What is Semantic Search?
Semantic search retrieves documents by meaning rather than by exact keyword match, comparing numerical representations called embeddings so that a query and a document can match even when they share no words in common.
Keyword retrieval matches strings. A page about "small business CRM" would not match a search for "customer database for a two-person shop" despite being the right answer, because no significant term overlaps.
Semantic retrieval converts both query and document into vectors — long lists of numbers positioned so that texts with similar meaning sit close together. Retrieval becomes a nearest-neighbour lookup in that space. The two texts above land near each other, so the page is found.
Most production systems use both, a pattern called hybrid retrieval: keyword matching catches exact terms, product names and rare strings where semantic matching is weak, while semantic matching catches paraphrase. Optimising exclusively for either one leaves half the retrieval path on the table.
The practical consequence for writing is that exact-phrase repetition matters far less than it did, and covering a topic in the vocabulary real people use matters far more. Stuffing a keyword does not move a vector closer to a query it does not actually answer.
Why Semantic Search matters
It is why keyword density stopped predicting retrieval. Content is now matched on whether it means the same thing as the question, not whether it repeats the same words.
Embeddings encode meaning as position
Similar texts sit near each other in vector space. Retrieval is a proximity lookup, not a string comparison.
Hybrid retrieval is the norm
Keyword scoring is combined with vector similarity, because exact terms and rare product names are precisely where embeddings underperform.
Chunking decides what gets compared
Long pages are split before embedding. A section that changes topic halfway produces a muddied vector that matches nothing well.
Synonyms and paraphrase are free
Writing the same idea in the wording an audience actually uses widens retrieval coverage without repeating a phrase.