Search that reads the sentence.
Four searches against one catalog: a vague description, a gift with a budget in it, a bulk trade enquiry answered in MOQs, and a search that finds nothing and says so. The page around the search box keeps personalizing throughout — banner, hero and recommendations. Switch the visitor below to run all four.
Built with AI Search — see the feature →
Everything for getting there and back
Outerwear, bags and the small things that make a commute bearable.
Insulated shell jacket£180 · rated to −5°Best fitCommuter backpack£95 · waterproof baseBought with
Thermal flask set£42 · 8-hour holdBought with
Guide: layering for a cold commuteEditorial · 4 min readContentOne catalog, four different searches.
Feed it the catalog you already have
Products and articles come from the same feed the recommendation engine reads — prices, stock and editorial together, re-read daily. Search and recommendations cannot disagree, because there is one source.
Let it read meaning, not keywords
A sentence with none of the catalog’s words in it still resolves, and a constraint written in plain English — a budget, a date, a quantity — is applied rather than ignored.
Rank per visitor, and log what is missing
The same query returns a different order for a returning customer, a gift buyer and a trade account. Searches that return nothing are written to a gap report, which is a merchandising list you did not have to commission.
Free AI tool · Inspire me
See ideas like this for your own site.
How Personyze personalizes search on this store.
The sentence is understood, not tokenised
“Something warm for a cold commute” shares no words with any product title here, so a keyword search returns an empty page. Meaning-based matching returns the insulated shell, and the two things people buy alongside it.
Constraints in the query are obeyed
A budget, a size, a date or a quantity written into the sentence becomes a filter. The £145 watch is the closest match to “runner”, and it is excluded because the visitor said under £80.
Results carry editorial as well as products
Guides and articles are in the same index, so a buying question can be answered with a guide instead of being forced into a product page. That is content recommendation and product recommendation in one ranked list.
A trade visitor searches a different catalog
Firmographics from an ABM integration switch the index to wholesale — MOQs, trade price, lead time — and suppress the consumer promotions entirely.
History reorders the same query
For a returning customer the ranking knows what they already own, what they viewed three times, and what they are due to reorder. Same catalog, same words, different order.
The searches that fail are the useful ones
A query with no match says so, offers the nearest real thing and a notify-me, and is logged. Thirty-eight people asking for something you do not sell is a merchandising decision, not a bug.
See how it is built, and where else it fits
Questions about this example
How is AI search different from keyword search?
It matches what the shopper means rather than the words they typed: “something warm for a cold commute” returns insulated jackets and layering even when those words are not in any product title, and the results are ranked for the visitor.
Can a search respect a budget or a size?
Yes. A budget, size, date or quantity written into the sentence becomes a filter, so items outside it are excluded rather than just ranked lower.
What happens when nothing matches?
The search says so, offers the nearest real item and a notify-me option, and logs the query, so repeated searches for something you do not sell show up as a merchandising signal.
Put AI search on your catalog.
It reads the same feed as your recommendations, obeys the constraints people actually type, ranks per visitor, and tells you what they searched for and could not find.