Why AI Search Beats Keyword Search (Embeddings Explained)

An embedding is a list of ~1,500 numbers that represents the 'meaning' of a piece of text. Two pieces of text with similar meanings have similar embeddings, even if they share no words. That's how 'happy puppy' can match 'joyful dog' in AI search — the words differ, the meaning vectors are close.

Some examples

• Spotify's recommendations work this way: every song has an embedding; songs with nearby embeddings get recommended.
• Notion AI search uses embeddings: 'find my notes about anxiety' matches notes that say 'stress' or 'overwhelmed' even without the word 'anxiety.'
• OpenAI's text-embedding-3-small costs $0.02 per million tokens — cheap enough that hobby projects are free.
• Vector databases (Pinecone, Chroma, Weaviate) store and search embeddings; they're how every 'chat with my docs' app works.

Try it!

Sign up for OpenAI API access ($5 free credit). Run their embedding example (10 lines of Python) on a CSV of your own notes. Then do a similarity search. You just built the core of every modern AI search engine in an afternoon.