What is Retrieval?
- Retrieval
- Retrieval is the process by which an AI assistant fetches live documents while composing an answer, rather than relying only on what it learned during training. An assistant using retrieval searches the web, reads the pages it finds, and grounds its response in them, which is why crawlability and parseability decide whether a site can participate at all.
01 In practice
Retrieval and training memory are two different routes to being named, and they fail for different reasons. A brand absent from retrieval has a sources problem: nothing findable exists, or what exists cannot be crawled. A brand absent from memory has an authority problem: not enough was written about it, in enough trusted places, before the model's training cut-off.
The two also repair on different timescales. Retrieval problems are usually technical and can move within weeks of a fix being deployed. Memory problems require a body of published material to accumulate and be trusted, which is measured in quarters and years.
The work behind it
Tell the models what you are.
Markup that tells models what your business is, on pages that render without JavaScript.