What is an embedding?

Imagine every document your company owns placed on a map where distance means similarity of meaning rather than similarity of spelling. The complaint about a leaking valve and the ticket about water under the machine sit next to each other, even though they share no words. That map is what an embedding produces: a model reads a piece of content and outputs a long list of numbers — typically several hundred — positioning it in a space where related meanings end up close together. Comparison then becomes arithmetic, which computers are very good at. This is the quiet machinery behind most useful AI features in a business setting. Search that finds the right answer when the customer used different words than your manual. Chatbots that answer from your own documents, because the relevant paragraphs are retrieved by meaning before the language model writes anything. Duplicate detection across supplier catalogues, product recommendations, automatic ticket routing. For a buyer the useful takeaway is that embeddings are how your own content becomes usable by AI at all — and that the quality of the result depends far more on how your documents are organised than on which model generated the numbers.