An embedding turns text into a list of numbers, and text with similar meaning gets numbers that are close together. That closeness is what a search uses. The numbers themselves are meaningless to read, and the relationship between two of them is the whole point.
About 1 min of reading
See the shape of an embedding.
One vector
Before you run it, say out loud what it will print. Then open the answer.
The comparison
Before you run it, say out loud what it will print. Then open the answer.
Read the first example one line at a time. A line that prints shows what appears on screen; the rest set things up quietly.
Printing an embedding tells you nothing. The check is whether two texts you expect to be similar score high, which is a test you can run and read.
Keep the chunk text with its embedding. When a search returns a match, you need the words to show the reader, and a match with no text is a match you cannot use.
A model turns a string into a few hundred numbers. The same string always gives the same numbers, and two strings about the same thing give numbers that point the same direction.
Compare two embeddings by how close they point. A high score means similar meaning, a low score means unrelated. That score is what ranks the chunks for a question.
Embeddings from two different models are not comparable, because the numbers mean different things in each. Embedding the documents once and the query with the same model every time is the rule that keeps the scores valid.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.