How it works in a data room
A user opens a document, or selects a folder, and asks for a summary. The system sends the text to a language model with instructions to extract the essentials, and returns a few paragraphs or bullet points, ideally with page references. Some tools let users pick a template, such as “lease summary” or “employment contract key terms”. Summaries respect the user’s permissions, are usually marked as AI-generated and may be saved as notes beside the document rather than replacing it.
Why it matters in a deal
Reviewers face thousands of files under tight deadlines. Summaries help them triage: which leases have unusual break clauses, which supply contracts mention exclusivity, which board minutes discuss litigation. They do not replace reading. A model can omit a crucial clause or state something the document does not say, so findings that affect price or liability should be checked against the source text. See hallucination.
Example
A buyer’s legal team has two days to review 600 commercial leases. They run summaries on all of them, sort by notice periods and rent review terms, and spend their reading time on the 70 leases with unusual provisions. A spot check against the full text finds two summaries that missed a side letter, so the team reads every side letter in full.