I recently solved an 87-year-old family mystery. And yes, I am ridiculously proud of myself. Not because I magically found one document nobody else had ever seen. Actually, almost the opposite.
I solved it because a whole bunch of people had found pieces of the answer before me. Some of them got things wrong. Some of them got tantalizingly close.
One woman wrote a letter in 1939 and never mailed it. Her daughter found it decades later and resumed the search. Another branch of the family preserved emails from 2007. Somebody built a family tree with the wrong interpretation of a Virginia court case but attached the right document. Another researcher spent years studying an entirely different Johnson family and helped convince me that the line I was following probably wasn’t mine.
Then there was DNA and fragments of evidence:
- Digitized court archives.
- FamilySearch.
- Ancestry.
- Old newspapers.
- Handwritten family records.
And, yes, AI helping me transcribe, organize, compare and interrogate nineteenth-century documents without losing my mind.
Eventually all those fragments collided and I figured out who my third-great-grandfather’s mother was, identified his long-lost brother, reconstructed a family that had been obscured for generations, and got the answer back to the descendants of the woman who had been searching for it since 1939.
The full genealogy story — including the Civil War-era murder, the wrong Quaker ancestry, the Virginia Chancery cases, the lost brother who went to Texas, and the 1939 letter that survived long enough for me to finally answer it — is over on VintageReveries, because that is exactly where something this wonderfully old, messy and human belongs.
Read: Standing on the Shoulders of Other Researchers: How I Broke an 87-Year-Old Johnson Brick Wall
But the process has been rattling around in my brain for another reason.
It reminded me that the way I do genealogy is really not that different from the way I approach almost everything else.
I stopped trying to prove myself right
The brick wall involved a man named Joseph James Johnson. Which is almost comically unfortunate from a research standpoint. Johnson is one of the most common surnames in the United States, and there were already plenty of public family trees confidently assigning him to a particular Quaker Virginia family.
For a while, I tried to make that family fit. There were reasons it seemed plausible. There was a family story that appeared to support it. A few DNA matches looked promising. Except the more I tested the theory, the worse it got.
So I started doing something that I think is useful far outside genealogy: I tried to disprove myself. Instead of asking, “What evidence supports this?” I started asking, “If this were true, what else should I be seeing?”
If this really were Joseph’s family, descendants of his proposed siblings should appear among my DNA matches. They mostly didn’t. If the relationship were correct, the documentary trail should become clearer as I built outward. It didn’t.
Every time I tried to stress-test the hypothesis, it got weaker. Eventually I deleted the whole relationship from my working tree.
That is a surprisingly hard thing to do when you have invested hours into proving something. There is a little sunk-cost monster in the brain whispering: “But maybe if you just search one more database…”
No.
Sometimes the breakthrough is admitting the current model sucks.
Then I found useful bad data
The eventual breakthrough came from another person’s family tree. It was wrong. Or at least partially wrong. And incredibly useful.
The tree was not evidence. But it pointed me toward evidence. That distinction ended up being everything. The researcher had attached a Virginia Chancery Court case involving people whose names overlapped with mine. Their interpretation of the family structure was not correct, but the document itself was exactly what I needed.
I opened the original scanned jpgs of handwritten documentation. Then I found the related case. Then another record.
Then I rebuilt the family from the evidence rather than from the tree. And suddenly the DNA made sense.
I love this because we spend so much time talking about “bad data” as though the only useful dataset is a pristine one. Reality is messier.
- Bad data can contain good signals.
- An incorrect conclusion can contain a correct source.
- A failed experiment can narrow the search space.
- A person saying, “I tried this and couldn’t make it work,” may save the next researcher fifty hours.
The important skill isn’t avoiding imperfect information. It’s knowing what level of trust to assign it.
AI did not solve the mystery
I want to make this distinction because I use AI constantly, and this project is a perfect example of what I think it is actually good for. AI did not discover my ancestors. It didn’t magically know which Johnson was mine. It couldn’t decide whether a DNA match was genealogically meaningful. It certainly could not replace reading the original documents.
What AI could do was help me work faster. I used it to help decipher terrible handwriting. I used it to extract names and relationships from long legal documents. I asked it to explain nineteenth-century legal terminology that was unfamiliar to me. I used it to organize competing hypotheses. I could dump an ugly pile of notes into it and say, “Here is everything I know. Show me where these sources agree, where they conflict, and what I still have not proved.“
Then I went back to the source material and checked. That last step matters.
AI became the research assistant who could tolerate me saying, for the seventeenth time at 1 a.m., “WAIT. WHAT IF HOLLAND IS A WOMAN?”
But the proof still had to come from the records.
That is how I increasingly think about AI in any serious research environment. Not oracle. Leverage. The value wasn’t that AI knew the answer. It dramatically reduced the friction between having a question and being able to test it.
The real breakthrough was collaboration across time
This is the part that affected me the most.
In January 1939, a young woman named Goldie Johnson sat down in Clayton, Oklahoma and wrote to the Census Bureau. She wanted to know what had happened to her ancestor Joseph Johnson’s older brother. She knew his name was something like Euel or Ewal. She believed he had gone to Texas. She wrote the letter. And then, for reasons nobody knows, she never mailed it. She kept it. Her daughter found it decades later.
Her daughter resumed the research and exchanged emails with another branch of the family in 2007. Those people saved the emails. One of their descendants sent them to me in 2026.
I found records the earlier researchers either couldn’t access or didn’t know existed. DNA testing gave me evidence Goldie could not have imagined.
And then came perhaps my favorite part.
The daughter who had continued Goldie’s research was no longer active online. I agonized over whether contacting her family would be intrusive. I asked a genealogy group what they thought. Almost everyone said, “reach out.” So I contacted her granddaughter. She responded within hours. And I was able to send the family the records that finally answered the question Goldie had written down 87 years earlier.
I don’t know how to reduce that to “genealogy.” That is human information moving through time.
Publish the imperfect work
That experience reinforced something I already believe strongly:
Put useful things on the internet.
Not everything. Not private information that isn’t yours to share.
But your work? Your research? The documents you’ve transcribed? The weird historical thing you spent three weekends figuring out?
Publish it.
Somebody may need it.
One reason this mystery survived for so long is that the people working on it were scattered across states, generations and platforms. But the reason I could solve it is that enough of their work survived. Someone attached a court case. Someone preserved a letter. Someone kept an email. Someone uploaded a tree. Someone indexed a marriage record. Someone digitized a courthouse book. Someone took a DNA test.
None of those people individually solved my problem. Together, they practically built me a ladder.
We are all leaving breadcrumbs
Genealogy also does this slightly uncomfortable thing to your sense of time. You start looking at a person living in 1859 and realize that they weren’t “living through history.” They were living through Thursday. They were worrying about money. Getting married. Fighting with relatives. Moving because something went wrong. Losing property. Writing letters they forgot or chose not to mail. Trying to understand what the hell was happening around them.
Then 167 years later, somebody is staring at their name on a computer screen trying to reconstruct why they did what they did.
Which inevitably makes me think about us.
We are also somebody’s archive. Our emails. Our websites. Our photographs. Our ridiculous social posts. The things we document carefully and the things we assume everyone will remember.
They won’t.
Context disappears frighteningly fast. And maybe that is part of why I write so much. Why I save things. Why I keep a blog that has wandered across careers, vintage clothing, genealogy, technology, birds, family history and whatever other rabbit hole currently has me by the ankle.
At the time, those things don’t necessarily look connected. Neither did the Virginia lawsuit, the 1939 letter, a Tennessee marriage record, a Texas voter roll and a handful of DNA matches.
Until they did.
So yes. I broke a genealogy brick wall. But mostly I was the latest person in an 87-year relay race who happened to be holding the baton when enough pieces finally existed to cross the finish line.
And then I got to hand the answer back to the family.
I’m going to be obnoxiously proud of that one for a while.
And if you want to know exactly what was on the other end of that 87-year relay race — Holland, Joseph, Ewel, Goldie, the wrong family tree, the DNA, the true connection to Alexander Doniphan, the Virginia lawsuits and the rather spectacular amount of nineteenth-century family drama — I told the complete story on VintageReveries
0 Comments