An embedding model converts input into a vector, a fixed-length array of numbers positioned in a high-dimensional space. Closeness in that space corresponds to closeness in meaning, so "how much does it cost" and "what are your prices" land near each other even though they share almost no words. Similarity is usually measured with cosine similarity or a related distance metric.
Embeddings are the foundation of semantic search, recommendation systems, clustering and deduplication. In a retrieval pipeline, documents are split into chunks, each chunk is embedded once and stored, and an incoming question is embedded at query time so the closest chunks can be pulled back as context for a language model.
Clerkzo embeds the pages it crawls from a customer's website so that visitor questions can be matched to the right content by meaning rather than keyword. That is why a visitor can phrase a question in their own words and still get an answer drawn from the correct page.
Related terms
Browse all 142 terms- Core Web VitalsCore Web Vitals are a small set of measurements Google uses to describe how a page feels to real visitors, covering how quickly the main content appears, how fast the page responds to input, and whether the layout jumps around while loading.
- TokensTokens are the small pieces of text, usually a word or part of a word, that an AI language model reads and generates. Models measure everything they process in tokens rather than characters or words.
- AI OverviewsAI Overviews are AI-generated summaries that appear at the top of some Google search results, answering the query directly and linking to the pages the summary drew from.
- Quick repliesQuick replies are the tappable suggestion chips shown inside a chat widget that let a visitor answer a question or pick a topic with a single tap instead of typing. They present a short set of likely responses, and selecting one sends it as the visitor's message.