AI search glossary
Semantic search: how search by meaning and vector search work
Semantic search is a search method that matches queries to content by meaning rather than by exact keywords. It usually turns text into embeddings (vectors that capture meaning) and returns the items whose vectors are closest to the query's, a technique called vector search.
Also called: vector search, embedding search, neural search, dense retrieval
Why it matters for AI search
AI answer engines and RAG systems use semantic retrieval to pick the passages a model reads before it writes. A passage is retrieved when its meaning is close to the question, so a page that states the answer clearly, in the words of the topic, is easier to match than one that repeats a keyword without saying anything specific.
How semantic search works
A traditional keyword index stores which words appear on which pages. A semantic system stores what the text means. It does this with embeddings: Google's crash course describes them as dense vector representations that capture semantic relationships. Text with similar meaning ends up with vectors that sit close together.
At query time the system:
- Converts the query into an embedding with the same model used for the content.
- Finds the stored vectors nearest to it. At scale this uses approximate nearest neighbor algorithms such as Google's ScaNN, trading a little recall for speed.
- Returns the matching items, often reranked by another model.
Google's Vector Search documentation separates dense embeddings, which represent meaning, from sparse embeddings, which represent exact terms. Hybrid search uses both, so a query for a model number still finds the exact product while a loose description finds related ones.
Google Search itself has used language understanding models for years. In October 2019 Google said BERT helped it understand one in 10 U.S. English searches, reading small words such as "to" in context rather than ignoring them.
What it means for AI search
AI answer engines work in two steps: retrieve, then write. The retrieval step often uses semantic search over web results or an index of passages, which is the core of retrieval-augmented generation. With query fan-out, the engine runs many such searches per question.
This changes what earns a match:
- Meaning beats repetition. Repeating a keyword does not move a vector closer to a question; answering the question does.
- Passages matter. Systems often embed chunks of a page, so each section should make sense on its own.
- Exact names still count. Hybrid systems reward precise product names, model numbers and brand terms.
How to write for it
- Answer one question per section, with the answer in the first sentence.
- Use the plain words buyers use, plus the precise names of products and features.
- Keep sections self-contained so a retrieved chunk is still clear out of context.
- Cover related subtopics (price, fit, setup, comparisons) so more queries land near your content.
Related ideas
The publisher-side practice is semantic SEO. Engines also use entities to tie meaning to specific brands and products, and the model that writes the final answer is a large language model.
Questions
- What is the difference between semantic search and keyword search?
- Keyword search matches the literal words in a query. Semantic search matches meaning, so a query for 'cheap laptop for students' can find a page about 'affordable notebooks for college' even with no words in common. Many systems combine both.
- What is vector search?
- It is the usual way semantic search is built. Text is converted into embeddings, lists of numbers that capture meaning, and the system returns the items whose vectors are nearest to the query vector. Google's Vertex AI Vector Search uses its ScaNN algorithm for this.
- What is hybrid search?
- Hybrid search uses both dense embeddings, which capture meaning, and sparse embeddings, which capture exact terms. Google's Vector Search documentation describes it as combining keyword and semantic search, which helps with product codes and names.
- Is semantic search the same as semantic SEO?
- No. Semantic search is the technology on the engine side. Semantic SEO is the writing practice on the publisher side: covering a topic's meaning and intent so that systems which search by meaning can match your page.