The assistant suggests specific products in the conversation — as rich cards with image, price, and a buy button — based on what the visitor describes they need. Your "guides visitors to the right product" positioning, made literal.
A shopper arrives knowing what they want to *do*, not which product does it. "I need something warm for a hiking trip in November." "What's a good gift for someone who just started cooking?" "Which of these works with my model?"
These aren't support questions — they're buying questions. And a text-only answer that says "we have several options, check the jackets page" sends the shopper back to square one, scrolling a category page, hoping to recognise the right thing.
Product Recommendations close that gap: the assistant names specific products and shows them, right in the conversation.
Specific products, as rich cards. Instead of a wall of text, the assistant returns product cards — image, title, price, and a button that goes straight to the product page or adds to cart. The shopper sees exactly what's being recommended and can act on it without leaving the chat.
Recommendations grounded in your real catalog. Suggestions are drawn from your live, synced product data — so the assistant only ever recommends products that exist, are in stock, and are priced correctly. No invented products, no sold-out suggestions.
Needs-based matching. The shopper describes a need in natural language — "waterproof, under €150, for a child" — and the assistant matches against product attributes to suggest the best fits, explaining briefly why each one matches.
In the shopper's language. A German visitor describes their need in German and gets recommendations explained in German, with product details localized. The same conversation works in every supported language.
It is your positioning, made literal. Companin's promise is "guides visitors to the right product or page." A text answer with a link is a weak version of that. A recommended product card — image, price, buy button — is the promise delivered. This is the feature that makes the positioning visible in a demo.
It converts browsers into buyers. The hardest moment in e-commerce is the shopper who wants to buy but can't find the right item. Naming the product and putting a buy button in front of them removes the friction at exactly the moment of intent.
It beats category-page browsing. A category page is a grid the shopper has to interpret. A recommendation is a curated answer to their actual question. For stores with large catalogs, this is the difference between a shopper who finds the product and one who gives up.
No competitor does this well in the shopper's language. Generic chat widgets answer FAQs. A multilingual assistant that recommends specific products, in the visitor's language, grounded in live inventory — that's a sharp, demonstrable differentiator.
Product Recommendations build directly on [Product Catalog Sync](/roadmap/product-catalog-sync) — the assistant matches the shopper's described need against your synced product attributes (type, price, variants, tags) and returns the best matches as cards.
You control the behavior: how many products to suggest at once, whether cards link to the product page or add directly to cart, and which collections are eligible for recommendation. The assistant explains its picks briefly so the shopper understands the fit, rather than presenting an unexplained list.
When the assistant isn't confident a product genuinely matches, it asks a clarifying question instead of guessing — "is this for an adult or a child?" — so recommendations stay relevant rather than scattershot.
"Check our [category] page." Static, rule-based product quizzes that break when the catalog changes. The shopper scrolling a grid of forty products trying to find the one that fits a need they already described to your assistant.
This feature is planned, and depends on Product Catalog Sync shipping first. When it lands:
Once live, the assistant doesn't just answer the shopper's question — it puts the right product, with a buy button, directly in the conversation.