A good case study can give us excellent ideas for tackling our challenging research. Enjoy this blog post by one of our Family Locket team members, Jill, who shares her success using AI as an assistant.

by Jill Leonard Nock, AG® 

Historical deed research can be simultaneously exhilarating and exhausting. The thrill of discovery often comes after hours of deciphering 18th- or 19th-century handwriting, tracking chains of title, and piecing together family relationships. But what if artificial intelligence could handle the tedious parts while you focus on the detective work? This case study demonstrates how AI transformed my approach to analyzing a complex series of Pennsylvania land transactions—and helped me solve a genealogical puzzle in real time. 

The Research Question 

Having wrapped up a client project, I decided to investigate my own family tree. I chose to focus on Martha Williamson, my 4th great-grandmother, born around 1789 in Ireland and died in Neshannock Falls, Lawrence County, Pennsylvania, on 3 October 1866. Past research of her husband Archibald Thompson’s War of 1812 pension record provided her maiden name and their August 1806 marriage date. 

Several online trees identified Martha as the daughter of James Williamson and Jane McEwen, but James’s will contained no daughter named Martha. Using FamilySearch’s Full Text search, I found deeds listing Martha and Archibald Thompson as heirs of Alexander Kerr, alongside several people with the Williamson and Chambers surnames. This unexpected connection opened a new research path. 

Building the Documentary Foundation 

Alexander Kerr’s will didn’t name Martha Williamson directly, but it did name Jean Williamson and Alexander Williamson as heirs of Alexander Kerr’s sister, Martha Kerr “otherwise Williamson.” The Chambers heirs were children of his sister Mary Kerr “otherwise Chambers.”¹

I found multiple deeds mentioning Alexander Kerr and his land “Fairy Green,” inherited by the Williamsons and Chambers. I tracked the land as each heir sold their interest, transcribing each deed and uploading transcriptions to Claude Sonnet 4.5 (AI) to create abstracts following a standardized format I’d developed.²

The Power of AI-Assisted Analysis 

The AI partnership proved valuable for maintaining consistency across more than a dozen abstracts. Claude extracted information precisely as it appeared—grantor, grantee, consideration, property descriptions, boundaries, chains of title, witnesses, and recording information. Each abstract included source analysis: who provided what information, whether it was primary or secondary evidence, and what the document proved directly or indirectly. I double-checked each data point as well as the source analysis and then added my own image analysis to my document. 

As each new deed added information, I would query Claude about what it might mean in relation to the families or land. The AI excelled at pattern recognition and organizing complex information, though I had to remain vigilant. Sometimes Claude would get relationships backwards—once telling me Jean and Alexander Williamson were Mary Kerr’s children when the opposite was true. Even with multiple deeds and the will transcript uploaded, I still needed to carefully verify every conclusion. 

The Breakthrough Moment 

But brainstorming with AI proved valuable — just as brainstorming with fellow genealogists—and is more readily available when transcribing at 10 o’clock at night! I had added the new branches to my Ancestry tree with Jean and Alexander as siblings, initially thinking Alexander Williamson was Martha Kerr’s husband. But something nagged at me, and it just felt “off.” 

I set the feeling aside to keep transcribing, thinking that it may resolve itself. Then I found deeds naming Jean Williamson “wife of Charles Haney” as an heir, and another deed naming Martha Williamson “wife of Archibald Thompson.” In discussing this with Claude, it asked, “Well how is Jean related if Alexander is the heir?” 

It dawned on me that I had missed a generation. Alexander wasn’t Martha Kerr’s husband—he was her son. Therefore, Jean Williamson was Martha’s daughter, and Martha Williamson (daughter of Alexander) was not Martha Kerr’s daughter, but her granddaughter. That made much more sense! The “nagging” feeling was resolved. 

Confirming the Hypothesis 

The breakthrough came with a 1832 power of attorney from “Esther Buchanan, late Esther Williamson, widow and relict of Alexander Williamson, deceased.” Alexander Williamson’s widow had remarried, moved to Mercer County, and was still alive 41 years after her first husband’s death, retaining dower rights in the Fairy Green tract. ³

The 1811 deed provided crucial evidence: it stated that Alexander Williamson “died intestate, leaving a widow and one child to wit Martha Williamson, since intermarried with the said Archibald Thompson.” This confirmed that Martha was Alexander’s sole heir through whom the Fairy Green land passed. Significantly, Esther Buchanan chose Archibald Thompson—her son-in-law—as her attorney in fact, referring to him as “my trusty friend” in the power of attorney. Despite the distance, she entrusted him to handle the dower rights in Washington County—in the very property her daughter had inherited from Alexander decades earlier. 

The naming patterns now made perfect sense. Martha Williamson was named for her grandmother Martha Kerr, mother of Alexander Williamson. Esther P. Thompson (daughter of Martha Williamson and Archibald Thompson) was named for her grandmother Esther [—? —] Williamson Buchanan. Another Thompson child was named Alexander Williamson Thompson—after his grandfather. This family used family names extensively, and recognizing the pattern confirmed the generational structure. 

What AI Does Well 

Consistency: Every abstract followed the same format without the transcription fatigue that affects manual work. No overlooked boundaries, forgotten witnesses, or inconsistent formatting. 

Pattern Recognition: Claude quickly identified connections across documents—chains of title, family relationships, chronological sequences that might take hours to notice manually, especially if the work is done over several days. 

Detail Preservation: Every spelling variation (Jean/Jehin, Haney/Heney/Heiney, Williams/Williamson), every mark versus signature, every measurement appeared accurately in the abstracts. 

Asking Questions: Even when Claude got details wrong, the questions it asked—“How is Jean related if Alexander is the heir?”—prompted critical thinking that led to breakthroughs. 

What Requires Human Expertise 

Claude couldn’t determine I was missing a generation—I had to notice something felt wrong. AI couldn’t search for additional records or make inferential leaps beyond the documents I provided. Understanding historical context, legal implications of dower rights, and the significance of naming patterns required my expertise. 

Most importantly, I verified everything against original documents. AI can misread relationships or make logical errors. The genealogist remains responsible for accuracy. 

The Collaborative Advantage 

The real power came from combining AI efficiency with human knowledge. Claude handled the mechanical aspects of deed abstracting with perfect consistency, freeing me to focus on interpretation, hypothesis testing, and creative problem-solving. 

Working through problems with AI was like having a methodical colleague who never tired, never skipped steps, and asked logical questions. Sometimes, just explaining the evidence to Claude revealed the answer. The technology excels at organization and pattern recognition but works best under the direction of a professional genealogist who provides context, verification, and interpretive expertise. 

Conclusion 

AI didn’t replace my genealogical expertise—it amplified it. By maintaining perfect consistency across more than a dozen deed abstracts, Claude allowed me to focus on the detective work: noticing what felt wrong, recognizing naming patterns, understanding family dynamics, and connecting documentary evidence into a coherent narrative. 

When I finally documented that Martha Kerr had children, Jean and Alexander Williamson, that Alexander married Esther and had a daughter, Martha, and that Esther lived to see her daughter’s family thrive, it wasn’t just finding names—it was confirming an entire family structure through systematic documentary analysis. 

That’s the power of the partnership: AI maintains the details while humans make connections that transform documents into family stories. 

Disclosure: I collaboratively wrote this blog with Claude Sonnet 4.5 based on our past chat histories of the will and deed abstracts to recreate what was accomplished. I also wrote a draft of this blog post and uploaded it to Claude to give it context to what I recognized about the family naming patterns and other missing background information. Much of my draft was incorporated into the finished blog. Claude helped clarify my thoughts into a coherent narrative. Still, during today‘s chat creating this blog post, I had to correct Claude on several points and remind it not to hallucinate. 

 

1. Washington County, Pennsylvania, wills, vol. 1:23-24, Alexander Kerr, 4 Jun 1783, recorded 19 Jul 1783; “Will books v. 1-2 1781-1814,” browsable images, FamilySearch (familysearch.org : viewed 27 Nov 2025); citing image group number (IGN) 5537968, item 1, images 18-19.
2. John H. Campbell, chief draftsman, South Strabane Township, Washington County, Pennsylvania, map, No. 1, “Fairy Green” and No. 2 “Witches Haunt” survey, Alexander Kerr, 11 Sep 1786, patented 21 Mar 1787 by Thomas Kerr; “r017Map3198WashingtonSStrabaneWeb.pdf,” digital image, Pennsylvania Historical and Museum Commission (phmc.state.pa.us/bah/dam/rg/di/r17-522WarranteeTwpMaps/r017Map3198WashingtonSStrabaneWeb.pdf : downloaded 13 Dec 2025); citing Pennsylvania State Archives, RG-17 Records of the Land Office, Warrantee Township Maps {series #17.522}, 2 Nov 1921.
3. Washington County, Pennsylvania, deeds, vol. 42, 2R:305, power of attorney, Esther Buchanan to Archibald Thompson, 12 May 1832, recorded 9 Dec 1832; “Deeds v. 42-43, 2R-2S 1834-1836,” browsable images, FamilySearch (familysearch.org : viewed 12 Dec 2025); citing image group number (IGN) 8084642, image 170.