Can I Use ChatGPT to Review a Contract? (And How to Do It Safely)

Yes, you can use ChatGPT to review a contract, and for a first pass it is genuinely useful. What it cannot do is replace a lawyer's judgement on whether a specific clause is enforceable in your situation. This guide is the honest middle: how to prompt it so the output is checkable, the specific ways it gets Indian contracts wrong, and what happens to the text after you paste it in. (Adira, which publishes this guide, sells contract review and CLM software, so we have a commercial reason to want you comfortable with AI-assisted review. Nothing below requires buying anything; the workflow works with ChatGPT alone.)

The short answer: yes for triage, no for legal advice

Treat ChatGPT as a fast reader, not a lawyer. It is good at surfacing what a long, unfamiliar contract actually says, restating dense clauses in plain English, and flagging where a document looks unusual for its type. It is not good at telling you whether a clause will hold up if challenged, because that depends on facts it cannot see: your leverage in the deal, wording a court has already ruled on, and Indian statutory overrides it was not reliably trained on. Use it to understand faster, not as the final word on whether to sign.

What ChatGPT is actually good at when you paste in a contract

Three things, if you prompt it properly. Summarising: turning a 12-page vendor agreement into a clause-by-clause plain-English list. Comparing against a stated standard: tell it your position ("liability capped at 12 months' fees, 30-day payment terms") and it scans for deviations far faster than a manual read. Drafting suggestions: once a weak clause is identified, it can draft two or three alternatives to choose from or send to a lawyer for a final check. What it is not reliably good at is knowing, unprompted, which of these clauses actually matters under Indian law, the subject of most of this page.

How to prompt it so the output is checkable

A vague prompt gets a vague, overconfident answer. "Is this contract okay?" invites ChatGPT to guess what "okay" means for you and answer anyway, because refusing feels unhelpful. A structured prompt fixes most of this.

Go clause by clause or category by category (parties, term, payment, liability, indemnity, IP, termination, dispute resolution), and for each ask three things: what does it say in plain English, what does it require of each side, and does it deviate from what is normal for this contract type. Then add two instructions that do most of the work: require an exact-sentence quote behind every point it raises, with the clause number if there is one, and state explicitly that Indian law governs, so it stops defaulting to US or UK assumptions. Both are free to add, and neither is optional if you want an answer you can actually check.

Where it goes wrong: confident, wrong, and hard to catch

The single biggest risk is not that ChatGPT misses something. It is that it states something false in exactly the same confident tone as something true. Ask "does this contract have a most-favoured-customer clause?" and it may answer "yes, Clause 14.3 states..." and paraphrase something plausible-sounding that is not actually in the document, or a real clause number attached to the wrong content. This is a known failure mode of large language models generally, not a ChatGPT-specific bug, and it worsens with longer, denser documents. The fix is not to trust it less in general, but to make every claim independently checkable, which is why the exact-sentence-citation instruction matters more than any other single prompt tweak.

What ChatGPT reliably misses about Indian contract law

General-purpose models train overwhelmingly on US and UK contracts and commentary. Left to its defaults, ChatGPT applies common-law assumptions that are wrong once an Indian statute overrides them, and it will not warn you unless asked directly. Three gaps are worth checking on every Indian contract.

Non-compete clauses. ChatGPT will often assess a post-employment non-compete on "reasonableness," duration, geography, scope of restriction, because that is how it has learned US courts approach the question. Indian law does not run that test. Section 27 of the Indian Contract Act, 1872 says:

"Every agreement by which any one is restrained from exercising a lawful profession, trade or business of any kind, is to that extent void." Source: Section 27, Indian Contract Act, 1872

The narrow carve-out is a seller of business goodwill agreeing not to compete with the buyer, within reasonable limits. A standard post-employment non-compete is, on the weight of Indian authority, void to that extent. Ask ChatGPT to review an employment agreement without telling it this, and it will debate whether a two-year, India-wide restriction is "reasonable" instead of flagging that it is very likely unenforceable as written.

IP assignment clauses silent on duration or territory. ChatGPT, reading from US-heavy training data, tends to assume silence on duration or territory means "forever, everywhere." Indian copyright law has an explicit default for exactly this gap. Sections 19(5) and 19(6) of the Copyright Act, 1957 state:

"If the period of assignment is not stated, it shall be deemed to be five years from the date of assignment." "If the territorial extent of assignment of the rights is not specified, it shall be presumed to extend within India." Source: Section 19, Copyright Act, 1957

A clause that simply says "the Contractor assigns all rights in the Work to the Client," with no stated period or territory, does not hand over worldwide, permanent rights by default in India. It reverts to a five-year, India-only assignment, a real defect for a client who thinks they own the work outright and globally, and one a generic ChatGPT pass will not flag unless you specifically ask it to check for both a stated period and a stated territory.

Stamping. Western contract practice has no real equivalent to this, so ChatGPT usually has no trained reason to raise it at all. Under Section 35 of the Indian Stamp Act, 1899, an unstamped instrument cannot be used as evidence:

"No instrument chargeable with duty shall be admitted in evidence for any purpose by any person having by law or consent of parties authority to receive evidence... unless such instrument is duly stamped." Source: Section 35, Indian Stamp Act, 1899

This even reaches arbitration clauses. In In Re: Interplay Between Arbitration Agreements under the Arbitration and Conciliation Act, 1996 and the Indian Stamp Act, 1899 (2023 INSC 1066), a seven-judge Supreme Court bench held an unstamped arbitration agreement is not void outright, but the defect is curable and the document stays inadmissible until the duty is paid. Ask ChatGPT to review your dispute-resolution clause and it will comment on the arbitration seat and rules, not on whether the contract is under-stamped, since stamping is not a question a US or UK contract ever raises.

The data question: what happens to the contract after you paste it in

This is the part most people skip, and it matters more than any single clause. Paste a confidential contract into a free ChatGPT account and the text can be used to improve OpenAI's models unless you have turned that off. OpenAI's documentation confirms conversations may be used to improve model performance, controlled by a setting, "Improve the model for everyone," under Settings, Data Controls, on by default for Free, Plus, and Pro accounts. A separate Temporary Chat mode skips history and memory and is never used for training, though it is still retained for 30 days for safety review before deletion.

Claude changed its own consumer policy in 2025: Free, Pro, and Max accounts are now opted in to training by default too, unless turned off in Settings, Privacy, "Improve Claude for everyone." Opting out keeps the standard 30-day retention window; allowing training extends it to five years. An Incognito chat, like Temporary Chat, is excluded from training regardless of the account-level setting.

The practical rule: on a free account, on either tool, assume the default is "yes, this trains," and check the setting before pasting an NDA, a client contract, or anything a confidentiality clause already forbids sharing. For a contract you cannot risk uploading anywhere, mark it up and flag clauses free in Weave, in your browser, without sending the document to a server at all.

Two tests worth running before you trust the output

The privacy test. Before pasting anything sensitive, open Settings and find the training toggle described above (ChatGPT: "Improve the model for everyone" under Data Controls; Claude: "Improve Claude for everyone" under Privacy). Turn it off, or start a Temporary Chat or Incognito chat instead. This takes under a minute and is the single highest-leverage thing you can do before using either tool on a real contract.

The citation test. Pick a clause you are certain is genuinely absent, a specific IP indemnity, a most-favoured-customer clause, anything you have personally confirmed is not there. Ask directly: "does this contract contain a [that clause]? Quote the exact sentence." A correct "not present" is a good sign for this session. An invented clause number, quoting something that does not address what you asked, means it is fabricating on demand, and everything else that session deserves real suspicion.

Signs your ChatGPT review is trustworthy, and signs it isn't

Normal	Red flag	Why it matters
Every flag comes with a quoted sentence and clause number	Flags are stated as summaries, with no locatable quote	A paraphrase cannot be verified; only a quote can be Ctrl+F'd in the source
It says "not stated" when you asked about a clause that is genuinely absent	It states a specific number or period for something the document is silent on	A confident specific for genuine silence is the classic shape of a hallucination
Indian-specific issues (non-compete, IP assignment period, stamping) get raised once you ask	They never come up even after you say "this contract is governed by Indian law"	The model is still defaulting to US or UK training data
The training toggle is off, or you used Temporary/Incognito mode, before pasting	You paste first and check settings after, or not at all	Once text is sent, you cannot recall whether it was used for training
It passes the "find something absent" test above	It invents a clause number that is not in the document	Fabricating on a direct test means it fabricates elsewhere, less visibly
It notes the document may need stamping once you mention India	It never raises stamping, even for an arbitration clause	Stamping has no equivalent in ChatGPT's mostly Western training data

A bad prompt vs a better one

Bad: "Here is our vendor agreement. Please review it and tell me if there are any issues."

What is wrong: no stated position to compare against, no requirement to cite the source sentence, and "any issues" lets the model surface whatever is easiest to describe rather than what actually matters for your side.

Better: "You are reviewing this contract for us as the vendor, governed by Indian law. Our position: liability capped at 12 months' fees, indemnity excludes indirect and consequential loss, payment terms of 30 days. Go clause by clause. For each: what it says in plain English, whether it matches or deviates from our position, quoting the exact sentence, and if there is no matching clause, say 'not present,' do not infer one. Separately, tell me whether this contract contains a post-employment non-compete, and whether any IP assignment clause states both a period and a territorial extent."

A bad clause vs a better one: the IP assignment gap ChatGPT will miss

Bad (what many contracts actually say): "The Contractor hereby assigns to the Client all rights, title, and interest in the Work."

Why this is a trap in India: no period, no territory. Under Sections 19(5) and 19(6) of the Copyright Act, this defaults to a five-year, India-only assignment, not the permanent, worldwide ownership both sides probably assumed. Ask ChatGPT to review this clause without mentioning Indian law and it will likely call it "a standard broad assignment" and move on.

Better: "The Contractor hereby assigns to the Client all rights, title, and interest in the Work, in perpetuity, for the full term of copyright and any renewals or extensions thereof, and throughout the world."

What changed: an explicit period and an explicit territory, which removes the statutory default entirely instead of leaving it to apply by accident.

ChatGPT vs Claude vs a no-upload option

	ChatGPT (free/Plus)	Claude (free/Pro)	Weave
Cost to try	Free tier, paid from about $20/month	Free tier, paid from about $20/month	Free
Trains on input by default	Yes, unless "Improve the model for everyone" is off	Yes, unless "Improve Claude for everyone" is off	No upload, question does not apply
No-training mode	Temporary Chat, 30-day retention	Incognito chat, excluded from training	Nothing leaves the browser
Clause-by-clause depth	Strong, with a structured prompt	Strong, with a structured prompt	Manual markup, not AI analysis
Cites exact source sentences	Yes if asked; verify, hallucination risk remains	Yes if asked; verify, hallucination risk remains	You read the original text directly
Raises Indian-specific issues unprompted	Rarely, ask directly	Rarely, ask directly	Not applicable, it is a markup tool

FAQ

Is it actually legal to use ChatGPT to review a contract? Yes. The risk is not in using it, it is in what you paste (a confidential document you should not be sharing) and in relying on its output as legal advice.

Will ChatGPT tell me if a clause is unenforceable in India? Only if you ask directly and state that Indian law governs. Its training leans heavily on US and UK practice, so it will not spontaneously flag a void non-compete or an IP assignment that silently reverted to five years.

Can I trust ChatGPT with a contract that has a confidentiality clause? Check the clause first; pasting into a third-party tool may already breach it regardless of what ChatGPT then does. If you proceed, turn off model training or use Temporary Chat, and consider a no-upload option like Weave for anything genuinely sensitive.

Does asking ChatGPT to "cite the exact sentence" stop hallucinations? No, but it makes them checkable. A quoted sentence you can Ctrl+F in the source either exists or it does not; a summarised claim gives you nothing to verify against.

Is ChatGPT better or worse than Claude for this? Neither is reliably better at the Indian-law gaps above; both need to be told explicitly. The bigger difference is the data setting on whichever tool you use, and whether you checked it before pasting.

When should I stop using ChatGPT and call a lawyer instead? Anything touching indemnity caps, IP ownership, non-compete terms, governing law and dispute forum, or real money and a long lock-in, should get a lawyer's eyes first, however clean the ChatGPT review looked. For the fuller workflow, see How to Review a Contract With AI. For the complete privacy answer, see Is It Safe to Upload a Contract to AI?

ChatGPT can genuinely speed up how you read and triage a contract, and the workflow above makes its output checkable rather than something you take on faith. It cannot tell you whether a specific clause will hold up if challenged, whether you have real leverage to push back, or whether this contract is safe to sign. That judgement needs a lawyer who can see facts ChatGPT cannot. Nothing in this guide is legal advice, and it is not a substitute for one.