Vector database
A database built to store embeddings and find the nearest ones to a query vector quickly, which is what makes semantic retrieval fast enough to use.
In depth
What it really means
A vector database stores millions of embeddings and answers one question fast: which of these are closest to this query vector. Pinecone, Weaviate, Qdrant and pgvector are common implementations.
You do not need one, and this term is here mainly so the pipeline makes sense. It also explains a real constraint: these indexes are built ahead of time, so newly published content is not retrievable until the index updates, which can take days.
How it works
- Content passages are embedded and stored with metadata.
- The index is structured for approximate nearest neighbor search.
- A query vector is compared against the index using cosine similarity.
- The closest matches are returned, then reranked.
Pros & cons
Pros
- Makes semantic search fast enough to run at web scale.
- Metadata filtering allows recency and source constraints alongside similarity.
- Understanding it explains the lag between publishing and being retrievable.
Cons
- Pre-built indexes mean new content is invisible until reindexed.
- Approximate search trades some accuracy for speed.
- It is infrastructure most marketers will never interact with.
Common mistakes
- Expecting new content to be retrievable immediately. Index refresh takes time.
- Assuming AI search always fetches live. Much of it runs against a pre-built index.
- Republishing at a new URL and losing the indexed history.
Best practices
- Publish ahead of when you need the content retrievable, not on the day.
- Keep URLs stable so indexed content is not orphaned.
- Update in place rather than republishing at a new URL.
- Make sure new pages are linked and in your sitemap so they get crawled promptly.
FAQs
What is a vector database?
A database built to store embeddings and quickly find the ones closest to a query vector. It is what makes semantic retrieval fast enough to be practical.
Do I need a vector database?
Only if you are building AI features yourself. For AI visibility, it is worth understanding as the reason new content takes time to become retrievable.
Why isn’t my new page showing up in AI answers?
Often because the retrieval index has not been rebuilt yet. Pre-built indexes update on their own schedule, which can take days or longer.
Is AI search live or indexed?
Both, depending on the system. Some fetch live pages, many query a pre-built index. That mix is why results vary between assistants for the same question.
Keep reading
Related on LymLyt
Beyond LymLyt
Further reading
Want this working on your site?
We build the content behind the term, ranked in search and cited by AI.
Book a 30-min call →