The short version
- A contract is enforceable if there is offer, acceptance, consideration, capacity, certainty of terms and an intention to be bound. Drafting method is not on the list.
- The real risks are content risks: a clause that does not match your jurisdiction, an internal contradiction, or a confidently invented citation.
- AI drafting is strongest on structure and completeness — the clauses people forget — and weakest on local statutory requirements and formalities like witnessing.
- Get a lawyer for anything high-value, cross-border, regulated, or that takes effect on death. Bring the AI draft with you; it makes the review cheaper, not redundant.
Why the answer is straightforwardly yes
A contract is formed when parties with capacity agree definite terms, exchange something of value, and intend to create legal relations. That is the whole test in common-law systems, with civil-law equivalents differing in detail but not in the relevant respect: none of them ask about the drafting process.
Contracts drafted from templates found online, copied from a previous deal, or written by a non-lawyer are all routinely enforced. A contract drafted with AI assistance sits in exactly the same category. In practice you would not usually know how a document was produced, and there is no obligation to disclose it.
What the drafting method does and does not change
Does it change anything that AI drafted the contract?
Not for enforceability
Offer, acceptance, consideration, capacity, certainty of terms and an intention to be bound. No legal system asks how the document was produced.
Yes for content
Jurisdiction mismatch, missed formalities, internal contradictions, invented citations. Ordinary document risks — and still entirely yours.
What actually goes wrong
The failures worth planning for are content failures, and they cluster into four types.
- 1
Jurisdiction mismatch
A clause that is standard in one legal system and void in another. Restrictive covenants, liability exclusions, deposit terms and termination rights are all heavily jurisdiction-dependent. This is why a well-built tool asks which law governs the document before it drafts, rather than defaulting to an unnamed jurisdiction.
- 2
Formalities missed
Some documents need more than words on a page: witnessing, notarisation, registration, specific statutory wording, or delivery by a prescribed method. A model can produce a perfect-looking will that fails because it was not executed properly, and that failure surfaces at the one moment nobody can fix it.
- 3
Internal inconsistency
A defined term used before it is defined, a clause cross-referencing a schedule that does not exist, two clauses giving different notice periods. Longer generated documents are more prone to this, and it is the easiest category to catch by reading.
- 4
Confident invention
Language models can produce citations, statute numbers and case names that do not exist, stated with complete assurance. This has produced sanctions for lawyers who filed AI-written briefs without checking. Treat any specific legal citation in generated text as unverified until you have looked it up.
Notice that none of these are problems with enforceability. They are problems with the document being wrong — which is a much more ordinary risk, and one that also applies to a template downloaded from a search result.
What AI drafting is genuinely good at
It is worth being specific rather than either enthusiastic or dismissive.
| Task | How reliable | Why |
|---|---|---|
| Producing a complete clause structure | High | Completeness is a pattern problem — it does not forget the notices clause at 11pm |
| Explaining what a clause does | High | Translating dense prose into plain language is a core language task |
| Spotting missing standard protections | High | Absence against a known structure is exactly what models are good at detecting |
| Rewriting a clause to be fairer or clearer | High | Constrained rewriting with the meaning preserved |
| Applying local statutory requirements | Medium | Depends entirely on the jurisdiction being specified, and on how well-documented it is |
| Getting execution formalities right | Low to medium | Witnessing, notarisation and registration are procedural, local and frequently updated |
| Citing statutes and case law | Low | The failure mode is confident fabrication — verify every citation |
| Commercial judgement | Not applicable | Whether the price is right or the counterparty will perform is not in the document |
The pattern is consistent: reliable on structure and language, less reliable the closer you get to local procedural law, and no use at all for facts it cannot see.
How to check an AI-drafted contract
Before you send it to anyone
- Read it end to end once. Generated documents read fluently, which makes skimming tempting and skipping errors easy.
- Check every party name, date, address and amount against your own records — these are the fields most often carried over from an example.
- Confirm the governing law clause says the jurisdiction you actually intended, and that it appears only once.
- Look up every capitalised defined term and confirm it is defined, once, before it is used.
- Follow every cross-reference: clause numbers, schedule references, exhibit names.
- Verify any statute, regulation or case cited by name. If you cannot find it, delete it.
- Check the execution block: who signs, whether a witness or notary is needed, and how many copies.
- Run the finished draft through a review pass — ideally one that reads it against the 12 red flags rather than just proofreading it.
Draft with the jurisdiction as an input
The app asks which law governs the document before it writes anything, then composes the clauses around your answers — 136 document types, three lengths, 43 languages. Three documents a month free.
The question about lawyers
AI drafting does not remove the need for legal advice; it changes what you are buying. Reviewing a draft you bring in is faster and cheaper than drafting from a blank page, and arriving with three specific questions costs a fraction of arriving with "please look at this deal".
How far up to go, and what decides it
- Minutes
Draft it and send it
Defensible where the worst case is a month of unpaid work, and the counterparty is known.
- An hour
Draft it, then work the checklist above
Catches all four content failures. Most routine agreements should stop here.
- One billed hour
Draft it, then buy an hour of review
You are paying to have a draft read rather than written. Arrive with three specific questions.
- Several hours
Have it drafted for you
Cross-border, regulated, security over assets, anything taking effect on death or incapacity, anything already in dispute.
Rung four is not a judgement about AI. It is where a template downloaded from a search result would also have been the wrong tool.
What about contracts negotiated or signed by AI agents?
This is the genuinely open question, and it is different from AI-assisted drafting. Where an automated system makes an offer or accepts one without a human in the loop, the analysis turns on authority: was the agent acting within authority the principal granted, and did the counterparty reasonably believe so? Electronic transactions legislation in several jurisdictions already contemplates contracts formed by automated systems, and the general answer is that the principal is bound.
The practical guidance for anyone deploying that today is unglamorous: define and document the authority limits, log what the system did and on what basis, keep a human approval step above a value threshold, and make sure the terms presented to the counterparty are the ones you intended. Those controls are what an argument about authority will eventually turn on.
The honest summary
Enforceability was never the interesting part of this question. A contract produced with AI is as binding as one typed by a partner at a law firm, and treating that as reassurance is the mistake — because it means an unchecked clause you did not understand is also fully binding. Use the tools for what they are good at, check the output where it is known to be weak, and escalate the categories that deserve a professional. That is the same discipline that has always applied to templates; the tools are simply much better now.
General information, not legal advice. This guide explains how these documents and rules generally work. Law varies by jurisdiction and changes, and none of it is applied to your circumstances here. For anything consequential, consult a licensed attorney where you are.
Frequently asked
Do I have to tell the other party the contract was AI-drafted?
There is no general disclosure obligation. Contracts are judged by their terms, and nobody asks who typed them. Specific professional contexts differ — some courts now require lawyers to certify how filings were prepared — but that is a rule about litigation conduct, not about contract validity.
Can an AI-generated contract be challenged for being unfair?
On exactly the same grounds as any other contract: unconscionability, misrepresentation, duress, or consumer protection rules on unfair terms. The origin of the drafting makes no difference to those arguments, though a term nobody can explain the purpose of is harder to defend when challenged.
Is a contract generated in another language binding?
Yes, but be deliberate about which version governs. Most cross-border agreements name one authoritative language and treat translations as convenience copies. Where a document is generated directly in the target language rather than translated from English, the legal register tends to be better — but the same rule applies: name the governing version in the contract.
What if the AI produces a clause that turns out to be void?
The usual severability analysis applies — a court strikes out or narrows the offending clause and generally enforces the rest. The practical risk is that the void clause was the one protecting you, which is why jurisdiction should be an input to the drafting rather than something you check afterwards.
Is it safe to paste a confidential contract into a general AI chatbot?
Check the retention and training terms of whatever you are using, because they vary widely and some consumer tools retain inputs by default. If a document contains salaries, settlement figures, client names or personal data, that matters. Tools built for this work should tell you plainly whether your content is used for training and where it is stored.