“Winter wedding dress.” “Comfy couch for a small living room.” “Gold earrings that hug the ear.”
That’s how people shop now—less like they’re filling out a database form, more like they’re talking to the one helpful employee in the store who actually knows where things are.
And here’s the ugly little secret a lot of e-commerce sites don’t want to admit: their internal search bars still behave like it’s 2009. They’re built for rigid keywords, not real-life intent. When search whiffs—zero results, irrelevant junk, duplicate listings—customers don’t “browse more.” They bounce. Conversion takes the hit.
This isn’t a niche problem. On many retail sites, search might represent a smaller slice of total traffic, but it’s disproportionately packed with buyers. These are the people who already have a budget, a size constraint, a style in mind, a deadline. They’re raising their hand and saying, “I’m ready.” If your site answers with “0 results,” you’ve basically told them to go spend their money somewhere else.
“Zero results” is the digital equivalent of shrugging at a customer
Most e-commerce search still runs on a simple idea: match the words a shopper types to the words sitting in your catalog—product titles, descriptions, attributes. If the strings overlap, the product shows up. If they don’t, the system panics.
This works fine when shoppers know exactly what to ask for: a brand name, a model number, a specific product term. But it falls apart the moment the query turns descriptive: “shoes for standing all day,” “warm coat that isn’t too long,” “round table for a narrow kitchen.”
The first failure is obvious: the dreaded zero-results page. It’s a blunt message—your store either doesn’t have what they want, or doesn’t understand them.
The second failure is sneakier: results appear, but they’re off-topic. The algorithm latches onto one shared word and serves a pile of products that technically “match” while completely missing the point. Either way, the shopper has to rewrite the query, mess with filters, or leave.
Synonyms make it worse. The French example “earrings that hug the ear” often maps to a known category—ear cuffs—but the catalog might list them under a different phrase, or only in English, or inconsistently across brands. Same headache with materials (“merino wool” vs. “merino”), activities (“hiking” vs. “trekking”), or constraints like “small living room,” which really means “don’t show me anything deeper than X inches”—assuming you even have depth data.
And then there’s the “help me choose” layer. “Dress for a winter wedding” isn’t a SKU. It’s a bundle of requirements: sleeves, warmth, formality, color (and yes, don’t show up in white), something that works with a coat. Keyword matching can’t infer that unless your catalog is obsessively structured and someone has hand-built rules for every scenario—which turns into a maintenance nightmare fast.
The irony is brutal: internal search works best when the customer already knows what to ask for. It breaks right when it should act like a good salesperson.
Mobile made shoppers wordier—and your catalog data is getting exposed
Long, chatty queries aren’t people being “extra.” They’re a predictable result of shopping on phones and getting used to autocomplete, chat interfaces, and voice assistants. People type the way they talk.
That changes what a search box has to do. “Comfy couch for a small living room” includes a space constraint, a comfort preference, and a room context. A strict engine will overweight “couch” and “living room” and basically ignore “small” unless the catalog has clean, standardized dimensions it can actually use.
Many catalogs don’t. Dimensions are missing. Attributes aren’t consistent. Descriptions read like marketing copy instead of usable specs. So conversational search becomes a stress test for data quality—and a lot of retailers are failing it.
Fashion is the same story. “Winter wedding dress” implies seasonality, formality, color expectations, and practical stuff like whether it works with closed-toe shoes. Without structured attributes, the engine grabs anything with “wedding” in the description, or any “winter dress” that’s totally wrong for the occasion.
And shoppers compare experiences. If they’ve been trained by big platforms to expect search that “gets it,” your clunky results don’t just feel inconvenient—they make your whole store feel smaller and worse stocked than it really is. Relevance becomes brand perception.
Semantic search vendors are selling “intent,” not keywords
This gap—human language vs. catalog logic—is why “semantic search” is having a moment in e-commerce. The pitch: stop counting exact word matches and start interpreting meaning.
The French article points to Kimera Technologies as one of the companies playing in this space, with an example deployment on Percentil (a European secondhand fashion platform). The idea is straightforward: take messy, natural-language queries and map them to product attributes even when the words don’t line up perfectly.
So “gold earrings that hug the ear” can be understood as a style/category/shape signal—ear cuffs—without requiring the exact phrase to appear in every product listing.
But here’s the part vendors don’t always put in 72-point font: semantic search can’t conjure missing facts. If your catalog doesn’t include dimensions, materials, or coherent categories, no amount of “AI” is going to magically filter couches by depth or dresses by warmth. The best outcomes come from pairing smarter query understanding with boring, disciplined catalog work: standardized attributes, required fields, QA checks, automated enrichment, and human validation.
When it works, the business case is simple: fewer dead-end searches, more clicks on relevant results, fewer backtracks, less abandonment. In e-commerce, every extra step is a conversion tax. Fix search, and you stop paying it.
If you don’t measure search failure, you’ll keep funding it
The reason this problem lingers is that it hides inside “overall conversion rate.” Teams track revenue, CAC, and checkout drop-off. They often don’t track the health of internal search with the same seriousness.
They should. The telltale metrics aren’t exotic:
• Zero-results rate
• Query reformulation rate (how often people immediately try again)
• Abandonment after search
• Click-through rate on search results
• Share of searches that lead to add-to-cart or purchase
A bad search experience creates a domino effect: confusion, then doubt, then retries, then exit. In competitive categories—where shoppers can jump to Amazon, Google, or a rival in seconds—that exit is immediate. And because the loss gets smeared across aggregate metrics, search rarely gets blamed.
The fix starts with unglamorous detective work: pull search logs. Find the top 100 queries with low clicks. Look for “intent words” like “for,” “with,” “without,” “that,” “small,” “warm.” Track recurring synonym problems (“gold” vs. “gold-tone,” “compact” vs. “small”). Pair that with session samples and you’ll usually see the mismatch fast: your catalog is written like a warehouse inventory sheet; your customers shop like humans.
Yes, you can make the zero-results page less of a dead end—suggest nearby categories, pre-filled filters, helpful content, query rewrites. That’s a bandage. The wound is comprehension.
And the budget angle is where executives should actually get mad: paying to acquire traffic—then letting internal search sabotage the visit—is paying twice. First for the click, then again in lost sales. Retailers who treat search like a core conversion system, on par with payments and shipping, plug a quiet leak that’s been draining revenue for years.
