A shopper types "red summer dress" into your search bar. Your catalog holds thirty red dresses that fit the season perfectly. She gets three unrelated results, then leaves the page. The product existed. The engine just did not find it.

When this kind of scene repeats, the instinct is to blame the search engine. In a large share of cases, the problem lies elsewhere: in the product page itself. A search engine, however advanced, can only find what your data describes. If the word "red" appears only on the photo and nowhere in the text, if the description is empty, if the color is filled into no attribute at all, then the product is invisible to search - even though it is live on the site.

30%
Of e-commerce visitors use the search bar during their session
Multiple studies
~50%
Of zero-result queries involve products that actually exist in the catalog
Market estimate
2-3x
Higher conversion rate for visitors who use site search
Multiple studies

These three figures combine into one simple conclusion. Site search is used by your best buyers, a significant share of their queries fail on products that do exist, and the most frequent cause is not the algorithm but the data you feed it. Improving your search starts with improving your product pages.

Your search engine does not read products the way you do

When you look at a product page, you see a photo, a price, a brand - a visual that is enough to grasp what it is. A search engine does not see the image. It indexes text and structured fields: the title, the description, the attributes (color, size, material, brand), the category. Anything not written into those fields does not exist for it.

On a modern engine like Vectail, powered by Google Vertex AI Search for Retail, two mechanisms work in parallel. Lexical search matches the words of the query against the words of the page. Semantic search goes further: it understands that a "waterproof jacket" is close to a "windbreaker" - provided your descriptions contain enough text for the model to grasp what the product is. In both cases the raw material is the same: what your product page actually contains.

The principle to remember: a product page is not only a sales page read by a human. It is also the data source that feeds your search engine. A thin page produces thin results, whatever the algorithm behind it.

The 5 product data mistakes that sabotage your search

These mistakes show up in almost every catalog we audit. They share one trait: they stay invisible until you look at search from the engine's point of view. Once identified, most can be fixed at the product feed level, with no redesign.

1

Titles written for the catalog, not for search

The title is the strongest signal for a search engine. Yet many pages display a title inherited from the supplier: manufacturer reference, technical jargon, internal codes. The customer never types "REF-4471 cherry colorway" - they type "red dress". If the title does not contain the words people use, the match never happens.

What the customer sees vs what the customer searches
Catalog title: "TF-2200 Black Std" - Customer query: "black coffee table" - Catalog title: "Running shoe M. Gel-X9" - Customer query: "men's running trainers"
The fix: structure each title around customer vocabulary - product type, brand, key attribute - before the technical references. A good title reads like a query: "Red summer dress short sleeves - Brand X" rather than "RED-SUMMER-SS-X-4471".
2

Empty descriptions, or copied from the manufacturer

The description is the fuel of semantic search. A page with no description, or with identical manufacturer text repeated across hundreds of products, gives the engine no usable signal. Natural-language search ("warm coat for winter", "gift for a coffee lover") collapses with no text to understand.

The supplier copy-paste trap
300 products with the exact same manufacturer description "Premium quality product built to last": none stands out, and semantic search has nothing to work with to tell them apart.
The fix: write unique descriptions that cover use, material, context and concrete benefit. These are precisely the words semantic search uses to connect an intent to a product. It is also the heart of product page SEO: the same text serves your visitors, Google and your internal engine.
3

Missing or inconsistent attributes

Color, size, material, brand, gender: these structured attributes power filters, facets and part of the relevance signals. When they are empty, misspelled, or filled inconsistently from one product to the next, filters become unusable and some queries never reach their target.

Classic product feed inconsistencies
"Red", "red", "Rouge", "Burgundy", "Vermilion" used at random for close shades - sizes sometimes as "M / L / XL", sometimes as "38 / 40 / 42" - brand sometimes filled, sometimes missing.
The fix: normalize attribute values across the whole catalog (a controlled vocabulary) and consistently fill the key feed fields. Clean attributes make filters reliable and strengthen result relevance.
4

Loose categorization

A product's category drives search, filtering and merchandising. A misfiled product, or one attached to a category that is too generic, becomes hard to surface at the right moment. And a Google product category field (google_product_category) left empty deprives the engine of an important cue for understanding what the product is.

Common symptom
An "espresso machine" filed under "Home > Misc" instead of "Home > Kitchen > Small Appliances > Coffee Makers": it never shows up when a visitor narrows their search by category.
The fix: maintain a consistent taxonomy and fill in the most precise product category in the feed. The simple rule: each product belongs where a customer would naturally go looking for it.
5

Ungrouped variants

When every size or color of the same item exists as an independent product, with no link between them, search degrades in two ways: results are flooded with near-identical duplicates, and the right product gets buried. Properly grouped variants, on the other hand, let you display one clean result per model, with a consistent "from" price.

With vs without grouping
Without: 8 lines for the same t-shirt (one per size) monopolizing the first page. With: 1 result "Organic cotton t-shirt" showing "from $19", with sizes handled as variants.
The fix: link the variants of a single item through the feed's group identifier (item_group_id). Vectail then automatically groups those variants under one representative product and computes the starting price, with no extra configuration.

Your product pages deserve a search engine to match

Vectail connects to your Google Merchant Center, imports your catalog and activates AI search powered by Google Vertex AI - in one line of code, no plugin.

Start free - 14 days, no credit card

How to reveal the blind spots in your product pages

You do not fix product data by guessing. You fix it by looking at what customers actually search and what fails. Two sources complement each other.

The zero-result queries from your engine are the most direct signal. In the Vectail dashboard they are listed and sorted by frequency. A query that comes back often and returns nothing almost always points to missing data: a word absent from a title, an empty attribute, an isolated variant. That is your task list, ranked by impact.

GA4 with site search tracking completes the picture. By cross-referencing high-volume terms with those that lead to no click on a product page, you spot the queries that technically "find" results, but not the right ones - often a sign of titles or categories to revisit.

Not to be confused: a vocabulary gap (the customer says "fridge", your catalog says "refrigerator") is solved with synonyms, not by rewriting the page. Missing or wrong data, on the other hand, is fixed at the source, in the product page. The two levers are complementary.

Where to start

There is no need to redo the whole catalog at once. The most cost-effective approach is gradual and data-driven.

  • Export the 20 most frequent zero-result queries from the dashboard. That is your starting point, ranked by volume.
  • For each one, identify the cause: product absent from the catalog, missing word in the title, empty attribute, or a plain vocabulary gap to handle with a synonym.
  • Fix it at the source in the product feed: readable titles, unique descriptions, normalized attributes, grouped variants.
  • Re-run the catalog sync and measure the drop in your zero-result rate over the following days.

A well-built product page serves everyone at once: your visitors who quickly understand what they are buying, your Google ranking, and your internal search engine that finally has something to work with. It is one of the rare e-commerce projects where a single effort improves three channels at the same time.

Back to blog