Find & ask

How Claude reads your rooms

Your rooms aren't only for people to read — they're a knowledge source your AI can read back. When you ask the assistant to work from what's already written, it pulls just the relevant text from your documents — not whole files — so it's fast, grounded in what you actually have, and never a guess.

You don't call this — Claude does. There's no button or command to learn. When the assistant needs to reason over your existing content — before drafting something new, or to pull together an answer — it reads the corpus for you automatically. This page just explains what's happening underneath.

Reading the point, not the whole page

Opening a document hands over the whole page — every word, heading, and bit of formatting. That's exactly right when you want to read or edit it. But when the assistant just needs to know what your documents say — to write a new one that's consistent with the rest, or to gather the facts behind an answer — reading entire documents is slow and wasteful. Instead it retrieves a short, relevant excerpt from each document that matters and reasons over those.

"Draft an onboarding email that matches our existing benefits and PTO documents."

"Summarize what our security policies say about vendor access."

In both cases the assistant reads the pertinent passages across several documents at once — you get a grounded result in seconds, without opening anything yourself.

Always scoped to what you can see

Reading the corpus obeys the same permissions as everything else. The assistant can only ever pull text from rooms and documents you are allowed to open — a private room you're not in never contributes so much as a sentence. See Access & people.

Only published content is ever read — drafts and works-in-progress never leak into what the AI reasons over.

Your own AI tools can use it too

Because you connect rooms to your Claude app, this capability is available to the assistant wherever you work — and to other AI agents you authorize. They read your rooms as a live, access-scoped knowledge source, always pulling the current published text, so an agent can reason over your company's real knowledge without anyone copy-pasting documents into a chat.

It stays current on its own

You never re-index or re-upload anything. The moment a document is saved, its text is captured and ready to be read back — so what the assistant reasons over is always the latest published version, not a stale copy. This is the same living corpus behind Search & Ask; keeping it healthy is covered in Content health & curation.