Somewhere out there, someone is asking ChatGPT to “write me a will” the same way they’d ask it to write a birthday speech. It’s a reasonable instinct. AI is fast, free, and confident – three qualities that make it dangerously persuasive when the stakes are your entire estate.
Here’s the twist, though: even South Africa’s own government got caught out by this exact problem in 2026. The Department of Communications and Digital Technologies published a draft National AI Policy for public comment, only to withdraw it weeks later after it emerged the document contained fake, AI-hallucinated references. If a national policy on AI can trip over AI’s own confident inaccuracy, an AI-drafted will deserves the same scepticism.
This article looks at what AI genuinely can do for your legal documents, what it can’t, and why the gap between “sounds legally correct” and “is legally correct” is exactly where estates go wrong.
What Actually Makes a Will Valid in South Africa
A will isn’t valid because it reads like a will. It’s valid because it satisfies section 2(1) of the Wills Act 7 of 1953, which requires:
- The will must be in writing.
- The testator must sign it (or direct someone to sign on their behalf, in their presence, following specific rules).
- Two competent witnesses, both over 14, must be present at the same time as the testator signs, and must sign the will themselves.
- The testator must sign every page, not just the last one.
None of this is about the quality of the wording. A one-page will scrawled by hand, signed and witnessed correctly, beats a flawless twenty-page AI-generated document that nobody signed properly. The law cares about the ceremony, not the composition.
This is precisely where AI tools fall short, because a chatbot cannot witness a signature, cannot confirm testamentary capacity, and has no idea whether your specific family situation (a second marriage, a minor child, an offshore asset, a business interest) needs a clause it didn’t think to include.
Where the “AI Wrote It, But Nobody Signed It” Problem Gets Real
In Perumal v Janse van Rensburg NO and Others [2025] ZAGPPHC 145, the deceased had made amendments to his will on his laptop and forwarded the updated document to his executor, but died before signing it. The executor tried to rely on a voice note as evidence of the deceased’s intentions. The matter had to go to the High Court under section 2(3) of the Wills Act, which allows a court to direct the Master to accept a non-compliant document as a valid will, but only if the court is satisfied the deceased genuinely intended it to be their final will.
Section 2(3) is sometimes described as a safety net. It isn’t a reliable one. It requires an expensive, time-consuming High Court application, with no guaranteed outcome, at a moment when your family is already grieving. An AI-generated draft that was never properly signed sits in exactly this danger zone. Technically, it might be rescued, but only after your loved ones have paid legal fees and waited months to find out.
Where AI Genuinely Helps, and Where It Doesn’t
AI tools are not the enemy here. Used well, they’re a decent starting point:
What AI can do:
- Help you organise your thinking before a consultation: who you want to benefit, what assets you hold, questions you hadn’t considered.
- Explain unfamiliar legal terms in plain language.
- Generate a checklist of documents and information to bring to your attorney.
What AI can’t do:
- Confirm your document meets the Wills Act’s formal requirements.
- Witness a signature or verify testamentary capacity.
- Catch the legal nuance specific to your circumstances: a trust structure, a blended family, an asset in another jurisdiction.
- Guarantee that any legislation, case law, or “facts” it cites are real.
That last point matters more than people assume. South African courts have already dealt with attorneys submitting AI-generated legal arguments that cited cases which simply don’t exist. The Legal Practice Council has flagged this as a serious professional risk, and the profession’s own ethics guidelines (developed through the Law Society of South Africa’s AI Committee, with input from the Bench and senior academics) now explicitly require every AI output to be independently verified before it’s relied on. If qualified legal professionals are required to double-check AI’s work, it’s a reasonable rule of thumb for anyone using it to draft their own legal documents too.
There’s also a quieter risk worth flagging: pasting your personal and financial details into a public AI tool to “help write your will” means that information leaves your hands. Firms handling client information have obligations under the Protection of Personal Information Act 4 of 2013 (POPIA); individuals doing their own drafting don’t have that same oversight, and should think twice about what they’re typing into a chat window.
Practical Implications for Your Estate
If you’ve already used AI to draft or amend a will, that document is not automatically worthless, but it is not automatically valid either. Treat it as a starting draft, not a finished product. The steps that actually matter are the ones AI cannot perform for you: proper execution, correct witnessing, and a review by someone who can spot what’s missing.
Firms that use AI responsibly (as a research and drafting aid, checked and finalised by a qualified attorney) get the benefit of AI’s speed without inheriting its risk. That’s a meaningfully different approach to typing a prompt into a free chatbot and printing the result.
Conclusion
AI can help you think about your will, but it cannot make your will valid. That job still belongs to the Wills Act’s formalities (written, signed, witnessed, every page) and to an attorney who can catch the gap between what you typed and what the law actually requires. Given the cost, delay, and uncertainty of a section 2(3) rescue application, it’s a far cheaper exercise to get it right the first time.
While every reasonable effort is taken to ensure the accuracy and soundness of the contents of this publication, neither the writers of articles nor the publisher will bear any responsibility for the consequences of any actions based on information or recommendations contained herein. Our material is for informational purposes.