25 AI Prompts for Reviewing Contracts (With Legal Caveats)

TL;DR: These 25 prompts get AI to do what it's actually good at with a contract: plain-English summaries, lawyer-ready question lists, red-flag spotting, checklists, and side-by-side comparisons. None of them ask it to say whether a clause is enforceable or whether you should sign. Read the confidentiality note before you paste anything.

Can You Paste a Contract Into ChatGPT or Claude Without Breaking Confidentiality?

Ask this before you ask anything else, because it's the one mistake here you can't undo. Most NDAs, client agreements, and employment contracts define confidential information broadly, and "broadly" usually includes handing the document to a third-party AI tool, whether or not that tool trains on what you paste. Check the specific document, not your general instinct about what counts as sharing.

Even a vendor that doesn't train on your conversations by default isn't a black box with nobody behind it. Anthropic's own privacy documentation for Claude's consumer plans states that conversations are used for training only if you opt in, but adds a second condition: "conversations are flagged for safety review (in which case we may use or analyze them to improve our ability to detect and enforce our Usage Policy, including training models for use by our Safeguards team, consistent with Anthropic’s safety mission)" (Anthropic Privacy Center, "Is my data used for model training?", accessed September 3, 2026). That's a narrow, safety-triggered case, not routine reading of your contract. It is still a real answer to "who could ever see this," and "nobody" isn't it.

The practical version of that answer matters more than the policy language: a safety review is triggered by content in the conversation, not by whether you meant to share it, so a pasted contract that happens to mention something the classifier reads as concerning (a threat, a dispute, a safety issue) can get a second set of eyes regardless of intent. That's one more reason to excerpt rather than paste whole documents by default.

The safe version of every prompt below works the same way: strip names, dollar figures, and anything that would identify the parties before you paste, and describe the clause type instead ("a non-compete clause," "the indemnification section") rather than posting the whole document when a smaller excerpt does the job. If you genuinely cannot tell whether a document is safe to paste anywhere, that question itself belongs on the list you send your lawyer.

What Can These Prompts Actually Do, and What Won't They Do?

Every prompt on this page sits firmly on one side of this line. Read it once before you use any of them, because the honest answer to "can AI review my contract" is "yes, for six specific jobs, and no for everything that requires a legal opinion."

Keep this table open in another tab while you work through the prompts below. Every single one of them is written to stay on the left column on purpose, and a prompt that starts drifting toward the right column is a sign to stop and rephrase it, not a sign the model finally got helpful.

How Do You Get a Plain-English Summary of a Clause?

The point of a summary prompt is translation, not judgment: it takes dense contract language and restates what it does, so you know what you're actually looking at before deciding whether it needs a closer read. A good summary tells you the mechanism. It should never tell you whether that mechanism is fair.

1. Plain-English summary of one clause

Role: a plain-language explainer, not a lawyer.
Task: summarize what the clause below actually does, in the fewest words that keep it accurate.
Clause: [PASTE THE EXACT CLAUSE TEXT]
Format: 3-5 sentences, no legal jargon without a one-line explanation next to it.
Constraints: describe only what the clause says; do not state whether it is fair, standard,
or enforceable, and do not guess at intent that isn't in the text.
Tone: neutral, clear, written for someone with no legal background.

What to change: paste the exact clause, not a paraphrase of it. A summary of your own summary compounds whatever you got wrong the first time.

2. Whole-document plain-English overview

Role: a plain-language explainer summarizing a full contract section by section.
Task: read the document below and summarize each numbered section in one to two sentences.
Document: [PASTE THE CONTRACT TEXT]
Format: one line per section, in the document's own section order and numbering.
Constraints: quote the section number and heading exactly as written; if a section is unclear
or ambiguous, say so instead of guessing at what it probably means.
Tone: neutral, descriptive.

What to change: for a long contract, run this in chunks by section rather than pasting the whole thing at once. A model summarizing 40 sections in one pass tends to compress the last ones harder than the first.

3. "Explain this like I'm not a lawyer" for one confusing sentence

Role: a plain-language explainer.
Task: rewrite the single sentence below in plain English, keeping every condition it states.
Sentence: [PASTE THE EXACT SENTENCE]
Format: one plain-English sentence, then a one-line note on any condition you had to drop
to make it readable (if any).
Constraints: do not drop a condition silently — a "plain English" version that quietly loses
an exception is worse than the confusing original.
Tone: direct, precise.

What to change: this one is worth running on any sentence you've read three times and still can't restate to yourself. If the plain-English version still doesn't make sense, that's a sign the sentence itself is genuinely ambiguous, which belongs on your lawyer question list below rather than something to keep re-reading alone.

How Do You Turn a Contract Into Questions for Your Lawyer?

This is the single most useful thing AI does with a contract: it turns a document you don't fully understand into a short list of specific questions, so the twenty minutes you get with a lawyer goes to the parts that actually need judgment instead of the parts a plain-English pass already answered.

4. Generate a lawyer question list from a full contract

Role: someone preparing for a paid consultation with a lawyer about the contract below.
Task: read the document and generate a list of specific questions to ask a lawyer about it —
not answers, questions.
Document: [PASTE THE CONTRACT TEXT]
Format: numbered list, each question naming the specific section or clause it's about.
Constraints: do not answer any of the questions yourself, even partially; if something looks
concerning, phrase it as a question ("does this section mean X?") rather than a conclusion.
Tone: direct, specific, no hedging language padding out each question.

What to change: if you already know the two or three clauses you're worried about, paste just those instead of the whole document. A shorter, sharper question list is more useful in a paid consultation than a long one.

5. Turn a vague worry into a specific question

Role: someone who has a vague concern about a contract clause but can't articulate it.
Task: read the clause below and my worry, and turn it into 2-3 specific, answerable questions
a lawyer could respond to directly.
Clause: [PASTE THE CLAUSE]
My vague worry: [DESCRIBE WHAT'S BOTHERING YOU, EVEN IMPRECISELY]
Format: numbered questions, each one sentence.
Constraints: do not resolve my worry yourself; sharpen it into a question instead.
Tone: precise, no reassurance, no opinion on whether the worry is justified.

What to change: describe the worry honestly, including the parts that feel silly or half-formed. A vague worry that gets smoothed over before you type it usually produces a smoothed-over question that misses the actual concern.

6. Prioritize a long question list before a short consultation

Role: someone with 15 questions about a contract and a 20-minute call with a lawyer.
Task: given the question list below, rank the questions by how much they'd change my decision
to sign, most important first.
Questions: [PASTE YOUR LIST]
Format: numbered, most important first, one-line reason for each ranking.
Constraints: do not remove any question, only reorder them; do not answer any of them.
Tone: blunt, prioritization only.

How Do You Flag a Clause That Looks Unusual or One-Sided, Without Calling It Illegal?

Flagging is different from judging. A flag prompt's job is to say "this is not what you'd typically see in this kind of agreement, and here's specifically why that's worth a second look," never "this clause is illegal" or "this won't hold up." The second sentence in every flag matters as much as the first: a flag with no reason attached is just an alarm nobody can act on.

7. Flag anything that looks one-sided

Role: someone comparing this contract against what's typical for a [CONTRACT TYPE, e.g.
freelance services agreement].
Task: read the document and flag any clause that looks unusually one-sided, and explain
specifically why it stands out — what it would typically look like instead, in general terms.
Document: [PASTE THE CONTRACT TEXT]
Format: for each flagged clause: the clause name, one sentence on why it's unusual, and
"worth asking a lawyer about" as the closing line every time.
Constraints: never state that a flagged clause is unenforceable, illegal, or invalid — only
that it departs from what's typical and warrants a specific question.
Tone: observational, not alarmist.

What to change: name the contract type as specifically as you can. "One-sided for a freelance agreement" and "one-sided for an enterprise vendor contract" are different baselines, and the prompt needs to know which one you mean.

8. Compare one clause against what's typical, without a verdict

Role: someone checking whether a specific clause is typical for a [CONTRACT TYPE].
Task: describe, in general terms, what this type of clause usually covers in this kind of
agreement, then note anything about the clause below that differs from that general pattern.
Clause: [PASTE THE CLAUSE]
Format: two short paragraphs — "typically" and "this one specifically."
Constraints: describe general patterns only, not a specific jurisdiction's law; end with a
question to ask a lawyer, not a conclusion about whether the difference matters.
Tone: measured, comparative.

What to change: run this on the two or three clauses that made you pause on a first read, not on every clause in the document. Flagging everything as "worth a second look" is functionally the same as flagging nothing.

9. Flag missing protections a contract of this type usually has

Role: someone checking a contract for what's absent, not just what's present.
Task: given the document below and the contract type, list anything a document of this type
commonly includes that this one appears to be missing.
Document: [PASTE THE CONTRACT TEXT]
Contract type: [e.g. "freelance services agreement", "residential lease"]
Format: bulleted list, each item naming what's commonly present and noting its absence here.
Constraints: "commonly includes" describes general patterns, not a legal requirement in any
specific jurisdiction; say so if asked to be more specific than that.
Tone: neutral, checklist-style.

What to change: an absence is not automatically a problem. Plenty of legitimate agreements deliberately leave out a clause that's standard elsewhere because the deal doesn't need it. Treat every result from this prompt as a question to ask, not a defect to assume.

What Should an NDA, Offer Letter, or Vendor Agreement Typically Contain?

A checklist prompt answers a different question than a flag prompt: not "what's wrong with this one" but "what does this whole category of document usually cover," so you can see at a glance what's present, what's absent, and what you should ask about before you assume an absence is an oversight rather than a deliberate choice.

10. Build a checklist for a contract type

Role: someone building a general checklist of what a [CONTRACT TYPE] typically contains.
Task: list the sections or clauses that documents of this type commonly include, described
in general terms, not tied to any specific jurisdiction's requirements.
Contract type: [e.g. "mutual NDA", "employment offer letter", "SaaS vendor agreement"]
Format: bulleted list, one line per typical section, with a one-clause note on what it usually
covers.
Constraints: state clearly that this describes common practice, not a legal requirement, and
that requirements vary by jurisdiction and negotiated deal.
Tone: neutral, reference-style.

What to change: run this once per contract type you deal with regularly and save the output. A founder signing NDAs, vendor agreements, and offer letters in the same month benefits from three separate checklists, not one generic one.

11. Check a real document against the checklist

Role: someone comparing a real contract against a general checklist for its type.
Task: given the checklist below and the document below it, mark each checklist item as
present, absent, or unclear in the document, quoting the relevant section where present.
Checklist: [PASTE THE CHECKLIST FROM THE PROMPT ABOVE]
Document: [PASTE THE CONTRACT TEXT]
Format: a row per checklist item, quoted section or "not found."
Constraints: mark "unclear" rather than guessing when a section is ambiguous; do not infer
that an absent item was intentionally left out.
Tone: neutral, factual.

What to change: regenerate the checklist for the specific contract type first, rather than reusing a checklist built for a different kind of agreement. A checklist built for a lease will misjudge a services agreement on almost every row.

12. Checklist for a document you're about to send someone else

Role: someone about to send a [CONTRACT TYPE] to another party and wanting a self-check first.
Task: using the general checklist below, note anything the draft below is missing or
underspecified before it goes out.
Checklist: [PASTE OR REGENERATE A CHECKLIST FOR THIS CONTRACT TYPE]
Draft: [PASTE YOUR DRAFT]
Format: bulleted list, one line per gap found.
Constraints: flag gaps only; do not draft replacement clause language to fill them.
Tone: practical, pre-send review.

What to change: run this before you send, not after the other party has already commented. A gap you catch yourself costs nothing; the same gap pointed out by the other side reads as sloppiness.

13. NDA-specific: what a mutual NDA typically defines

Role: someone checking a mutual NDA for its core defined terms.
Task: list what the document below defines for "confidential information," the disclosure
obligations, the term of confidentiality, and any carve-outs, quoting each definition exactly.
Document: [PASTE THE NDA TEXT]
Format: four short sections (definition, obligations, term, carve-outs), each with the
exact quoted text it's based on.
Constraints: if any of the four is missing or vague, say so rather than filling the gap
with what an NDA "usually" says.
Tone: precise, definitional.

What to change: pay particular attention to the carve-outs section. What an NDA explicitly excludes from confidentiality (information already public, already known, independently developed) matters as much as what it protects, and it's the section most likely to be thin or missing entirely in a template pulled off the internet.

How Do You Extract Dates, Parties, and Obligations Into a Table?

Extraction prompts turn a contract into something you can actually scan: who the parties are, what each one owes the other, and when. This is the most mechanical of the six jobs in the table above, and also the one where a wrong extraction is easiest to catch, because you can check every cell against the document in seconds.

14. Parties, dates, and key terms into one table

Role: a document analyst pulling structured facts out of a contract.
Task: extract the parties, effective date, term length, renewal terms, and governing law
clause (if stated) from the document below.
Document: [PASTE THE CONTRACT TEXT]
Format: a table — field, value, and the exact quoted source text for each value.
Constraints: leave a field blank and say "not stated" rather than inferring a value that
isn't explicitly in the document.
Tone: factual, source-quoted.

15. Every obligation, who owes it, and when it's due

Role: a document analyst listing every obligation in a contract.
Task: extract every obligation either party has under this contract — what, who owes it,
and any deadline or trigger condition.
Document: [PASTE THE CONTRACT TEXT]
Format: a table — obligation, owed by, deadline/trigger, source section.
Constraints: include an obligation only if it's explicitly stated; do not infer an implied
obligation that isn't written down.
Tone: exhaustive, factual.

16. Payment terms and amounts into a table

Role: a document analyst extracting every payment-related term.
Task: extract every dollar amount, payment schedule, late-fee provision, and refund or
cancellation term stated in the contract below.
Document: [PASTE THE CONTRACT TEXT]
Format: a table — term type, amount/condition, source section quoted exactly.
Constraints: quote figures exactly as written, including currency; do not do arithmetic
on the figures or estimate a total.
Tone: exact, numeric.

How Do You Compare Two Versions of the Same Contract?

Comparing a redline against the original is where cross-document recall earns its keep: instead of you scanning two documents side by side for what moved, the model states every change and you check the ones that matter. Our own ChatGPT vs. Claude comparison names this exact task as a place where Claude's recall of details buried in a long document tends to hold up better across a full analysis, which is worth knowing before you pick which one to paste two long contract versions into.

17. What changed between two versions

Role: a document analyst comparing two versions of the same contract.
Task: list every substantive difference between version A and version B below — added,
removed, or reworded clauses that change meaning. Ignore pure formatting differences.
Version A: [PASTE FIRST VERSION]
Version B: [PASTE SECOND VERSION]
Format: a table — clause/section, what changed, quoted text from both versions.
Constraints: do not judge whether a change favors either party; describe the change only.
Tone: neutral, comparative.

What to change: if the two versions are long, run this section by section rather than end to end. A model comparing two forty-page documents in one pass will find the obvious changes and quietly miss a small reworded clause buried in the middle.

18. Which changes shift risk, without saying which side benefits

Role: a document analyst flagging changes worth a closer look.
Task: given the change list below, flag which changes look substantive enough to warrant
a specific question to a lawyer, versus which look like minor wording cleanup.
Change list: [PASTE THE OUTPUT FROM THE PROMPT ABOVE]
Format: two groups — "worth a question" and "likely just wording" — with a one-line reason
for each item's group.
Constraints: do not state who the change benefits; note only that it changes the meaning.
Tone: careful, non-committal on outcome.

19. Summarize a redline for someone who hasn't read either version

Role: someone briefing a colleague who hasn't seen either contract version.
Task: given the change list below, write a short summary of what changed overall, for
someone who needs the gist before a meeting, not every individual clause.
Change list: [PASTE THE CHANGE LIST]
Format: one paragraph, under 150 words.
Constraints: mention only changes already identified above; do not add new ones or
speculate about ones you weren't given.
Tone: brief, meeting-ready.

What to change: keep this to what actually changed, and resist the urge to add your own recommendation into the same summary. A colleague reading it before a meeting needs the facts first; your opinion belongs in the meeting, not folded silently into a status update.

What About Termination, Renewal, and Notice Periods?

These clauses are exactly where "extract the fact" and "state the law" get confused most often, because a notice period reads like a fixed rule and often isn't one. What counts as sufficient notice, whether an auto-renewal clause is enforceable as written, and how a termination-for-cause provision actually plays out all vary by jurisdiction and change over time. The job here is extraction and flagging, never a verdict on what the rule requires.

20. Extract every termination and renewal condition

Role: a document analyst extracting termination and renewal terms.
Task: extract every condition under which either party can terminate this contract, the
notice period required for each, and any auto-renewal terms.
Document: [PASTE THE CONTRACT TEXT]
Format: a table — condition, notice required, source section quoted exactly.
Constraints: quote the stated notice period exactly; do not state whether that period is
legally sufficient. That varies by jurisdiction, and this prompt does not know yours.
Tone: factual, source-quoted.

What to change: put the extracted notice periods straight into a calendar reminder set well before the deadline, not on the deadline itself. A missed cancellation window is usually a missed calendar entry, not a missed legal argument.

21. Flag an auto-renewal clause worth a second look

Role: someone checking whether an auto-renewal clause needs a calendar reminder or a lawyer's
opinion.
Task: describe exactly how the auto-renewal clause below works — what triggers it, how much
notice cancels it, and what happens if that window is missed.
Clause: [PASTE THE CLAUSE]
Format: 3-4 sentences, mechanism only.
Constraints: do not state whether the clause is enforceable or standard for your jurisdiction;
end by suggesting a calendar reminder for the cancellation window, since that's a practical
step, not a legal opinion.
Tone: practical, mechanism-focused.

How Do You Catch AI Hallucinating a Clause That Isn't There?

This is not a hypothetical risk you're guarding against out of caution. In 2023, a federal court case's own docket record describes its holding plainly: attorneys were "sanctioned for using fake case law citations generated by ChatGPT." Neither citation had been checked before it was filed (Mata v. Avianca, Inc., S.D.N.Y., decided June 22, 2023). The same failure mode applies to a contract clause: a model can hallucinate a provision confidently and be entirely wrong about what it says, or claim a document contains something it doesn't. The fix isn't more caution about phrasing. It's insisting every claim comes with the exact source text attached, every time, without exception.

22. Force every claim to cite its source

Role: a document analyst who must show its work.
Task: for every claim you make about this contract, quote the exact sentence it comes from.
If you cannot find a direct quote supporting a claim, say "not directly supported in the
document" instead of the claim.
Document: [PASTE THE CONTRACT TEXT]
Question: [YOUR SPECIFIC QUESTION ABOUT THE DOCUMENT]
Format: answer, then "Source:" followed by the exact quoted text for every factual claim.
Constraints: never present an inference as if it were a direct quote.
Tone: precise, source-first.

What to change: run this even on questions you think you already know the answer to. The whole value of the prompt is exposing the gap between what you assumed the contract said and what it actually says, and that gap is invisible until you force the citation.

23. Self-audit a previous answer for unsupported claims

Role: a reviewer checking a previous AI answer against the source document.
Task: given the answer below and the document it was about, mark every sentence as either
directly supported by a quote from the document, a reasonable inference, or unsupported.
Previous answer: [PASTE THE EARLIER OUTPUT]
Document: [PASTE THE CONTRACT TEXT]
Format: the answer broken into sentences, each tagged [SUPPORTED] / [INFERENCE] / [UNSUPPORTED]
with the source quote where supported.
Constraints: be stricter about "supported" than feels natural; a paraphrase that changes
meaning even slightly does not count.
Tone: skeptical, auditor-style. For more on why the same self-check habit matters across any
high-stakes AI output, not just contracts, see [red-teaming your own prompt](/blog/224-red-team-your-own-prompt-before-you-trust-the-output).

Should You Ever Let AI Draft the Actual Contract Language?

Not as something you paste in and sign. A model turning your position into a negotiation talking point or a redline comment is a discussion draft, and the discussion is exactly where a human is supposed to stay in the loop. Whether specific wording actually protects you, closes a loophole, or holds up if it's ever tested is a drafting judgment that belongs to a lawyer, not a prompt that has never seen the rest of the document, your jurisdiction, or the deal you're actually trying to make.

24. Turn a negotiating position into talking points, not final language

Role: someone preparing to discuss a clause with the other party, not draft final wording.
Task: turn my position below into 3-4 talking points I could raise in a conversation or email
about this clause — not a replacement clause, just what to say and why.
Clause: [PASTE THE CLAUSE]
My position: [WHAT YOU WANT CHANGED AND WHY, IN YOUR OWN WORDS]
Format: bulleted talking points, each one sentence.
Constraints: do not draft replacement clause wording — talking points only. Final language
is a lawyer's job, not this prompt's.
Tone: assertive but reasonable, negotiation-ready.

What to change: state your position honestly, including "I don't know if this is standard, it just worries me." The prompt turns a real, even unpolished concern into something sayable out loud; it can't manufacture a concern you don't actually have.

25. Draft a redline comment explaining a proposed change

Role: someone commenting on a contract redraft, not finalizing it.
Task: write a short comment explaining what I'd want changed in the clause below and why,
in the style of a tracked-changes comment bubble, not a rewritten clause.
Clause: [PASTE THE CLAUSE]
What concerns me: [YOUR CONCERN, IN YOUR OWN WORDS]
Format: 2-3 sentences, comment style ("Suggest revisiting this — ...").
Constraints: explain the concern; do not supply replacement contract language.
Tone: collegial, specific, not adversarial.

What to change: send this as one comment among several, not as the entire redline. A single bubble reading like a legal memo stands out from the rest of the document and reads as more adversarial than it means to be.

Does the Free Plan Cover This?

Yes, for occasional use, with a caveat worth repeating: every prompt above is plain text that runs in ChatGPT, Claude, or Gemini exactly as written, with no extra tool required. Prompt Architects is an optional layer that turns the role/task/format/constraints/tone structure each prompt uses into a sidebar you fill in, inside the chat you already have open, instead of retyping the whole block from a notes app each time.

The Free plan includes 5 architected prompt generations a day, forever, no card required, per the FAQ page, which is enough to run a handful of these on a document that needs a few different passes. What Free doesn't include: saving your own versions of these prompts as reusable templates with tags and search, or a personal context library where the contract types you deal with most (freelance agreements, vendor contracts, whatever's actually on your desk) would sit once instead of you retyping the same brackets every time. Those are Pro-plan features, priced at the time of writing on /pricing; check the current figure there, since it's running a launch discount that won't last.

None of this replaces a lawyer, and nothing here was built to make it feel like it does. What it replaces is showing up to a paid consultation having not read the document at all, or signing something because reading it felt like more effort than the decision seemed to deserve. Read it, question it, flag what's unusual, and bring the specific, sharpened list to the person whose job is actually to answer it.