Custom corpus
A custom corpus is a chosen set of documents an AI system searches for its answers, in place of the open web or alongside it.
A custom corpus is a chosen set of documents that an AI system searches when it answers, in place of the open web or alongside it. It might be a company's internal files, a product's manuals, a library of licensed articles, or a set of trusted sources picked for one subject.
Google describes the idea in a published patent application: a language model answers a question using documents from a collection that an organisation or a person manages.
Why it matters to a marketer
Not every AI answer comes from the public web. When a business gives its staff an AI assistant, that assistant often answers from the business's own documents: supplier lists, contracts, internal reviews, past purchase orders. Your website may never come into it.
For example, a café chain's office manager asks the company's internal assistant, "Which coffee supplier do we use for our offices, and is there a cheaper option?" The assistant answers from the chain's purchase records and a supplier comparison someone wrote last year. If Harrow & Pine isn't in those files, it isn't in the answer, however well it does in public AI search.
What you can do about it
You can't see inside someone else's corpus, but you can make it more likely you end up in one. Make your product information easy to download and share: a clear spec sheet, a price list, a short PDF a buyer can save. When a buyer files your details, they become part of the corpus their assistant reads.
Licensing deals work the same way on a larger scale. When an AI company licenses a publisher's archive, that archive becomes a trusted corpus, and what it says about you can carry extra weight.
In AnswerOrbit
We ask the public versions of six AI platforms, as someone who isn't signed in. Answers inside someone's own account, or from a company's private assistant, are among the things we can't see. See How we count.