Yes, for the first pass. AI can sort a data room by document type, pull the clauses your checklist asks about, flag change of control, assignment, exclusivity and termination terms, and link every finding to the page it came from. Lawyers still verify each finding, judge what matters for the deal and write the report. This is not legal advice.
#Key takeaways
- AI is useful for the mechanical first pass: sorting, extracting and flagging against a written checklist.
- Every finding must point to the exact page and sentence, or a reviewer cannot verify it.
- Legal AI tools still produce wrong or misgrounded answers, so verification is part of the workflow, not an extra.
- Measure recall on a sample a lawyer has reviewed by hand before trusting the first pass on a live deal.
- For a small data room with clean English documents, a commercial review tool or careful manual review is enough.
#What does first-pass due diligence look like today?
The data room opens with hundreds or thousands of files: scanned leases, supplier agreements, employment contracts, board minutes, licences, some in Arabic, some as photos of signed pages. Junior lawyers or paralegals open each one, note the parties, term, renewal, change of control and assignment, and fill a spreadsheet the partner will turn into the report.
The reading is repetitive, the checklist is the same on every deal, and the risk is the one document nobody opened because it was misfiled under "other".
#What can AI do reliably in a data room?
- Classify each document by type and counterparty, including scans after text recognition.
- Extract the fields on your checklist: parties, dates, term, renewal, governing law, notice addresses.
- Flag the clauses that matter for this deal type, such as change of control, assignment, exclusivity and termination for convenience.
- List documents it could not read or classify, so a person opens them first.
- Draft the findings table, with a link to the source page for every row.
Document review is the top use of generative AI that legal professionals report (Thomson Reuters, 2025). The first pass is where that reading load sits.
#Why must every finding be verified by a lawyer?
Legal AI tools still get things wrong. A Stanford study found leading legal research tools produced incorrect or misgrounded answers in more than one in six benchmark queries, where misgrounded means the cited source does not support the statement (Stanford HAI, may 2024). In diligence, the equivalent is a clause summary that points to the wrong page.
Vendors say the same: one legal software company's due diligence guide calls the obligation to validate AI output non-negotiable (Clio, checked november 2026). Design for it: every row links to its source, and a reviewer marks it confirmed or corrected.
#How do you run an AI first pass safely?
- Write the checklist for the deal type before uploading anything.
- Run the first pass on a closed past deal where your team already knows the answers, and count what it missed.
- Agree what the AI output is: a draft findings table, never a conclusion.
- Route unreadable, handwritten and mixed-language documents to a person first.
- Have a lawyer confirm or correct every flagged row, and sample unflagged documents to check what was missed.
- Keep the data inside the hosting the engagement letter and the seller's data room terms allow.
#When is a commercial tool or manual review enough?
For a small data room of clean English contracts, careful manual review with a good checklist is often faster than setting anything up. If your firm already licenses a legal AI platform with data room review, test it on a closed deal before building anything. A custom workflow fits when deals repeat the same checklist, documents are Arabic and English scans, and findings must land in your own templates and systems.
#Where does Stennir fit in due diligence?
Stennir's AI consulting audit compares diligence with your other document-heavy work and checks whether the volume and repetition justify a build. With AI implementation, Stennir builds the first-pass step on your accounts, in Arabic and English, hosted where your rules and the data room terms require, with a named lawyer confirming every finding and each action logged.
AI training teaches associates and paralegals to check source links, spot misgrounded findings and feed corrections back. For the same pattern on single contracts, see contract review with a full audit trail.
We are
Stennir builds a first pass that links every finding to its source page and leaves the conclusions with your lawyers.
We aren't
Stennir is not a law firm and does not give diligence opinions or sign off on findings.
#What should a deal lead do next?
Pick a closed deal, list the checklist items your team extracted and test any tool against that answer key. See how legal and compliance teams use AI on AI for compliance, take the free readiness assessment or book a 30-minute call.
#Common questions
Can we upload a seller's data room into an AI tool?
Only if the data room terms, the engagement letter and your confidentiality obligations allow it, and only into a tool hosted where those terms permit. Check with the responsible partner first. Stennir decides the hosting before anything is built, so documents never pass through a service your terms exclude.
Does AI due diligence work on Arabic documents?
It can, but test it. Arabic scans, stamps and handwritten annotations reduce text recognition quality, and legal Arabic uses terms general models misread. Stennir tests extraction on your own Arabic documents first and routes anything below the agreed confidence to a person.
What do you need from us to set up a first-pass review?
Your diligence checklist for one deal type, a closed past deal with its final findings as an answer key, and a lawyer who can judge the output. Stennir uses that answer key to measure what the first pass finds and misses before it touches a live deal.
What could go wrong with AI first-pass diligence?
A misfiled document can be skipped, a clause can be summarised from the wrong page, and reviewers can stop checking once the output looks good. The safeguards are source links on every finding, a list of unread documents, sampling of unflagged files and a lawyer confirming each row.
