Ecommerce search terms, defined.

Attribute extraction is the process of identifying structured product attributes — like material, color, or occasion — from unstructured text such as a product description or title.

Autocomplete suggests likely completions or corrections for a query as a shopper types, before they finish entering it or press search.

BM25 (Best Matching 25) is a widely used keyword-ranking algorithm that scores how relevant a document is to a query based on term frequency, adjusted for document length and how common each term is across the catalog.

Boosting is a merchandising technique that manually raises specific products higher in search or category results — for example, to promote a new arrival or high-margin item — independent of their computed relevance score.

Search click-through rate (CTR) is the percentage of search queries where the shopper clicked at least one result, used as a proxy for whether the results returned looked relevant.

Embeddings are numerical representations of text (or other data) that place semantically similar items close together in a multi-dimensional vector space.

Faceted navigation is the UI pattern of structured filters — checkboxes, sliders, dropdowns — that lets a shopper narrow a product list by attribute, typically shown alongside a category or search results page.

Faceted search lets shoppers narrow results using structured filters — like size, color, price, or brand — applied on top of a search query or category browse.

A head query is a short, high-volume search term — as opposed to a long-tail query — that accounts for a large share of total searches despite being a small number of distinct phrases.

Hybrid search combines keyword (lexical) matching with semantic (vector) matching in a single search system, so exact queries stay fast and precise while complex or descriptive queries still return relevant results.

An inverted index is a data structure that maps each term to the list of documents (products) containing it, allowing a keyword search engine to find matches quickly without scanning the entire catalog per query.

A long-tail query is a longer, more specific, and less common search phrase — as opposed to a short, high-volume "head" query — that often expresses more detailed shopper intent.

A merchandising rule is a specific, manually defined instruction — like "always boost this product in this category" — that a merchant sets to shape search or browse results independent of the underlying ranking algorithm.

The no-results page is the page a shopper lands on after a search returns zero matching products — typically a dead end unless the store recovers it with alternative suggestions.

Pinning fixes a specific product to an exact position in a results list — for example, always first — rather than just raising its relative rank the way boosting does.

A query fallback is a broader or related match a search system returns when it can't find a strong direct match for the original query, instead of returning an empty results page.

Query intent is what a shopper is actually trying to accomplish with a search — find a specific product, browse a category, or compare options — as distinct from the literal words they typed.

Query reformulation is when a shopper searches again shortly after an initial query, usually with different wording, typically because the first search didn't return what they wanted.

Retrieval-augmented generation (RAG) is a technique where a language model generates a response using information retrieved from an external source — such as a product catalog — rather than relying only on what it learned during training.

A ranking signal is any individual input a search system uses to help decide the order of results — such as textual relevance, popularity, recency, or a merchandising rule — before they're combined into one final score.

A ranking weight is the relative importance assigned to a specific ranking signal — like popularity or recency — when it is combined with others into one final relevance score.

A relevance score is the numerical value a search system assigns to how well a given product matches a given query, used to rank results from most to least relevant.

Search abandonment is when a shopper leaves a store, or gives up searching, after one or more failed search attempts — most commonly following a zero-result or low-relevance response.

Search attribution is the methodology a store uses to decide how much credit a purchase gets for having involved a search interaction, when a shopper's session touched multiple paths like browse or recommendations too.

Search conversion rate is the percentage of search sessions that result in a purchase, used to measure how effectively search turns queries into revenue.

Search exit rate is the percentage of search sessions where a shopper leaves the store shortly after searching, without clicking a result or continuing to browse.

Search suggestions are recommended queries or products shown to a shopper, either before they type anything or after a search that returned weak results.

Semantic search matches a query to results based on meaning rather than exact keyword overlap, so paraphrased or descriptive queries can still surface relevant products.

Session-based personalization adjusts search or browse results using only the current visit's behavior — recent clicks, dwell time, items viewed — without relying on a shopper's identity or purchase history.

Vector search finds matches by comparing numerical representations (embeddings) of a query and catalog items in a shared vector space, rather than comparing text directly.

A zero-result search is a storefront search query that returns no matching products, even though the catalog may contain items a shopper would consider relevant.