Definition: what is semantic search?
Semantic search evaluates what a query means — not just which words it contains. Modern systems combine several technologies for this: vector embeddings for query and document, transformer-based relevance models, entity linking with Knowledge Graph integration, and intent classification (informational, transactional, navigational). A document can rank without containing the query words — and be irrelevant despite exact word matches.
The evolution
The milestones: Hummingbird (2013) brought entity orientation into Google Search, RankBrain (2015) machine learning for unseen queries, BERT (2019) contextual language understanding at query level, MUM and successors multimodal understanding. With LLM-based systems from 2022, semantic ranking became semantic answer synthesis — search no longer just understands, it writes (RAG).
Consequences for content strategy
- Intent clusters instead of single keywords: you optimise for question spaces and topic clusters; the tool for that is the topical map.
- Depth beats frequency: semantic search rewards exhaustive topic treatment (topical authority) — not keyword density. Thin content with keyword stuffing no longer ranks structurally.
- Entities instead of strings: consistent entity signals (author, organisation, field) anchor content in semantic space.
- Fewer but deeper articles: the opposite of the 2015–2022 content-farm logic.
Related terms
The technical foundation of semantic search is embeddings and vector search; the strategic response is topical maps and topical authority. The paradigm's continuation into generative systems is described under RAG.
Search understands — content has to as well
Since BERT, search scores meaning, not word lists. Anyone still thinking in keyword frequencies is optimising for a system that has not existed for years.
FAQ on semantic search
What distinguishes semantic search from classic keyword search? ▾
Classic search matches strings and weights term frequencies. Semantic search scores meaning and intent via embeddings, language models and entity linking — content can rank without exact word matches and be irrelevant despite them.
Since when has Google Search been semantic? ▾
Gradually: Hummingbird (2013) and RankBrain (2015) were precursors; with BERT (2019), contextual language understanding became standard for practically all queries. LLM-based systems from 2022 extended the paradigm into answer synthesis.
Are keywords irrelevant now? ▾
No — they remain the expression of user language and a research instrument. What has become irrelevant is keyword density as an optimisation target. Relevance comes from semantic completeness, entity clarity and intent fit, not repetition.
What does semantic search mean for GEO? ▾
GEO builds directly on it: generative systems retrieve semantically (RAG), and a brand's semantic position — shaped by co-occurrence and entity consistency — decides whether it is found and cited as a source.