Verified against ChatGPT · 2026-08-09
Build a signing checklist for an NDA that flags one-sided terms before you agree to it
Walks through a draft NDA against a plain checklist of common one-sided terms — scope, duration, mutuality, carve-outs — and produces a go/hold list, framed as prep for a lawyer's review rather than a substitute for one.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Check the NDA below against the standard points that make an NDA either reasonable or one-sided, and produce a checklist I can use before deciding whether to sign or push back. NDA TEXT Two-page mutual NDA from a prospective client before a scoping call. WHO'S ASKING ME TO SIGN A potential client we'd be pitching a paid engagement to; not an employer or investor. WHAT INFORMATION IS ACTUALLY AT STAKE Our pricing model and a rough project approach; no source code or trade secrets involved. CHECK EACH OF THESE AGAINST THE TEXT - Is the confidentiality obligation mutual (both sides protect each other's information) or one-way, and does that match the actual relationship described above? - How long does the confidentiality obligation last, and is that duration reasonable for the type of information at stake, or unusually long (e.g., indefinite, or 10+ years for information that won't stay sensitive that long)? - Is the definition of "confidential information" narrow and specific, or so broad it could cover information that was already public or that I already knew? - Are there standard carve-outs (information that becomes public through no fault of mine, information I already had, information I develop independently, information I'm legally compelled to disclose)? Flag if any of these is missing. - Does the NDA try to restrict anything beyond confidentiality — non-compete language, IP assignment, or non-solicitation quietly folded in under an NDA label? - What governing law and dispute resolution process does it specify, and does that match where I actually operate? For each point, state plainly what the text actually says, then mark it Standard, Worth a Question, or Push Back — do not assign a legal severity beyond that three-way flag, and do not guess at how enforceable a given term would be in any specific jurisdiction. OUTPUT FORMAT 1. A six-row checklist table (Point | What the text says | Flag). 2. A short list of any items marked Push Back with the specific plain-language reason. 3. A closing note: this checklist is a preparation aid to help you read the NDA critically before a conversation with a lawyer, not a legal opinion on the document — anything marked Push Back, and anything involving non-compete or IP language folded into the NDA, should go to a qualified lawyer before you sign.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
NDAs are short enough that a model asked to "review" one in the abstract will often produce a generic list of NDA concepts rather than checking this specific document against them, because nothing in an open-ended request forces a claim-by-claim comparison; giving GPT-5.1 a fixed six-point checklist and instructing it to state what the text actually says before assigning a flag converts a vague review into a verification task, which is a much more reliable mode for a language model than open-ended risk judgment. Requiring the counterparty relationship and the actual information at stake as inputs is what lets the mutuality and duration checks mean anything — a one-way NDA is unremarkable when you're the one receiving someone else's trade secrets and alarming when the relationship is peer-to-peer, and a model without that context has no way to tell which situation it's looking at. Naming the specific carve-outs to check for (public-domain information, prior knowledge, independent development, legal compulsion) matters because their absence is a silent risk — a clause that isn't there doesn't announce itself, and a model scanning for "problems" tends to notice what's present and overlook what's conspicuously missing unless it's told exactly what to look for. The three-way flag system (Standard / Worth a Question / Push Back) keeps the output actionable without tipping into a legal severity rating the model has no basis to assign, and restricting the enforceability question entirely keeps the checklist a reading aid rather than something that could be mistaken for legal sign-off on a document with real confidentiality exposure.
What you get back
Mutuality: text says obligations apply to "both parties" — Standard, matches the relationship. Duration: confidentiality survives 7 years after termination — Worth a Question, longer than typical for pricing/approach info, ask if it can be reduced to 2-3 years. Carve-outs: missing an independent-development carve-out — Push Back, add standard language. Closing: this is a preparation checklist, not a legal opinion — take the Push Back item to a lawyer before signing.
Verified against
ChatGPT GPT-5.1 · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
