A lot has happened here at LupaSearch!
We packed this release with improvements that quietly boost your site’s performance while noticeably upgrading your shoppers’ experience.
Here’s everything we’ve been building behind the scenes.
New Vector Search model (yes, it’s much smarter)
We’ve upgraded our vector search model, so it now understands not just keywords, but the actual intent behind a user’s query.
This means searches like “kids/for kids” will instantly surface children’s products, and “literature” will lead users to books — even if that exact word doesn’t appear anywhere in the product data.
Our previous version was good, but this one is sharper, faster, and far more context-aware (P.S. and still quicker than an OpenAI request).
It’s a powerhouse for eliminating zero-results pages, so we recommend turning it on when traditional keyword search comes up empty. The result? Fewer dead ends, and more happy shoppers.
Deeper Recommender Analytics (see what shoppers actually click)
Good news: you can now dig much deeper into how your shoppers interact with suggested products.
For clients using our plugin, you now get richer stats like most-clicked recommended products, plus filtering by source. See what people click on:
- blog widgets,
- “Frequently bought together”,
- homepage blocks,
- “More like this” on product pages,
- or “Popular now”.
You can even filter by the original product of the click. For example, look at how many users viewed a hair dryer and whether they clicked on another dryer, a diffuser, or a styling attachment.
This opens the door to much deeper insights into user behavior and intent, helping you fine-tune both merchandising and UX.
Minor but exciting: smarter handling of smashed-together words
Your shoppers can now search using up to three words without spaces, and LupaSearch will still understand what they meant.
This is a small UX boost with a big impact — perfect for accidental typos (“toysforboys”), fast mobile typing, or products with apostrophes that used to break search.
Statistical Filtering for ultra-precise search results
For stores where accuracy is non-negotiable, we’ve added a new logic that filters search results using real clicks and category data. This helps reduce irrelevant or loosely related products appearing in results.
With Statistical Filtering, you can set thresholds like “show only products with 5+ clicks for this keyword” — perfect for large catalogs, strict B2B setups, or anyone who wants zero noise.
It’s also a lifesaver when users sort by lowest price, preventing odd items (like “fridge drawers” or “cooling sprays”) from landing in results for “fridge”.
Easy Recommendation Rule configuration
Configure recommendation rules without touching any code. This means your store can automatically show accessories or related items based on product attributes — all through simple rule setup.
For example, if your business already defines what counts as “phone accessories”, you can enable a rule that recommends accessories whenever someone searches for “phones”.
New languages are here!
We’ve added several new languages — Serbian, Hungarian, Danish, Greek, and Bulgarian — and the system is handling the extra load beautifully.
More markets, smoother scaling, same high-speed performance.
Additional notes
We remain committed to improving our e-commerce search solution and delivering the best experience for our customers.
If you encounter any issues or have suggestions on improving it, contact our team at [email protected]. It will help us further improve our service.