Write your NDA positions down as a clause-by-clause playbook, let AI compare each incoming NDA against it, and route the result: standard paper goes to a contract manager to sign off, deviations go to a lawyer with the clause, the rule it breaks and a suggested redline. The lawyer keeps the decision. This is not legal advice.
#Key takeaways
- NDA triage fails without a written playbook; AI only makes an unwritten one faster and less consistent.
- The useful output is a routing decision plus the clauses that deviate, not a summary of the whole document.
- A named person approves every NDA. AI decides which person, and what they should read first.
- Measure the share of NDAs that reach a lawyer and how often the lawyer overrules a flag.
- If you see only a handful of NDAs, a good template and a checklist are enough.
#What does NDA review look like for an in-house team today?
Sales wants to talk to a prospect, procurement wants to share a spec, and the counterparty sends its own paper. The request lands in a shared inbox or a teams chat with "urgent" in the subject. Someone opens the word file, reads it top to bottom, and checks the same things every time: mutual or one-way, the definition of confidential information, the term, residuals, non-solicit, governing law.
Most of these NDAs are close to standard. The cost is a qualified lawyer reading easy ones while the business waits. In the Gulf the same document often arrives in Arabic and English, which doubles the reading.
#Why does AI NDA triage need a playbook first?
A model can read an NDA, but it cannot know that your company accepts a residuals clause for vendors and never for customers. That knowledge sits in a senior lawyer's head. Until it is written down as positions, fallbacks and walk-away points per clause, AI will produce confident opinions that differ from document to document.
Adoption is rising: a survey of in-house teams found 52% already using or evaluating AI for contract review (LegalOn via Business Wire, january 2026). Document review is also the top use of generative AI that legal professionals report (Thomson Reuters, 2025). The teams that get value write the playbook before they switch anything on.
#What goes into an NDA playbook you can use without any tool?
- List the clauses you check on every NDA, in the order you check them.
- For each clause, write your standard position, one or two acceptable fallbacks and the point where it goes to a lawyer.
- Add the routing rule: standard and mutual goes to a contract manager; any fallback goes to a lawyer; anything past a walk-away point goes to the general counsel.
- Keep a clause library: approved wording in English and Arabic for each position, with a note on why.
- Log every NDA, who approved it and which clauses moved. That log is your test set later.
#What should AI do, and what stays with the lawyer?
Let AI do the reading and the sorting: identify the parties and direction, extract each clause, compare it to your playbook, label it standard, fallback or outside, and draft a redline from your clause library. It should cite the exact sentence it relied on, so the reviewer can check in one look.
Keep with people: approving any NDA, accepting a fallback, judging anything the playbook does not cover, and talking to the counterparty. Track how often reviewers overrule a flag. A rising overrule rate means the playbook or the prompt needs work, not that people should stop checking.
#When is a template or an off-the-shelf tool enough?
If your team sees only a handful of NDAs, a one-page playbook and a standard template you push first will do more than any AI. If your NDAs arrive only in English, sit in one contract system and follow your paper, the AI review feature in a contract management tool you already pay for may be enough. Try it on your last fifty NDAs first.
A custom workflow earns its place when volume is steady, the paper is the counterparty's, documents arrive in Arabic and English, and the result must land in your own matter log with an audit trail.
#How does Stennir run NDA triage?
Stennir starts with AI consulting: the AI opportunity audit checks whether NDA triage is the right first workflow against your other legal and compliance work, and turns your senior lawyer's habits into the written playbook. With AI implementation, Stennir builds the triage step on your own accounts, in Arabic and English, hosted where your rules require, with a named reviewer approving each NDA and every decision logged.
AI training then teaches contract managers and lawyers to read the flags, overrule them and feed corrections back. We showed the audit trail side in contract review with a full audit trail.
We are
Stennir writes your playbook with your lawyers, builds the triage step around it and leaves every approval with a named person on your team.
We aren't
Stennir is not a law firm, gives no legal advice and does not sign or send anything on your behalf.
#What should a general counsel do next?
Pull your last fifty NDAs and sort them by hand into standard, fallback and outside. That split tells you whether triage is worth building. 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
What do you need from us to set up NDA triage?
A sample of past NDAs with the final signed versions, the person who knows your positions best, and your current template. From those, Stennir drafts the playbook for your lawyers to correct. We also need to know where documents may be stored and processed, because that decides the hosting.
Does AI NDA review work in Arabic?
Yes, when it is tested on your own Arabic documents. Stennir checks extraction and clause matching on real Arabic and bilingual NDAs before anything goes to users, and the clause library holds approved Arabic wording. If the two language versions differ, the step flags it for a lawyer instead of choosing one.
Who is responsible if the AI misses a bad clause?
Your team is, because a named person approves every NDA. That is why the design shows the clause and the rule next to each flag, logs every decision, and samples standard NDAs for a lawyer to spot check. The overrule rate tells you whether the triage is still trustworthy.
What could go wrong with AI NDA triage?
The playbook can be incomplete, a counterparty can hide an unusual obligation under a standard heading, and staff can start approving flags without reading them. The fixes are a fallback rule that sends anything unmatched to a lawyer, spot checks on standard NDAs, and tracking how often reviewers overrule the AI.
