Testing on-site search for five brands with their own product copy

A while back, I shared how a mid-size merchant was losing shoppers to thin catalog data.

The obvious next question: does the same thing happen to the big, well-known brands, the ones with real search infra and engineering teams behind them?

So I ran a few tests on five brands from five different categories.

I took language straight off their product pages - the exact claims and words, typed it into their on-site search bar. The words a shopper who has just landed on the brand’s online storefront would be typing to find a product.

Search does not read a page. It compares attributes and text strings just like an AI agent does. Everything below is a different shape of that one fact.

Zero results for the exact words from the product

Our Place sells cookware that claims (on the product page) to be nonstick, made without PFAS, PTFEs, lead, or cadmium, and compatible with induction cooktops.

Search "induction compatible" on their site and you get zero results.

Search "PFAS free" and you get zero results too. Both times, the site falls back to a bestseller grid instead of saying so plainly.

One space can break the search

Here's the mechanical reason for Our Place’s search failing.

Search "nonstick" as one word - works.

Search "non stick" as two words, the same claim, and you will get nothing.

Search is matching whatever string got indexed against what I typed.

If the catalog has "nonstick" and the query is "non stick" - those are two different strings as far as search is concerned, even though they mean the same thing to a shopper.

This is the same reason "PFAS free" fails while the product page spells it "without PFAS". A shopper doesn't know which exact phrasing the catalog uses. Why would they?

One dropped letter, sixteen products gone

Glossier - if you search "lipstick" and spell correctly, 17 items come back. Drop one letter, search "liptick" - the result count falls to 1.

That's the difference between a shopper finding what they came for and a shopper assuming the brand doesn't carry it, over one dropped letter.

The top result has the lowest rating

Gymshark's on-site search works, in the sense that it returns real, relevant products.

The question is which one it puts first.

Search "gym top for men" and the top result is a t-shirt rated 2.4 stars, and later on the same row: 3.7, 4.4, and 4.5 star shirts - matching the same query.

Most default search setups rank on text-match strength first, and star rating or relevance to intent only get factored in if someone deliberately wires it that way.

Structured data that search uses

Mejuri's website has a real, working material attribute.

Search "18k gold", and every single result that comes back is labeled 18k Gold Vermeil.

A similar version of catalog data working correctly at Cotopaxi. The Tasra 16L and Allpa 26L Daypack both genuinely have laptop compartments.

Search "backpack for laptop".

In both cases, the structured data already exists somewhere in the catalog.

What it looks like when it works

Mejuri's search handled the query "ring stacking set" cleanly - the live preview and the full results page both returned the same four items - stacker sets and a bundle, matching exactly what a shopper would have meant.

Brands that didn’t do well at some tests performed well at others.

It means the gaps are easy to miss internally - the team searches for the things that already work, not the phrasing that happens to break.

Where this leaves you

The catalog and the search never agreed on language.

That's the whole pattern, repeated across five different brands. Nothing here required guessing what a shopper meant, and none of it needed a big team or a bigger budget to fix, just systems reading the same catalog data.

This is exactly what AI Search by Atomz is built to close.

It reads the same structured catalog that our AI writes, so natural language queries like the ones above actually match. Collective Shoes started using Atomz and search became their second-biggest revenue channel (NZ$600K+ in search-driven revenue).

Install Atomz on Shopify - try it out for free - enrich the catalog once, and every surface that reads it gets smarter at the same time.

Book a free strategy call - if you want a demo of this on your store.

- Ankit