An embedding is a list of numbers produced by a model for a piece of text. Texts with similar meaning end up with vectors that point in similar directions, so a search system can compare a query vector with a corpus of document vectors and return the nearest ones. Embeddings are how most retrieval systems find passages without keyword overlap.
They are not a summary and not readable. The dimensions have no individual meaning, and two embedding models produce incompatible spaces: a vector from one model cannot be compared with a vector from another. If you change the embedding model, re-embed the corpus.
Embedding quality depends on the domain and on how the text is chunked. A model trained on general web text may do poorly on legal clauses, source code, or product SKUs. Evaluate retrieval on real queries before you blame the generator. See RAG for how the vectors are used.