AI Drafts a Will and Leaves Out Everything That Matters

The Experiment That Revealed the Cracks

A British law practice recently put generative AI to the test in estate planning, and the results underscore a fundamental tension in legal automation: the difference between producing a document and understanding what should go into it. SE Solicitors asked a chatbot to draft a last will and testament for a fictional client named Daniel - unmarried, 42 years old, father of two adult children, owner of a £440,000 home, £10,000 in savings, a car, and a dog. The firm then had Tom McInerney, a senior associate specializing in wills and probate, review the output.

At first glance, the AI-generated will looked credible. The prose flowed smoothly, the formatting met conventional standards, and the clauses appeared coherent. But McInerney's analysis revealed a cascade of omissions that would trouble any estate planning professional. The chatbot had produced a document that fit the prompt - but missed the person behind it entirely.

What the Chatbot Missed

The AI-generated will made no mention of inheritance tax planning, lifetime gifts, pension arrangements, life insurance policies, digital assets, trusts, or the financial vulnerability of either child. It failed to ask whether Daniel had a romantic partner, wanted to leave bequests to friends or charitable organizations, or intended to disinherit anyone. In one clause addressing care for Daniel's dog, the chatbot inserted a placeholder instead of a specific sum and neglected to make the bequest conditional on the named beneficiary actually accepting the animal.

These gaps are not merely stylistic. Inheritance tax rules in the UK can significantly affect the net value of an estate, particularly for properties approaching or exceeding the nil-rate band. Digital assets - from cryptocurrency wallets to social media accounts - are increasingly part of estate inventories, yet many AI-generated documents treat them as afterthoughts or ignore them altogether. Trusts can protect vulnerable beneficiaries or minors; without them, a will may expose heirs to financial risk or legal disputes.

McInerney pointed out that the chatbot could only work with the information it received. A human solicitor, by contrast, would spend time asking follow-up questions, probing for details the client might not think to volunteer. The AI had no mechanism to identify what was missing because it had no framework for understanding what a complete estate plan requires.

The Limits of Prompt-Based Legal Work

At DailyTechWire, we've tracked the adoption of large language models in professional services across Asia and Europe, and the pattern is consistent: LLMs excel at text generation but struggle with interrogation. In legal practice, the hardest part of drafting a will is not writing clauses - it is surfacing the facts, preferences, and contingencies that inform those clauses. An AI trained on vast corpora of wills can recognize standard language and structure, but it cannot ask why a client chose not to mention a sibling, or whether a second marriage creates blended-family dynamics, or whether a business interest should pass through a trust.

SE Solicitors emphasized that a will is a legally binding instrument that must withstand scrutiny in probate court. Errors or omissions can trigger disputes among heirs, invalidate bequests, or expose an estate to avoidable tax liability. The firm warned that consumers using AI tools may overlook the need to revisit their wills after major life events - marriage, divorce, the birth of children, property acquisitions - because the chatbot offers no reminders and no ongoing relationship.

The data-privacy dimension also looms large. Anyone who feeds personal and financial details into a commercial AI platform is transmitting sensitive information to a third party, often without clarity on how that data is stored, processed, or shared. Solicitors in the UK operate under strict confidentiality obligations enforced by professional regulators; AI platforms, by contrast, are governed by terms of service that many users never read.

Rising Demand, Uneven Readiness

Search interest in legal AI has surged. SE Solicitors cited Google Trends data showing a 312 percent increase in searches for "legal AI" over the past year, a trajectory that mirrors what we have observed in parallel verticals - contract review, due diligence, compliance monitoring. Separately, research commissioned by the Association of Lifetime Lawyers found that 72 percent of UK adults aged 30 to 34 would consider using AI to write or update a will.

That appetite reflects both cost sensitivity and the perception that AI can democratize access to legal services. Drafting a will with a solicitor in the UK typically costs several hundred pounds; a chatbot offers an alternative at near-zero marginal cost. For younger adults with straightforward estates - no property, few assets, no dependents - an AI-generated will might suffice. But as wealth and complexity accumulate, the risk of relying on a tool that cannot ask questions grows correspondingly.

What This Means for Legal Automation

The SE Solicitors experiment does not prove that AI has no role in estate planning. It does, however, clarify where the technology sits today. Generative models are capable document assemblers, able to synthesize standard clauses and adapt them to simple fact patterns. They are not yet capable advisors, and the gap between those two functions is not merely a matter of training data or parameter count. It is a gap in reasoning architecture: the ability to recognize uncertainty, to surface latent issues, and to guide a client through decisions they did not know they needed to make.

We are beginning to see hybrid models in which AI handles initial drafts or flags potential issues, and human professionals review, interrogate, and finalize. That division of labor makes sense in high-stakes domains where the cost of error is high and the value of judgment is difficult to automate. The challenge for the legal profession - and for the AI vendors serving it - is to design workflows that preserve the interrogative function while capturing the efficiency gains of automation.

For now, the lesson from this trial is straightforward: AI can write a will that looks like a will. Whether it can write a will that reflects the client's intentions, protects their heirs, and withstands legal scrutiny depends entirely on what the client thought to include in the prompt - and what they did not.