I spent years collecting wills. Not writing them, but collecting them. Photographs of handwritten pages from courthouse archives, scanned probate records, typewritten documents on crumbling paper. Over time, I assembled roughly 700 historical wills from ten archival collections across six countries: the United States, Canada, Australia, New Zealand, and, through the movement of people and paperwork, traces of England and France.

I am a legal educator in Thailand. I teach will drafting. And I had a problem that no textbook could solve.

Textbooks teach you the rules. They tell you what a residuary clause is, what an attestation clause must contain, how a life estate works. What they do not tell you is what real wills actually look like. The messy, human, sometimes heartbreaking documents written by ordinary people who knew they were running out of time.

I wanted my students to learn from the wills themselves. So I collected them. And then I waited.

The Wait

Here is the honest truth: I could not read 700 wills by myself. Not carefully, not systematically, not in any reasonable timeframe. Each will is a scanned image, a photograph of a page in a probate book, often handwritten, often faded, sometimes in looping nineteenth century cursive that takes ten minutes to decipher per paragraph.

I needed technology that could look at a photograph of a handwritten will from 1878 and not just read the words, but understand the document. Identify the clauses. Recognise the legal structure. Notice what was present and what was missing.

Traditional optical character recognition could not do this. It chokes on old handwriting, and even when it works, it produces raw text without understanding. I did not need a transcription machine. I needed a reader.

In 2026, I found one. I sat down with Claude, Anthropic’s AI, and asked it to read the first will in my collection. It was a photograph of a West Virginia will book page from 1944. The AI read it, and not just the words but the structure. It identified the preamble, the bequests, the executor appointment. It noticed the Catholic religious provisions. It distinguished this will from the unrelated will printed on the adjacent page.

I asked it to read another. Then another. Then I started sending it batches. One agent reading New York wills while another read Maryland, a third read West Virginia, a fourth read Australia and New Zealand. All at once.

In a single day, I read more wills than I could have read manually in months.

What We Found

I recently completed a formal write up of the findings as an academic article titled “Structural Patterns in Common Law Testamentary Instruments: A Corpus Based Analysis of 700 Historical Wills Across Six Jurisdictions, 1873 to 1967.” The full article contains the methodology, data tables, and detailed analysis. Here, I want to share the findings in plain language, because the wills deserve to be understood by everyone, not just lawyers.

The skeleton never changes

Every will in the corpus, every single one, across six countries and ninety four years, follows the same structural sequence: identify yourself, cancel prior wills, pay debts, give away your property, appoint someone to handle it, sign, and have witnesses certify. The words change. The complexity varies enormously. But the order never does.

A handwritten will from rural New York in 1873 and a solicitor drafted instrument from a New Zealand farm in 1942 follow the same architectural logic. This is not prescribed by any single law. It emerges from practice, as if there is an inherent logic to how human beings organise their final wishes.

Spiritual fades from the document

The oldest wills open with the words “In the name of God, Amen.” By the 1920s, this formula begins to disappear. By 1935, it is gone entirely, replaced by the phrase “being of sound mind and memory.”

The change is not about faith declining. It is about lawyers discovering that the capacity declaration has more legal weight in court than the religious invocation. The old formula was spiritual comfort for the testator. The new one is litigation armour for the estate.

One will stands apart. Zephirin Roy, a Quebec farmer in 1891, opens his will with the words: “I give my soul to God, begging Him to give it His mercy.” He is not reciting a legal formula. He is speaking directly to God before speaking to his heirs.

The wife gets income; the children get capital

About forty five per cent of all wills in the corpus use the same pattern. The wife receives use of the estate during her lifetime, and the children receive the capital when she dies. This appears everywhere, in Illinois, New York, New Zealand, and British Columbia, and across every decade.

The logic is twofold. Protect the widow, and at the same time preserve the family wealth. Some wills go further: if the wife remarries, she loses everything. The fear is not about the wife herself. It is about the next husband.

The five dollar trick

Multiple wills give a child exactly five dollars. Not as a gift, but as a legal shield. The token amount proves the testator knew the child existed and deliberately chose to leave them almost nothing. Without it, the child could argue they were simply forgotten.

But one will from New York in 1915 does something braver. Instead of the five dollar trick, the father writes a paragraph explaining why he favours his daughter Alice: she stayed home and cared for him while the others married and left. Modern courts respect this approach more than the token bequest. Explaining your reasoning is wiser than hiding behind a legal device.

People cared deeply about their graves

About a quarter of the wills include provisions for grave maintenance. One woman in West Virginia created a three thousand dollar trust, equivalent to roughly sixty five thousand dollars today, solely for flowers at her family plot. In perpetuity. Another testator in Maryland gave five hundred dollars to each of two churches in two different states to maintain cemetery lots where family members were buried.

The dead wanted to be remembered.

Wills track migration

A Quebec will probated in Saskatchewan. An Illinois will in Manitoba. A British Columbia man owning one hundred acres in Nova Scotia. A Yukon hotel keeper distributing property across five jurisdictions to siblings scattered from Quebec to Rhode Island.

Wills are migration records. They tell you where people came from, where they ended up, and what they carried with them.

Women’s power grows visibly

In the earliest wills, women appear as beneficiaries, receiving life estates controlled by male executors. By the 1930s and 1940s, women are testators with substantial estates, corporate trustees, and complex trust arrangements. The transition from dependent to autonomous actor is visible, datable, and traceable across the corpus.

What Weak Wills Teach

The most valuable lesson from the corpus is not found in the strong wills. It is found in the weak ones.

The same drafting failures appear across every jurisdiction and every decade. No plan for what happens if the spouse dies first. No backup executor. No clause catching property not specifically mentioned. No provision for what happens if a beneficiary dies before the testator and leaves children of their own.

These are not exotic mistakes. They are the same gaps, over and over, for ninety four years. They suggest that legal education has been failing to teach one critical skill: asking the question, “What if?”

A will is not a list of gifts. It is a set of instructions for every scenario the testator can foresee.

The Technology Question

People ask me whether AI wrote the research. The answer is no. The AI read the documents. I did the research.

The AI looked at 1,313 scanned images, photographs of handwritten and typewritten pages, some over a century old, and told me what they said. It identified clause structures, bequest types, executor appointments, and attestation formats. It distinguished between two different wills printed on adjacent pages of a probate book.

What it did not do is decide which patterns mattered, design the analytical framework, choose the teaching examples, or write the template. Those require judgment that comes from years of teaching, studying, and thinking about what students need.

The right analogy is the microscope. It lets the researcher see what was previously invisible. It does not perform the science.

But the scale matters. One person reading 1,313 archival images manually would need years. The AI made it possible in days. That is not a small thing. It is the difference between a research project that stays on the shelf and one that gets finished.

For My Students

Everything I found is now contained in two documents that sit alongside the original wills in my teaching collection.

The first is the Will Analysis Framework, a seven point scoring rubric for evaluating any will across intent, structure, language, bequests, executor appointment, safeguards, and modern fitness. It includes two worked examples scored in detail.

The second is the Universal Will Template, the structural skeleton that emerged from reading the entire corpus. Nine elements in a fixed order. Frequency data from 700 wills. Three distribution patterns for different family situations. A table of the most common drafting failures and what goes wrong when they appear.

The formal research behind all of this is documented in the academic article referenced above. But the real teaching happens when a student picks up one of the 700 original wills, reads it through the framework, and sees for themselves what works, what fails, and why.

That is why I collected them. That is why I waited for the technology. And that is why, when the technology arrived, the research could finally be completed.

Aphiwat Bualoi is a legal educator based in Chiang Mai , Thailand. The 1,000 Wills Project is an ongoing research initiative. The academic article, “Structural Patterns in Common Law Testamentary Instruments,” is available as a companion document. Contact: bualoi.law@hotmail.com.