NLP Fundamentals
Semantic Search
Search that retrieves results by meaning rather than exact keyword match, typically using embeddings and vector similarity.
Definition
Semantic search interprets a query in terms of its meaning and returns documents that match conceptually, even when no words overlap. The technique relies on embeddings: query and documents are converted into vectors, and similarity in that vector space stands in for similarity in meaning. The result is a search experience closer to how humans look for information, where 'cancel my plan' and 'end my subscription' should obviously return the same answer.
Why it matters
Semantic search has changed user expectations across products. Customers no longer accept 'no results' for queries that are clearly answerable with rephrasing. Internal knowledge bases that rely on exact-match keyword search now feel broken. For content teams, the practical implication is that semantically rich writing — content that genuinely covers a topic in the language a reader might use — outperforms keyword-optimised thin pages. Most real-world systems use hybrid search: keyword for precision on names, IDs, and exact terms; semantic for recall on conversational queries. The combination is reliably better than either alone.
Examples
Keyword fails
Query 'turn off email notifications' returns nothing because the help article is titled 'manage your communication preferences.'
Semantic succeeds
Same query embeds near the help article and is returned at the top of results.
Frequently asked
Does semantic search replace SEO?
No, but it changes the rules. Writing that answers questions the way people ask them performs better than writing optimised purely for keyword density.
How does it handle typos?
Reasonably well — small errors usually do not move the query vector far enough to change the top results.
Is it always better than keyword search?
For natural-language queries, usually. For exact-string lookups (SKUs, error codes), keyword still wins.
Related terms
Embeddings
Numerical vector representations of text whose distances capture semantic similarity between pieces of content.
Vector Database
A specialised database optimised for storing high-dimensional vectors and retrieving nearest neighbours efficiently.
Retrieval-Augmented Generation (RAG)
A pattern in which a model is given relevant context retrieved from your own sources before generating an answer.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite