Or: What happens when a genealogist turns artificial intelligence loose on 200-year-old tax data

Every family historian knows the thrill of discovery—that moment when scattered pieces of evidence suddenly click into place, revealing relationships that were hidden in plain sight. Recently, I decided to enlist some high-tech help in solving a genealogical puzzle that had been nagging at me: the Wright family’s settlement patterns in Wayne County, Kentucky, from 1802 to 1813.

My 6th great-grandfather Amos Wright lived in Wayne County during those crucial early years before heading off to Washington County, Indiana around 1808. But Amos wasn’t alone—the tax records showed multiple Wrights scattered across the county’s creeks and hollers. Were they related? How were they connected? And could artificial intelligence help me see patterns I’d missed?

Inspired by a podcast episode about using AI to analyze tax data, I decided to conduct my own experiment. What followed was part genealogical investigation, part technology test drive, and entirely fascinating.

Setting the Stage: Garbage In, Garbage Out

Before I could turn my AI assistants loose on the data, I had to confront every genealogist’s nemesis: inconsistent 19th-century spelling. Standard genealogical practice in transcribing is to write it exactly as it is in the document you are viewing. However, the minefield of variants—”Right” versus “Wright,” “Amos” versus “Amus”— if left as is in my spreadsheet— would likely confuse any algorithm into thinking we were dealing with entirely different people.

So, cleanup — on a copy of my originally transcription — was my first task. Each variant had to be standardized, each waterway name corrected. After all, what good is artificial intelligence if you’re feeding it artificial confusion?

Round One: ChatGPT Takes the Stand

Armed with my cleaned dataset, I posed my first question to ChatGPT: “Taking into consideration the columns ‘Person Chargeable w/ Tax’, ‘Water Course’, ‘Name Entered’ and ‘Name Surveyed’, suggest possible relationships between the various taxpayers.”

ChatGPT approached the problem like a seasoned detective, immediately zeroing in on geographic clustering:

The Beaver Creek Connection: The AI noted that Amos Wright, Evan Wright, and Philip Copple all had land along Beaver Creek, suggesting they were neighbors—or possibly kin. This observation proved remarkably astute, as I knew from my research that Amos and Evan were indeed brothers, and Philip Copple had married one of Amos’s daughters.

The Henry Biggs Mystery: ChatGPT flagged Henry Biggs as a recurring figure whose name appeared in both the “Name Entered” and “Name Surveyed” columns for multiple Wright properties. The AI theorized that Biggs was either a surveyor or someone whose land abutted Wright holdings—a hypothesis I hadn’t fully explored.

The Surveyor Theory: Most intriguingly, ChatGPT suggested that Amos Wright himself might have been active as a surveyor, noting his name appeared in surveying columns. This painted a picture of Amos as a community leader—a “patriarch,” as the AI put it.

But here’s where it got interesting: ChatGPT completely ignored William Wright, Elijah Wright, John Wright, and Isaac Wright in its initial analysis. When an AI overlooks data, it’s worth asking why.

The Plot Thickens: New Evidence Emerges

Suspecting I’d missed some entries, I went back to the original tax records. Sure enough, I’d overlooked Samuel Wright, Moses Wright, and Jesse Wright. After adding these men to my dataset, I decided to test something: would ChatGPT give me the same answers with the expanded data?

The answer was no—and that’s when this investigation took an unexpected turn.

Day Two: The Same Question, Different Answers

Using identical prompts with the revised dataset, ChatGPT offered notably different conclusions. Instead of focusing solely on Beaver Creek clustering, it now suggested:

• Chronological Settlement Patterns: William Wright at Elks Spring, Amos at Beaver Creek, Jesse at Meadow Creek might represent an “order of settlement”
• Land Transfer Networks: The appearance of names like Stacter and Biggs as “predecessor patentees” suggested the Wrights were systematically acquiring nearby tracts
• Three Wright Lines: The AI now theorized three different Wright family lines establishing themselves simultaneously in Wayne County around 1802

This inconsistency raised a red flag. If AI is supposed to be deterministic, why were my answers changing? It reminded me that these tools, powerful as they are, aren’t infallible oracles—they’re sophisticated pattern-matching systems that can be influenced by data variations.

Revealing the Facts: Guided Analysis

At this point, I decided to level the playing field. I fed both ChatGPT and Claude AI the facts I’d uncovered through traditional research:

• William, Evan, and Amos Wright were brothers
• Philip Copple was Amos Wright’s son-in-law
• There were two different John Wrights in the records
• Elks Spring was a tributary of Beaver Creek
• Biggs and Stacter might have been county surveyors

With these revelations, both AIs refined their analyses significantly. ChatGPT now correctly identified the geographic relationships (William and Amos as neighbors along the same creek system) and properly distinguished between professional relationships (surveyors) and family ties.

Claude AI Enters the Investigation

When I presented the same data to Claude AI, it took a somewhat different approach. Where ChatGPT had been expansive in its theorizing, Claude was more cautious—but it also made some notable errors.

Claude correctly identified the surveyor relationships and community connections, but missed the marriage connection between Philip Copple and the Wright family entirely. More concerning, it made what appeared to be a complete fabrication, claiming that “Jesse’s land is often ‘Name Entered’ or ‘Name Surveyed’ under William Wright’s entries”—something that simply wasn’t in the data.

This reminded me of a crucial lesson: AI can hallucinate connections that don’t exist, just as easily as it can miss ones that do.

The Smoking Gun: Details Only Humans Notice

One detail that only Claude mentioned caught my attention: Evan Wright consistently appeared in the “Blacks” column of the tax records, indicating he was a slaveholder. This wasn’t a relationship pattern—it was a social and economic marker that added crucial context to understanding the Wright family’s standing in Wayne County society.

What the Evidence Reveals

After running this comparative analysis, several patterns emerged that neither AI fully captured on its own:

Geographic Clustering: The Wright brothers clearly settled along connected waterways, with William at Elks Spring (a tributary of Beaver Creek) and Amos directly on Beaver Creek itself. This wasn’t coincidence—it was family strategy.

Professional Networks: The recurring appearance of Henry Biggs and Samuel Stacter in survey records likely reflects their roles as county officials rather than family relationships, though neighboring land ownership remains possible.

Generational Succession: The appearance of younger Wrights (Elijah in 1809, likely William’s son coming of age) demonstrates how tax records can reveal family demographics over time.

Separate Wright Lines: Jesse Wright’s consistent association with Meadow Creek and different survey patterns suggest he represented a distinct Wright family line—not necessarily related to Amos, William, and Evan despite sharing the surname.

The Verdict: AI as Research Assistant, Not Replacement

So what’s the takeaway from this technological experiment? AI proved remarkably useful for pattern recognition and hypothesis generation, but it also demonstrated significant limitations:

Strengths:

• Excellent at identifying geographic and temporal clustering
• Good at spotting recurring names and potential professional relationships
• Capable of generating testable hypotheses about family structures

Weaknesses:

• Inconsistent results with identical queries
• Tendency to hallucinate connections not present in the data
• Missed obvious relationship indicators (like known family connections)
• Limited ability to distinguish between different types of relationships

The Human Element

Perhaps most importantly, this experiment reinforced why traditional genealogical research remains irreplaceable. The AI’s most accurate insights came only after I provided the human context—the family relationships I’d painstakingly documented through other sources.

Without that foundation, the AI was essentially reading tea leaves, finding patterns that may or may not reflect historical reality. With it, the technology became a powerful tool for exploring implications and connections I might have missed.

A Final Twist: The Copyright Question

One unexpected discovery emerged during this process: ChatGPT had apparently trained on some of my own blog posts, citing them as sources without my explicit permission. This raises intriguing questions about how AI systems acquire their knowledge—and reminds us that the information we freely share online may someday be reflected back to us in unexpected ways.

Closing the Case

The Wright family tax records of Wayne County, Kentucky tell a story of strategic settlement, family networks, and community building in the early American frontier. AI helped illuminate some of these patterns, but only human knowledge could properly interpret them.

For fellow genealogists considering similar experiments, my advice is this: use AI as you would any other research tool—with curiosity, skepticism, and the understanding that technology amplifies both our insights and our errors. The algorithms can spot patterns we miss, but they can’t replace the detective work, critical thinking, and contextual knowledge that make family history come alive.

After all, behind every tax record entry was a real person making real decisions about where to live, whom to marry, and how to build a life on the Kentucky frontier. No algorithm, however sophisticated, can fully capture that human story—but it might just help us see new chapters we hadn’t noticed before.

What patterns have you discovered in your own family’s records? Have you experimented with AI in your genealogical research? I’d love to hear about your experiences—and any Wright or Copple connections you might have uncovered along the way.

“AI Detective: Wayne County Tax Records,” Claude Sonnet 4, chat initiated by user Cathy Dempsey, Claude (https://claude.ai/chat/818f967f-d918-42f2-9293-873902f1cdf9 : accessed 28 August 2025)