Make your deal documents readable by AI assistants — safely.

Because it moves your confidential documents outside every control you set, and hands the answer to a model that will guess. When you email a PDF, the unsafe default takes over: an analyst on the other side drags it into a public consumer chatbot to draft their diligence notes. Now your CIM or investor package has been exfiltrated to a third-party tool that has no access control over who can query it, no audit trail of what was asked or answered, and no awareness of the room's permissions. On consumer tiers, the content of an uploaded file can be used to train models by default unless the user opts out — and you have no way to see, let alone set, which tier your counterparty's junior is on.

The subtler harm is the confident wrong answer. A general-purpose model asked about a clause it can't find won't always say so — it will often produce a fluent, wrong paragraph, and a fluent wrong answer attributed to your documents is worse than no answer, because it looks authoritative and ends up in the deal record. This is the story we tell in full in stop clients from uploading your reports to ChatGPT and in should you connect ChatGPT to your data room: the fix is not a policy asking people not to upload, it is giving them somewhere better to ask.

The safe version keeps the documents, the permissions, the AI, and the audit trail inside the same gated room the parties already agreed to under NDA. The AI reads only the files in the room, answers are scoped to the viewer's permissions, every answer cites its source, and every query is logged next to every document view. The disclosure boundary doesn't move — the AI is part of the environment everyone signed the NDA for.

It turns serial diligence into parallel self-serve. Instead of you drafting the same working-capital or churn answer for every bidder over email, each reviewer asks inside the room and gets an answer grounded in the documents they're permissioned to see, around the clock, with the source cited — fewer duplicate email threads, and a question log that shows what every reviewer actually cares about. In practice, buyers and investors self-serve the large routine tail of diligence questions this way, so you spend your time on the judgment calls that need a human.

The Q&A capability runs deeper than this section covers. For the full treatment — the question log as a diligence signal, permission-scoping across bidder groups, the moderated human-approved workflow, and the plan-by-plan detail — read AI document Q&A: let diligence ask the documents. When you want a moderated workflow where questions route to your team, get an AI-drafted cited answer, and are reviewed and approved before release, that is Smart Q&A on the Data Room plan.

We run our own marketing the way we think deal documents should work. peony.ink publishes an llms.txt index and literal .md URL variants of its pages, so an AI assistant can read the site natively — as plain, structured text — instead of scraping rendered HTML and hoping it parses the layout correctly. It's the same principle as an agent-ready room: make the material machine-readable at the source, rather than trusting a model to reconstruct it from a page that was built for human eyes.

We're careful about scope here, because it's an easy claim to overstate. The llms.txt and .md practice is how we publish our public marketing website. It is not a claim that the Peony data-room product serves an llms.txt index for your customer rooms. Inside a room, the machine-readable path is different and permission-bound: it's AI document Q&A (Business, $30/admin/month) and AI auto-indexing (Data Room, $52/admin/month), where the AI answers from your documents, respects each viewer's permissions, cites the source, and logs every query. We do the open, native-readable thing on the public site; we do the gated, permission-aware thing on confidential rooms. The common thread is that in both places, the document is built to be read by a machine on purpose.