AI wills are already showing up in probate disputes before a single court has ruled on how to treat one, and the evidentiary gap that creates is more serious than most people typing a prompt into ChatGPT realise. No law in the United States, Canada, or the United Kingdom specifically prohibits an AI-generated will. That is precisely the problem. The absence of a prohibition has let a wave of DIY estate planning move ahead of the legal system’s ability to say clearly what happens when the document at the centre of a contested estate was drafted by a chatbot rather than a person.
This piece looks at what actually happens when an AI-drafted will meets the formal rules courts apply to every will regardless of who wrote it, where the technology’s real limitations sit, and what a genuinely useful role for AI in estate planning looks like once you separate it from the risky one.
Why “No Law Prohibits It” Is Not the Same as “It’s Fine”
Wills have never been judged by how well-written they sound. They are judged against formal execution requirements that exist specifically to prove a document reflects a person’s genuine, considered wishes: the will must be in writing, signed by the testator, and witnessed by a required number of people present at the same time, requirements that vary by state and country but exist in some form almost everywhere.
An AI chatbot can produce text that reads exactly like a will. It cannot supervise a signing, confirm two witnesses were in the same room, or assess whether the person signing had the legal capacity to understand what they were doing. Every one of those steps still depends entirely on the human being who used the tool getting the execution right, unassisted, after the AI’s job is already finished.
The Evidentiary Gap That Makes AI Wills Different From Ordinary DIY Wills
AI-generated wills share the well-documented risks of fill-in-the-blank will kits, but they introduce a distinct problem that DIY kits never did. A detailed analysis published by the Canadian Bar Association’s British Columbia branch lays out the mechanism precisely: when a will ends up contested, courts rely on external evidence of what the testator actually intended, a drafting solicitor’s notes, correspondence, earlier draft versions, to establish that a document reflects genuine, informed wishes.
An AI-generated will typically leaves none of that trail behind. There is no solicitor’s file. The prompt that produced the document is usually unrecoverable. What a court is left with is the finished text and nothing that explains how it got there.
The author of that analysis ran a direct test of the problem, prompting several AI chatbots to draft a will clause modelled on a real, previously litigated Ontario case involving an ambiguous life-estate provision. Some chatbots successfully resolved the ambiguity that had actually required court intervention in the real case. Others, including two mainstream consumer tools, reproduced the exact same ambiguous structure that had already caused litigation once.
The finding is a useful, concrete illustration of a wider point: AI-drafted legal language can sound entirely fluent and still fail to resolve the specific contingency a trained drafter would have caught, and there is no way to know which outcome you got until the ambiguity surfaces after the person who could clarify it has died.
The Privilege Problem Nobody Warns You About
A separate risk sits underneath the drafting question entirely, and it has nothing to do with whether the AI produces good language. A detailed legal analysis from a California estate planning practice traces the mechanism through the state’s evidence code: a conversation with a licensed attorney about your estate is protected by attorney-client privilege and generally cannot be forced into a courtroom. A conversation with a consumer AI chatbot has no such protection, because there is no attorney-client relationship for privilege to attach to in the first place.
The analysis points to a concrete precedent: in February 2026, a federal court in United States v. Heppner ruled that documents a client had created using a public AI tool were not privileged, even after the client subsequently handed those materials to his own lawyer, because sharing them with a third party before an attorney ever saw them had already broken the confidentiality privilege depends on.
Applied to estate planning, that ruling has a direct and uncomfortable implication. Anyone who types their full financial picture, the reasons for disinheriting a family member, or a family conflict into a chatbot while drafting a will has created a written, potentially subpoenable record of exactly the kind of information a contest lawyer representing a disappointed heir would want to see.
Details typed into a consumer AI platform can be stored, reviewed, or retained well beyond the conversation itself, in ways a person drafting with an attorney would never have to think about.
The Undue Influence Trap AI Cannot See Coming
Several jurisdictions have built specific statutory presumptions around who drafted a will, precisely because a document that benefits its own drafter is a well-documented pattern in contested estates. California’s Probate Code, for instance, presumes that a gift to the person who actually drafted the instrument is the product of fraud or undue influence, a presumption that is difficult or impossible to rebut without an independent attorney’s involvement in the process.
That statutory scheme was built around the assumption that a human drafter exists and can be identified. An adult child who sits down at a parent’s computer and prompts a chatbot to draft a will amendment that happens to favour that child sits inside exactly the fact pattern these laws were written to catch, whether or not the child is technically the one who “drafted” the document in a traditional sense. AI does not remove the underlying risk of undue influence in vulnerable family situations. It simply removes the one safeguard, an independent professional asking hard questions before anyone signs anything, that the entire statutory framework assumes will be present.
Where AI Genuinely Helps
None of this means AI has no legitimate role near an estate plan. Used to answer general orientation questions, what is a revocable trust, what does probate actually involve, or to build a checklist of documents and questions to bring to an attorney, AI is a low-risk research tool that involves handing over no sensitive information and produces nothing anyone is going to sign.
Attorney-reviewed guidance aimed at consumers draws the line in the same place: AI can be a reasonable starting point for understanding your options, but drafting the actual language, deciding how to divide specific categories of property, or resolving how a beneficiary designation interacts with a trust requires a professional applying judgement to your specific facts rather than a plausible-sounding pattern-matched answer. The line that keeps recurring across every serious legal analysis of this technology is the same one: AI moving from explaining concepts to actually drafting or deciding is the point where the risk profile changes fundamentally, regardless of how confident or fluent the AI’s output happens to sound.
The Same Pattern Showing Up Across Every High-Stakes Legal Use of AI
The gap between what AI-drafted legal documents look like and what actually holds up once a human life or a family relationship is on the line is not unique to wills. LiveAIWire’s coverage of AI valuation tools entering divorce court found the identical structural problem in a different area of family law: an algorithm producing a confident-sounding number or recommendation that both sides accept, while nobody involved can actually explain the reasoning that produced it, and the parties with the least power to challenge that opacity are usually the ones who can least afford to.
The same dynamic runs through healthcare. LiveAIWire’s reporting on the AI doctor dilemma in diagnostic medicine found that the exact same underlying technology produces dramatically different outcomes, a 20 to 30 percent improvement in some deployments, a 66 percent failure rate to catch critical conditions in others, depending entirely on how much human oversight sits around it.
AI wills fit the same pattern precisely: the technology’s usefulness is not fixed. It depends entirely on whether a qualified professional remains in the loop, or whether the AI’s output is simply signed and trusted as final.
That dependence on oversight is not unique to law or medicine either. LiveAIWire’s broader reporting on AI recruitment tools screening job candidates found the same governance gap in hiring: a plausible-sounding algorithmic decision, adopted without independent scrutiny, that quietly shifts outcomes against the people least equipped to challenge it. Estate planning simply raises the stakes of that same pattern to their maximum, since the person affected is no longer available to correct the record once the error surfaces.
Who Is Actually Checking These Tools
Part of what makes AI wills a harder problem than other AI accountability failures is that almost nobody is independently checking the commercial products in this category before they reach a consumer. LiveAIWire’s coverage of the growing movement to build independent audit infrastructure for AI systems found that most sectors adopting consequential AI tools have moved faster than any external body’s capacity to evaluate them, and estate planning is a particularly stark example: there is no equivalent of a medical device regulator or a bar association certification process standing between a consumer will-drafting app and the person who downloads it.
What This Means for You
If you are considering using AI anywhere in your estate planning process, the practical dividing line the legal analysis converges on is specific rather than abstract. Use AI to understand terminology, build a list of questions, or organise your thinking before a meeting with an attorney, and you have taken on essentially no legal risk.
Use AI to draft the actual will or trust language, decide how property should be divided, or resolve any provision involving a beneficiary, a blended family, or a business interest, and you have taken on every risk described above: an execution that may not meet your jurisdiction’s formal requirements, a document with no evidentiary trail if it is ever contested, a permanent record of sensitive family information with no privilege protecting it, and, in some family situations, the exact fact pattern undue-influence law was written to catch.
An attorney-drafted estate plan typically costs a few thousand dollars. A contested probate matter arising from a flawed AI-drafted document routinely costs many times that in legal fees alone, stretches on for a year or more, and often damages family relationships in ways money cannot repair.
If you have already drafted something using AI, the safer move is bringing it to an attorney for review before it is signed and witnessed, not after someone has challenged it. A flawed provision is usually a quick fix during drafting. A signature ceremony that already happened incorrectly, or a privileged conversation that never existed in the first place, cannot be undone after the fact.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.