How do you search for a product when you don't know what it's called?

Use two different search boxes for two different jobs, because they behave in opposite ways. In Google, describe the thing in a full sentence, the way you would describe it to a friend. Google's own figures put AI Mode queries at two to three times the length of a traditional search, and its one-year review of AI Mode put the average at triple, so the long descriptive query is now the case the engine is built around. In a retailer's own search box, do the opposite and type three or four nouns. That box is usually matching your characters against a product title a merchandiser typed, so every extra word is one more word that had to be in the title for you to match. The reliable order is category, then silhouette, then material, then the one detail that separates it from its neighbours.

Why does a long descriptive search return nothing on a shop's website?

Because most site search is keyword matching against a product title, not an understanding of your sentence. Baymard Institute's benchmark found that a site's search logic is often based entirely on matching keywords against the product title or main description rather than the full product data set, that 60% of ecommerce sites do not support thematic queries such as looking for an outfit for an occasion, and that 16% of top-benchmarked sites fail to return their own product when you search the exact product name or model number printed on that product's own page. So zero results usually means your words and the seller's words did not overlap, not that the shop lacks the item. Cut the sentence to its nouns and search again before you conclude anything.

What order should the words go in when you search for clothing?

Match the order the seller was told to write the title in. Google's Merchant Center guidance for apparel recommends titles built as brand, gender, product type, colour, size and material, and tells merchants to front-load the important attributes because titles get truncated in most ad formats. Sellers follow that pattern, so searching in the same shape lands on their words. With no brand to lead with, start at the product type and work outward: square neck ribbed top, suede fringe jacket, cane webbing sideboard. Keep it to three or four words, then narrow with the site's own filters rather than by typing more words.

Where do you find the right word for a style detail you can't name?

Open any large retailer's category page and read the filter sidebar. Those filter names are the retailer's own taxonomy, which is the same vocabulary its search index was built on, and it is free to read without buying anything. A shoe category will list mule, slingback, Mary Jane, kitten heel and block heel; a dress category will list bateau, sweetheart, halter, cowl and square necklines. Click the one that looks like your item, look at what comes back, and you have both the word and a confirmation that it is the right word. It is faster than a glossary because it shows you pictures next to the terms.

← All posts

A shop's search box only knows the seller's words

September 14, 2026 · 6 min read

A clay-style magnifying glass leaning against a phone

Use two search boxes, and type differently into each. In Google, describe it in a whole sentence. In the shop's own search box, type three nouns.

Those are opposite instructions and both are right, because the two boxes do completely different things. One is trying to understand you. The other is matching your characters against a product title that a merchandiser typed on a Tuesday.

The two boxes want opposite things

Google has moved toward the sentence. Its own figures put AI Mode queries at two to three times the length of a traditional search, and its one-year look at AI Mode put the average at triple. The long rambling description, the thing everyone was trained out of typing, is now the case the engine is designed around. “Ribbed tank top with a straight neckline and thin straps that the girl in the video wore tucked into jeans” is a reasonable query there.

Paste that same sentence into a retailer's search box and you will usually get nothing. Baymard Institute's benchmark work found that site search logic is frequently based entirely on matching keywords against the product title or main description instead of the full product data, that 60% of ecommerce sites do not support thematic queries, and that 16% of top-benchmarked sites fail to return their own product when you search the exact model number printed on that product's own page.

That last number is the one worth sitting with. A shop that cannot find its own product from its own product name is not going to find it from your description of a two-second video. Zero results is not evidence the shop lacks the item. It is evidence that your words and the seller's words did not overlap.

The order the words go in

Sellers do not name things freely. Google's Merchant Center guidance tells merchants to build apparel titles as brand, gender, product type, colour, size and material, and to front-load the important attributes because titles get cut off in most ad formats. Feed tools enforce roughly that shape across thousands of shops.

So the words already exist in a known order. You have no brand, which means you start one slot in: category, then silhouette, then material, then the single detail that separates the item from its neighbours.

“square neck ribbed top” rather than that tight top with the straight-across neckline in the stretchy ribbed fabric.
“cane webbing sideboard” rather than the wooden cabinet with the woven rattan panels on the doors.
“gold curb chain bracelet” rather than the chunky flat gold bracelet with the links that lie sideways.

Each of those is three or four words and every one of them plausibly appears in a real product title. The long version contains maybe two words a seller would have typed, surrounded by eight that guarantee a miss.

The words themselves

You do not need to learn a glossary. You need one correct word for the part of the item that makes it distinctive, and the trade has already named every part of everything.

Necklines. square, sweetheart, halter, cowl, bateau (also written boat), keyhole, henley.
Sleeves and shoulders. puff, bishop, flutter, dolman, cap, drop shoulder.
Cut and waist. A-line, shift, wrap, peplum, empire waist, bias cut.
Shoes. mule, slingback, Mary Jane, kitten heel, block heel, ballet flat, loafer.
Bags. baguette, hobo, bucket, crossbody, top handle, clutch.
Fabric and finish. bouclé, ribbed, poplin, satin, corduroy, cable knit, sherpa.
Home. fluted, scalloped, cane webbing, arc lamp, travertine, slipcover.

One of those words in a four-word query does more than any amount of describing. “Peplum” eliminates every top in the catalogue that is not one.

The filter sidebar is a free dictionary

Here is the part almost nobody does. When you cannot name a detail, open a big retailer's category page and read the filters down the left-hand side.

Those filter names are the retailer's own taxonomy, and the search index was built on the same vocabulary. A shoe department will list mule, slingback, kitten heel and block heel as checkboxes. A dress department will list the necklines. Click the one that looks like your item and the results tell you instantly whether you picked the right word, which is something a glossary cannot do, because the glossary has no pictures of what the seller thinks the word means.

Two minutes in a filter sidebar is worth more than an hour of rephrasing. You leave with the exact term the index is expecting.

Add one word at a time, and stop when it gets worse

Searching is a ladder, not a sentence. Start with two nouns, look at what comes back, add one word, look again.

Start too broad on purpose. A thousand results you can filter beats zero results you cannot. Zero is a dead end that tells you nothing about whether the item is there.
Add the word that eliminates the most. Silhouette and material cut a catalogue down faster than colour, because colour names are where sellers are most creative.
Put attributes in filters, not in the query. Type the shape, click the colour and the size. The seller called it camel, oat, biscuit or tan, and their filter knows which one. Your keyword does not.
Back up the moment results get worse. The word you just added is not in their titles. Remove it and try a different one rather than adding a fifth.

When the catalogue you are searching is other people's resale listings rather than a retailer's, the vocabulary shifts to era and construction words and the filters stop being reliable. We took that case apart separately in the post on dating a piece with no label.

When the word does not exist

Sometimes there is no name to find, and recognising that quickly saves the evening.

A plain factory item has no distinctive vocabulary by definition. One mould, a dozen sellers, a dozen invented brand names, and titles that are just keyword soup. For those, the search that works is the boring category phrase plus a filter for price and rating, and the thing you are choosing between is sellers rather than products.

And the whole naming exercise is a workaround. You are reconstructing a word because the original source, the video, did not hand you one. The manual playbook for getting a name out of the video itself is usually a shorter path: captions, on-screen text, the comments, the creator's storefront. Description is what you fall back to when all four come up empty.

What we can and cannot do here

Stealsy removes the naming step rather than helping you through it. Share a video and it reads the post, the caption, the on-screen text, the audio and the comments, finds the products, prices them, and opens every shop link to check it resolves before you see it. Where a cheaper version of the same look exists, it shows that next to the original. Heart something instead of buying it and it watches the price and tells you when it drops.

The honest limit is that it needs the video. There is no box to type “that ribbed square neck top” into. If the thing you want exists only as a memory of something you scrolled past last Tuesday, everything above is still the method, and nothing we have shortens it.

It is iPhone only, iOS 18 or later, and free to try.

Get Stealsy for iPhone — share a video, get the products, prices and verified links.

Common questions

How do you search for a product when you don't know what it's called?: Use two different search boxes for two different jobs, because they behave in opposite ways. In Google, describe the thing in a full sentence, the way you would describe it to a friend. Google's own figures put AI Mode queries at two to three times the length of a traditional search, and its one-year review of AI Mode put the average at triple, so the long descriptive query is now the case the engine is built around. In a retailer's own search box, do the opposite and type three or four nouns. That box is usually matching your characters against a product title a merchandiser typed, so every extra word is one more word that had to be in the title for you to match. The reliable order is category, then silhouette, then material, then the one detail that separates it from its neighbours.
Why does a long descriptive search return nothing on a shop's website?: Because most site search is keyword matching against a product title, not an understanding of your sentence. Baymard Institute's benchmark found that a site's search logic is often based entirely on matching keywords against the product title or main description rather than the full product data set, that 60% of ecommerce sites do not support thematic queries such as looking for an outfit for an occasion, and that 16% of top-benchmarked sites fail to return their own product when you search the exact product name or model number printed on that product's own page. So zero results usually means your words and the seller's words did not overlap, not that the shop lacks the item. Cut the sentence to its nouns and search again before you conclude anything.
What order should the words go in when you search for clothing?: Match the order the seller was told to write the title in. Google's Merchant Center guidance for apparel recommends titles built as brand, gender, product type, colour, size and material, and tells merchants to front-load the important attributes because titles get truncated in most ad formats. Sellers follow that pattern, so searching in the same shape lands on their words. With no brand to lead with, start at the product type and work outward: square neck ribbed top, suede fringe jacket, cane webbing sideboard. Keep it to three or four words, then narrow with the site's own filters rather than by typing more words.
Where do you find the right word for a style detail you can't name?: Open any large retailer's category page and read the filter sidebar. Those filter names are the retailer's own taxonomy, which is the same vocabulary its search index was built on, and it is free to read without buying anything. A shoe category will list mule, slingback, Mary Jane, kitten heel and block heel; a dress category will list bateau, sweetheart, halter, cowl and square necklines. Click the one that looks like your item, look at what comes back, and you have both the word and a confirmation that it is the right word. It is faster than a glossary because it shows you pictures next to the terms.