Embeddings
Embeddings are lists of numbers that capture what a piece of text means, so a system can tell how close a page is to a question.
Embeddings are lists of numbers that stand for what a piece of text means. An AI system turns a question into an embedding, does the same for pages or passages, and then compares the numbers to find the text closest in meaning to the question.
They're part of why AI search works differently from matching keywords. A page can be judged a close fit for a question it never uses the exact words of.
How they work
Think of a map where every piece of text has a spot. Texts that mean similar things sit close together, and texts about different things sit far apart. An embedding is the set of coordinates for one spot, usually hundreds of numbers long instead of two.
For example, "cheap espresso machine for beginners" and "affordable first coffee maker for home espresso" share almost no words, but their embeddings sit close together. "Espresso martini recipe" shares a word with the first and sits much further away.
When a system retrieves pages for an answer, it often finds the passages whose embeddings sit nearest the question's, then reads those.
What it means for your pages
Since meaning counts more than exact wording, repeating a phrase over and over doesn't pull a page closer to a question. Covering what the person actually wants to know does. A passage that answers one question clearly tends to sit close to that question. A passage that wanders across five subjects sits somewhere in the middle of all of them, close to none.