Today we're introducing AI Search: AI site search for online stores that finds what a shopper means, not just what they typed. It runs on a model trained on your own catalog. The searches that used to fail quietly become the clearest signal you have about what your customers want.
The searches that return nothing
Your most valuable visitor is the one a search sends away empty-handed.
A shopper types "a dessert for ten people" into your store. Your catalog has three cakes that serve ten. Keyword search matches words, not meaning, so it returns an empty page, and the shopper leaves. The product was there. The sale was not.
This is the quiet, daily failure of on-site search, and it hits your most valuable visitor. Someone who reaches for the search bar has already decided to buy and is telling you exactly what they want. Baymard Institute finds that nearly half of ecommerce sites give that shopper no effective way to recover when a search returns zero results. The empty page just sends them to a competitor.
At the volume a large catalog does, those empty pages add up to a slice of revenue walking out the door, invisible until now. For a store where search drives real sales, closing that gap is one of the few levers that moves the top line without buying more traffic.
AI Search is built for that moment. It reads the request the way a person would, and it answers with the products that fit.
A model trained on your catalog
Your search is relevant from its very first query, before it has any traffic to learn from.
Most product discovery tools run one generic engine over everyone's data and hope your traffic slowly teaches it what you sell. We do the opposite. Every store gets its own model, trained on your catalog alone and running on our infrastructure. It knows only your shop, and it speaks your language from the very first search. It knows that in your shop a "napoleon" is a pastry, not an emperor, and a "devil's food" is a chocolate cake.
It reads "gluten free cake for a birthday" the way your sharpest shop assistant would, as one request rather than four keywords. It is built on a language model called BERT, the kind behind modern language understanding, tuned to one obsession: your products. And it gets sharper the more your customers search.
It rescues the search before it fails
A typo or an over-specific phrase should never cost you a sale.
When a search would otherwise come back empty, AI Search does three things before giving up:
- Reads through typos. So "choclate cake" still finds the chocolate cake.
- Learns your customers' words. So "vegan" finds the dairy free line and "for a gift" finds what you sell as a hamper, even when those are not the words on your labels.
- Widens the net. It loosens the strictest part of the query first, keeps the intent, and returns the closest real products rather than a dead end.
The shopper sees an answer. You keep the visit. And when someone would rather ask than search, our chat widget picks up the same intent and answers right in the cart.
Your customers are telling you what to sell
Your search bar is the most honest market research you own.
Every search is a sentence of intent. Not where a visitor came from, but what they actually want, in their own words. AI Search reads those sentences back to you.
Your site search analytics show what people search for most, which product they click after each query, and where attention is landing. A shelf view sorts your catalog by what earns its place. Some products get seen and bought. Some get seen and skipped. The quiet ones convert whenever someone finds them. It also separates two kinds of gap. A catalog gap is real demand for something you don't stock. A language gap is when you have the product but not the word your customers use for it.
The searches that return nothing are the most useful of all. Where other tools treat a failed search as an embarrassment to hide behind fallback products, we hand it to you as a list: the exact things your customers asked for and you could not answer. Export it, and it becomes a buying plan. Stock the item, rename the category, or bundle the products people keep pairing. For a merchandising team, that list is a standing read on what your market wants next.
Measured honestly
We count the revenue we can trace to a search, and nothing we cannot.
It is easy to dress up search with big numbers. Call every empty result a lost sale, add them up, and present a frightening total. We don't, because most of those searches are noise, and you cannot make decisions on a number you made up.
So we draw a hard line. We report attributed revenue, money from a sale we can trace back to a specific search, separately from the total you observe in the same period. One is what the search earned. The other is just what happened while it was running.
This is the number that pays for the platform. For a store where search moves real revenue, recovering a few points of failed searches is money you can see, attributed to the search that earned it.
A search that finds nothing is not a lost sale. It is a customer telling you what to sell next.
That honesty is the point. AI Search understands your customers, rescues the searches that used to fail, and turns every query into something you can act on.
Frequently asked questions
- How long does it take to set up?
- A few days. What takes time is training the model on your catalog, and that runs on our side.
- Does it work for a new store with little traffic?
- Yes. Because the model is trained on your products rather than on visitor behavior, it is relevant from the first search, before you have any clickstream.
- What happens when a shopper searches for something we don't sell?
- It returns the closest real products instead of a dead end, and it records the query in your zero results list so you can decide whether to stock it.
- What do the search analytics actually show?
- What people search, which product they click after each query, which searches find nothing, and whether a gap is missing stock or just missing vocabulary.