In plain terms
A search engine for the inside of the company. People type a question once and get results from the shared drive, the wiki, the CRM and the chat history together, where before they had to search each system in turn or ask a colleague. With a language model on top, they get a written answer with sources in place of a list of links.
Why it matters
It is the foundation under most workplace assistants: an assistant that answers from company knowledge is enterprise search plus a model. The difficult part is permissions. Every result must respect who may see what, in every source system. AI search also exposes old mistakes: a salary file that was technically open to everyone but that nobody ever found will now be found. Review access rights before switching it on.
Example
A new sales engineer asks, “What did we promise Acme about data residency?” The system searches CRM notes, the contracts folder and a chat thread, and answers in three sentences with links to all three. A colleague outside the account team asks the same question and gets an answer without the contract, because they have no access to that folder.
Most often confused with
Enterprise Search vs. Web search
Web search ranks public pages using signals such as links between sites, and shows everyone the same index. Enterprise search has no such link signals, many more content formats and a different view of the index for every user. It relies on freshness, authorship and how close a document is to the person asking.
Under the hood
Building blocks: connectors to each source system with incremental synchronisation; an index that stores each item's access-control list and enforces it at query time; identity mapping, so the same person is recognised across systems; hybrid retrieval with reranking; and ranking signals such as recency, author, team and usage. Many products also build a graph of people, projects and documents to personalise results. Two architectures exist: a central index, which is fast and ranks well, and federated search, which queries each system live and avoids copying data. The generative layer is RAG over this index with citations. Risks: over-permissive source systems, stale permissions in the index, and injected instructions inside indexed content. Products include Glean, Microsoft 365 Copilot and Gemini Enterprise; general assistants now offer connectors to the same systems.