days to 1 week once scope and access are ready. You can also see how it works step by step.