TL;DR: ChatGPT and other LLMs can draft a prenup that looks right, and that's the danger. A 2025 Stanford-led study found that even purpose-built legal AI tools from LexisNexis and Thomson Reuters hallucinate between 17% and 33% of the time, and AI tools can miss proper handling of state-specific rules like California's seven-day delivery requirement or New York's deed-style acknowledgment standard. An unenforceable prenup gives both partners false confidence, until a court throws it out.If you're reading this, there's a good chance you recently got an attorney quote that made you blink. Maybe $5,000. Maybe $10,000. Maybe more. And maybe you opened ChatGPT later that night and asked it to draft something, just to see. That's a reasonable thing to do. AI is a powerful tool, and prenups have been gatekept behind hourly billing for a long time. The question isn't whether you're allowed to use AI. It's whether the document you end up with will hold up. If you want a foundation on what a prenup does in the first place, our primer on premarital agreements is a good starting point.
This post gives AI credit for what it does well, then walks through the specific places it fails, with named statutes and a federal court case as evidence. If you decide AI is enough for your situation after reading this, you'll be making an informed call. If you decide it isn't, you'll know exactly why.
Key facts: AI and prenups at a glance A 2025 peer-reviewed study in the Journal of Empirical Legal Studies tested the leading purpose-built legal AI tools and found that LexisNexis Lexis+ AI, Westlaw AI-Assisted Research, and Ask Practical Law AI each hallucinate between 17% and 33% of the time . Westlaw's tool was found to hallucinate nearly twice as often as the others. General-purpose tools like ChatGPT hallucinate on specific legal queries between 58% and 82% of the time, according to Stanford research cited in the same paper. In Mata v. Avianca , an attorney submitted a brief in federal court containing six judicial opinions invented by ChatGPT. The court sanctioned both attorneys and their law firm, and the case became the canonical example of generative AI failing in a legal setting. That was 2023. The problem has not been solved since; it's gotten harder to detect. Newer models produce more polished prose around their fabrications, which makes the hallucinations look more like the real thing. General-purpose AI tools have no built-in connection to legal citator services like Shepard's or KeyCite, which are the industry-standard tools attorneys use to verify whether a case is still good law. An AI tool can cite a real case that's been overturned, or fabricate one that doesn't exist at all, and there's no automated check inside the model that catches either failure. Some purpose-built legal AI platforms used by major law firms integrate directly with Shepard's Citations to validate case law in real time. However, these tools are gated to enterprise legal customers and not realistically available to a consumer drafting a prenup at home. What AI does well in this context ChatGPT and other AI tools are useful at the front end of the prenup process. They're good at explaining what a premarital agreement is, what it can and cannot cover, and how the general categories of separate property, marital property, and spousal support work. If you and your partner are trying to decide whether to pursue a prenup at all, AI can help you organize the conversation. It can summarize concepts you'd otherwise have to read three Wikipedia tabs and a law-firm blog to piece together.
AI is also reasonable at drafting boilerplate. Generic separate-property clauses, debt-allocation language, and standard severability provisions read fine in AI output because that language is heavily represented in its training data. If your goal is to walk into a meeting with an attorney already knowing what you want and roughly how it might be worded, AI can shorten that prep meaningfully. Treating it as a thinking partner before legal work is a different exercise from treating it as the legal work itself.
The same goes for terminology. AI can explain what acknowledgment means, what community property is, what unconscionability refers to, in language a person without a law degree can follow. None of that requires the model to be right about your specific state, your specific timing, or your specific assets. Those are the places it falls apart.
The four things AI consistently gets wrong The first and biggest problem is state-specific procedural formalities . Prenup law is not federal. Every state has its own rules about waiting periods, signing requirements, financial disclosure, and the moments at which independent counsel is or isn't required. AI tools have no reliable way to know which state's rules apply to your signing, and they routinely describe rules from one jurisdiction as if they applied everywhere. California's seven-day delivery requirement under Family Code § 1615(c)(2)(B) is statutory, not a "best practice." New York requires acknowledgment in the manner of a deed under Domestic Relations Law § 236(B)(3), a stricter standard than the routine notary stamp most people picture. Florida requires two subscribing witnesses in addition to the parties' signatures when the agreement includes a waiver of spousal rights in the other's estate, such as the elective share or homestead, under Florida Statute § 732.702. AI drafts routinely miss or misdescribe all three.
The second is acknowledgment and notarization . Notarization is not one uniform action across states. In New York, the Matisoff v. Dobi decision held that without proper acknowledgment in deed form, the agreement is unenforceable. The Court declined to decide whether the defect could be cured after the fact, but made clear that the parties' later admissions in court were not enough to save it. A document that gets a generic notary stamp in the wrong state can fail entirely. AI output rarely flags this.
The third is financial disclosure . Most states require each partner to provide a full and fair disclosure of assets, debts, and income, or to expressly waive that disclosure in writing with knowledge of what's being waived. The schedules attached to a prenup are part of the agreement itself, and they have to reflect what you have at the time you sign. AI cannot generate a current, complete disclosure for you. It can produce a template that looks right and is empty where the substance should be.
The fourth is unconscionability analysis . A court asked to enforce a prenup years later examines whether the agreement was unconscionable at the time of signing, given the circumstances of both parties. That analysis is case-by-case and depends on facts AI doesn't have: the disparity in your assets, the timing relative to the wedding, whether either of you had counsel, whether anyone was under pressure. AI can recite the standard. It can't apply it to you.
Underneath all four is the hallucination problem the Stanford RegLab study quantified. Earlier Stanford research, cited in the same paper, found that general-purpose tools like ChatGPT hallucinate on specific legal queries between 58% and 82% of the time. Even tools purpose-built for lawyers, like LexisNexis Lexis+ AI and Thomson Reuters Westlaw AI, hallucinated between 17% and 33% of the time in a 2025 study published in the Journal of Empirical Legal Studies . Mata v. Avianca is the canonical example of what that looks like in practice: ChatGPT generated six judicial decisions out of thin air, complete with realistic citations and quoted language, and confirmed they were real when the lawyer double-checked. The attorneys were sanctioned. The cases never existed.
Why "looks right" is the most dangerous failure mode A prenup is a contract that often goes untested for ten or twenty years. You sign it, you file it away, and you (hopefully) never look at it again. The problem with an AI-drafted prenup isn't that it produces an obviously broken document. It's that it produces a polished, organized, legal-sounding document that reads like the real thing. Both partners feel protected. Both partners move on with their lives.
A decade later, if the agreement is ever tested, that's when the missing seven-day delivery window, the wrong acknowledgment form, or the inadequate disclosure schedule surfaces. By then, the rules of asset division revert to your state's default, which is what you signed the prenup to change in the first place. Today you sign the document feeling protected. Future you discovers the protection was never there.
This is what makes the failure mode so hard to catch in the moment. There's no error message. There's no red flag at signing. The agreement looks the part. The downside doesn't appear until the moment it was supposed to prevent.
The test that matters: would your prenup survive a challenge in your state? The right question to ask about any prenup, AI-drafted or otherwise, is whether it would survive a challenge in your specific state, under that state's specific procedural requirements, with the disclosure schedules and signing formalities each spouse would need to produce in court. That's the only test a court will care about. For a deeper walk-through of what enforceability requires, our full guide to getting a prenup goes through it state by state.
A few questions worth answering, with counsel if needed, before you decide your AI draft is enough:
Does the state where you'll be signing impose a waiting period between final draft delivery and signing? California does. Other states have judicial guidance favoring a window. Did the AI surface this, or describe it accurately? Does your state require a specific form of acknowledgment beyond a basic notarization? New York does. Several others do as well, in narrower ways. If you are waiving spousal support, does your state require each party to have independent counsel for the waiver to hold? California does. Is your financial disclosure schedule complete, current, and attached to the agreement? Does any clause in the draft cite case law? If so, has the AI verified that case law is still good law, or did it generate the citation? If you can't answer those questions confidently, the agreement isn't ready, regardless of how clean the prose looks.
What First does that AI can't First was built specifically for the situation that brings most people to AI in the first place: you want a fast, affordable prenup, and the traditional process priced you out. With the median lawyer in the U.S. earning $151,160 a year per the BLS, hourly billing on a contract that requires several rounds of drafting and review adds up quickly. The instinct to look for a better option is sound. The platform you choose still has to clear the enforceability bar.
First's drafting process is jurisdiction-aware. It applies the procedural formalities of the state where you'll be signing, including waiting periods, acknowledgment requirements, and disclosure standards. For couples who want a lawyer involved, First's Lawyer Review package allows each partner to connect with a separate, independent attorney who can review and negotiate on their behalf. The execution step is built around what each state requires for the agreement to be valid. No PDFs, no hourly rates, no waiting two weeks for a callback.
If you'd rather walk into a process with an attorney already prepared, our prenup checklist is a good way to organize your assets and questions in advance, whether you continue with First or somewhere else.
Frequently Asked Questions Can ChatGPT write a prenup? Yes. ChatGPT can produce a document that looks like a prenup, with organized sections and legal-sounding language. Whether it's enforceable is a separate question that depends on your state's specific procedural requirements: waiting periods, acknowledgment formalities, financial disclosure, and independent counsel rules that AI tools routinely get wrong or omit entirely.
Is an AI-generated prenup legally enforceable? It can be, but only if the agreement happens to comply with every formality your state requires. Many don't. Prenups have to satisfy state-specific rules around timing, signing, notarization, and disclosure, and AI tools have no reliable way to apply the right state's rules to your situation. A court tests enforceability years later, when it's too late to fix.
Will a lawyer review my ChatGPT prenup? Some will, but many won't. Reviewing AI-generated language often takes longer than drafting from a vetted template, because the attorney has to untangle clauses that don't fit your state's law. If a lawyer puts their name on the final agreement, they're taking responsibility for it, so most prefer to start over from a known foundation.
Why is "it looks right" the biggest risk with AI prenups? Because you may not discover the problem for ten or twenty years. An unenforceable prenup gives both partners false confidence during the marriage, then collapses at exactly the moment it was supposed to protect them. By then, the rules of asset division revert to your state's default, which is what you signed the prenup to change.
How does AI handle California's seven-day rule? Inconsistently, and usually wrong. California requires that the final draft be delivered to both parties at least seven calendar days before signing, regardless of whether either party has an attorney. AI tools tend to describe this as a general best practice rather than a statutory requirement under Family Code § 1615(c)(2)(B). Missing it can invalidate the entire agreement.
How does AI handle New York's acknowledgment requirement? Poorly. New York Domestic Relations Law § 236(B)(3) requires a prenup to be acknowledged in the same formal manner as a deed, a stricter standard than a routine notary stamp. The Court of Appeals held in Matisoff v. Dobi that without proper acknowledgment, the agreement is unenforceable, full stop. AI drafts often skip or misdescribe this step.
A middle option If you came here because an attorney quoted you $10,000 and AI looked like the only alternative, there's a middle option. First was built specifically for couples who want a fast, affordable, state-aware prenup without losing the parts that make it enforceable. Take a look at how it works. No commitment, no hourly rates.
First is not a law firm. The information and tools provided by First on this site are not legal advice and not a substitute for the advice of an attorney.
Statutes and case law cited in this post reflect the law as of publication. Readers in a specific state should confirm current requirements with a licensed attorney in that state. AI tools and their policies are changing quickly, and specific capabilities described above may shift after publication.
Sources Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools : Magesh, Surani, Dahl, Suzgun, Manning & Ho, Journal of Empirical Legal Studies (2025). 17–33% hallucination rate on purpose-built legal AI tools (Lexis+ AI, Westlaw AI-Assisted Research, Ask Practical Law AI); 58–82% hallucination rate on general-purpose LLMs for legal queries.Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models : Dahl, Magesh, Suzgun & Ho, Journal of Legal Analysis , Vol. 16, No. 1 (2024). 69–88% hallucination rate on specific legal queries.California Family Code § 1615 : Seven-day delivery requirement for premarital agreements.New York Domestic Relations Law § 236 : Acknowledgement requirement for prenuptial agreements; supported by Matisoff v. Dobi , 90 N.Y.2d 127 (1997).Mata v. Avianca, Inc. , 678 F. Supp. 3d 443 (S.D.N.Y. 2023): Documented case of ChatGPT fabricating six judicial decisions.U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Lawyers : Median annual wage for lawyers, $151,160 (May 2024).Uniform Law Commission, Premarital and Marital Agreements Act : State adoption of the Uniform Premarital Agreement Act.
