Ancestors and Algorithms

Website Contact the practice

Ancestors and Algorithms Where Family History meets Artificial Intelligence. Uncover your ancestors with the power of AI.

Clara J. Ashby was married less than three years.She died in August of 1875, at home near Holden, Missouri. Her daughter...
10/28/2026

Clara J. Ashby was married less than three years.

She died in August of 1875, at home near Holden, Missouri. Her daughter Ida was seventeen months old. Missouri wasn't recording deaths yet, so there's no death certificate for Clara, and her grave hasn't turned up.

Two years later, Martha Dowell married Clara's widower and took on a three-year-old girl as her own.

Ida died in 1938, at sixty-three. A clerk asked her daughter Opal for Ida's parents, and Opal gave the name of the only grandmother on that side she had ever known. Martha's name went into the box marked mother, and for nearly ninety years the family believed it.

Opal wasn't lying, and she wasn't careless. She was telling the truth about who had loved her mother.

The family in this week's episode is a composite I built for the show, but the pattern is real, and it's sitting in more trees than we'd like to think. The woman who raised a child gets remembered as the woman who gave birth to that child.

What brought Clara back was three words her husband wrote on a pension form in 1898: "By first wife."

Episode 61 shows how I found her, step by step, with two free AI tools. Listen at link in the first comment.

Have you ever found a stepparent remembered as a parent in your own tree?

A death certificate can answer your question in plain ink and still be wrong.Episode 61 of Ancestors and Algorithms is o...
10/27/2026

A death certificate can answer your question in plain ink and still be wrong.

Episode 61 of Ancestors and Algorithms is out today: "The Informant Test: Death Certificates and AI."

For nearly ninety years, one family trusted the name typed in the box marked mother on a 1938 Missouri death certificate. Most of us would have trusted it too. But the woman who answered the clerk's questions was born twenty-four years after the birth she described.

In this episode I walk through a case I built to show you how to question that informant with AI:
• Date every fact on the certificate against the informant's own life
• Ask AI to predict how that person would most likely get it wrong
• Build a test plan from the earliest records made by other people
• Count the witnesses behind your records, not the pieces of paper

Two free tools, Claude and ChatGPT, and three copy-paste prompts you can run tonight on the certificate you trust most. And the record that finally settled the case was one neither AI plan suggested.

Listen free wherever you get your podcasts, or start at the link in the first comment.

There's a sentence I could have written at the end of Episode 60, and I didn't."Leonard Ostlund is not Lorraine's father...
10/24/2026

There's a sentence I could have written at the end of Episode 60, and I didn't.

"Leonard Ostlund is not Lorraine's father."

Here's why it isn't in my notes. One of Lorraine's matches, M03, is Leonard's granddaughter, and she shares 297 centimorgans with Lorraine. If Leonard is Lorraine's uncle, M03 is her first cousin once removed. The Shared cM Project reports 102 to 980 cM for that, and 297 fits comfortably. If Leonard is her father, M03 is her half niece. The reported range there is 492 to 1,315, and 297 falls well below the bottom of it.

That's strong evidence. It isn't proof that something is impossible. The project's published ranges already trim off the extreme outliers, so a result outside them means rare, not never. My notes say "very unlikely," and that's the sentence I'll stand behind.

There's a second sentence I won't write yet: which of the two remaining brothers it was. The Army files that might have said so may well have been lost in the 1973 fire at the National Personnel Records Center, which the National Archives estimates destroyed eighty percent of Army files for soldiers discharged between 1912 and 1960. So the next step is target testing, which means kindly asking a living descendant of each brother whether they'd test.

This is what the Genealogical Proof Standard looks like in practice. Reasonably exhaustive research meant working every match in the 90 to 400 cM band, not only the ones with trees. Analysis and correlation meant two independent DNA groups joined by one documented marriage. And a soundly reasoned written conclusion meant choosing the words "very unlikely" on purpose.

AI is your research assistant, not your researcher. In this episode the AI sorted the list, and it was good at it. The words in the conclusion are mine, and so is the decision not to write "impossible."

What's one sentence in your own tree that you've written more confidently than the evidence allows?

The full case is in Episode 60 at the link in the first comment.

Four tools, four jobs. Here's what I used to sort a DNA match list this week, and why each one got the job it got.1. Cla...
10/22/2026

Four tools, four jobs. Here's what I used to sort a DNA match list this week, and why each one got the job it got.

1. Claude: the sorter
Claude took a coded table of twenty matches and sorted them into grandparent groups under five written rules. It named the link behind every placement, set one bridge match aside, and refused to place two matches it had no evidence for. I used an incognito chat (the ghost icon on a new chat), and the free plan's Sonnet 5.5 model is what I ran it on.
Cost: free plan works.

2. Gemini Notebook: the namer
Formerly NotebookLM, now at notebook.google.com. It answers only from the notes you give it and puts a numbered citation on every claim, so each step toward "this is my father's family" points back to a line you wrote. When one group's trees only shared surnames, it said so instead of inventing a couple.
Cost: free with a Google account.

3. Perplexity: the lookup
It searches the web and shows its sources, so it's the right tool for published shared-DNA ranges. But its opening sentence contradicted the body of its own answer, twice. The details were right and the summary was wrong.
Cost: free, no account needed.

4. DNA Painter's Shared cM Project tool: the source
This is the reference the AI tools quote. Type in your number and every relationship that fits stands out on the chart. Every range went through here before it went into my notes.
Cost: free, no account needed.

The pattern I'd take from this: give each tool one job, and let the last word go to the source, not to a chatbot.

One more cost to know about. On Ancestry, seeing more than three shared matches per person needs a paid membership. The method still works with three, but plan for it.

Which of these four have you used for DNA work? I'd love to hear what's worked for you.

The full walkthrough is in Episode 60 at the link in the first comment.

The space for her father's name has been empty for eighty-two years.Lorraine was born in Brisbane in 1944. She's a compo...
10/21/2026

The space for her father's name has been empty for eighty-two years.

Lorraine was born in Brisbane in 1944. She's a composite I built for this week's episode, but every record and every match pattern in her story is the kind you'll find in your own results.

Her birth registration names her mother, Joan, and leaves the father's line blank. Joan raised her alone and told her exactly one thing about him. He was a Yank.

That isn't a wild story for that city and that year. The National Archives of Australia says nearly eighty thousand Americans were in Brisbane at the peak of the war. So "a Yank" narrows it down to tens of thousands of men.

For her eightieth birthday, Lorraine's daughter gave her a DNA test. It came back with more than two thousand matches.

I know a lot of you have been exactly where she was that afternoon. You open the list and scroll past names you don't recognize. A few have trees. Plenty don't. Nothing on that screen tells you which side of the family anyone is on.

That feeling isn't a lack of skill. It's a sorting problem.

By the end of Episode 60, Lorraine has something she's never had before: a surname, a couple, a city and two names. We don't know yet which brother it was, and I won't pretend we do. The last step means writing to real people, gently, and nobody owes her an answer.

For eighty-two years that space was blank. Now it holds a short list.

Have you ever opened a match list and not known where to start? Tell me about it below.

Hear Lorraine's whole case at the link in the first comment.

Your DNA match list has two thousand names on it and not one label. This week I'll show you how to sort it.Episode 60 of...
10/20/2026

Your DNA match list has two thousand names on it and not one label. This week I'll show you how to sort it.

Episode 60 of Ancestors and Algorithms is out: "Sort Your DNA Matches: The Leeds Method and AI."

The idea fits in one sentence. You have four grandparents, and almost every cousin on your match list is related to you through only one of them. Get everyone standing beside the right grandparent and your list turns into four groups.

Dana Leeds created this method in 2018. By hand, it's a spreadsheet and a long weekend of clicking. With AI, the sorting takes about a minute, and you stay in charge of what it means.

I teach it on one case, a composite I built for the episode. Lorraine was born in Brisbane in 1944, with a blank where her father's name should be. Her mother told her one thing about him. He was a Yank.

Here's what you'll learn:
• How to build a coded match table, so no living person's name ever goes into a chatbot
• Five written rules that keep the AI from guessing
• The shuffle test, a check I haven't seen taught anywhere yet
• How one "bridge" match pointed to the marriage that joined her father's two families

I'll be straight with you about the ending. This one is a partial answer. We went from two thousand matches to two brothers, and that's where the evidence stops for now. I'd rather show you exactly where that line is than pretend it isn't there.

The 26-page Companion Guide picks up right where the episode stops.

Listen free on our website, or wherever you get your podcasts.

The sentence I won't write yet.At the end of this week's episode, I found a baptism from 1846 in Neuenkirchen, a village...
10/16/2026

The sentence I won't write yet.

At the end of this week's episode, I found a baptism from 1846 in Neuenkirchen, a village in the Altes Land about sixteen miles from Hamburg. Friedrich, son of Hinrich Eckhoff and his wife Anna. Right name, right year, right father, in the same district as his wedding witness's village.

It's tempting to write "Friedrich Eckhoff was born in Neuenkirchen" and close the file. I won't, not yet. Here's why, and it's the Genealogical Proof Standard doing its job.

1. Explain a conflict before you choose. His records say Hanover in 1870 and Prussia in 1880. That isn't two places. Prussia annexed the Kingdom of Hanover in 1866. Same man, same village, new border.

2. Weigh who answered, not how many papers agree. His death certificate had a birthplace line, but his Baltimore-born son filled it in, and the obituary most likely came from the same family source. One firsthand answer, Friedrich himself telling a pastor in 1874, outweighs all of them.

3. Correlate. Name, year and father lining up in records made an ocean apart is strong evidence.

4. Say how strong. At least one other Neuenkirchen in the province has no church book images online, so I can't rule out a second Friedrich from my desk. And the marriage entry says "aus," which means "from." It can mean where he was born, or only where he last lived.

So here's what I will write: it's probable that our Friedrich is the boy baptized in Neuenkirchen in 1846, and it isn't proved until those other parishes are checked. "Probable" is a respectable word.

AI did a lot of work in this case. It interviewed me, tested my records and read old German script. But it didn't decide anything, and I didn't let it. AI is your research assistant, not your researcher.

The whole case is in Episode 59 at the link in the first comment.

What's a conclusion in your own tree marked "proved" that might really be "probable"?

Nobody ever asked Friedrich for his village.That's the line from this week's episode I keep coming back to.Picture a coo...
10/15/2026

Nobody ever asked Friedrich for his village.

That's the line from this week's episode I keep coming back to.

Picture a cooper in Baltimore, building barrels for more than thirty years. He showed up for every census. He naturalized in 1875. When he died in 1903, his son Henry answered the questions on the death certificate. Henry was born in Baltimore and had never seen the old country, so he said what he knew: Germany.

Every clerk who ever wrote Friedrich down handed him a blank the size of a country. The census taker wanted a state. The naturalization papers wanted to know which king he was giving up. Not one of those forms had room for the place he grew up.

Except one.

In 1874, a German Lutheran pastor in Baltimore wrote Friedrich's marriage into the church book, and he wrote down where Friedrich came from: "aus Neuenkirchen in Hannover." From Neuenkirchen in Hanover. Son of Hinrich Eckhoff. In the Low German Friedrich grew up speaking, Hinrich is Henry. It's the same name his son carried through Baltimore.

The answer was never lost. It was waiting for someone to ask in the right place.

Friedrich is a composite, but the pattern is real, and I'd bet some of you have an ancestor sitting in that same spot right now. Episode 59 walks you through how to find the one record that asked. Listen free at the link in the first comment.

Who in your tree has a birthplace that's only one word long?

Three free AI tools, three different jobs. Here's how I split the work in this week's episode, and why I didn't ask any ...
10/14/2026

Three free AI tools, three different jobs. Here's how I split the work in this week's episode, and why I didn't ask any one tool to do it all.

ChatGPT: the interviewer.
I used it in a Temporary chat set to "Unpersonalized," so it couldn't borrow anything I'd told it about another family. Its job was to ask me questions, not answer mine. Eight questions later I had a written Question Card. Free accounts began moving to a new model, GPT-6 Luna, on October 8, so your questions may come in a different order than mine. Judge your card by whether it names one person, one fact and a record that could hold it.

Perplexity: the fact-checker with receipts.
I needed current facts about old record forms, with sources attached. Perplexity searches the web and puts a small source link beside each claim. No account needed. I ran the same prompt twice, a day apart. It got a different thing wrong each time, and the source links are how I caught both.

Gemini in Google AI Studio: the handwriting reader.
For an 1874 church entry in old German script, I used Gemini through aistudio.google.com, which is separate from the Gemini app. It's free to use as long as you haven't added an API key. I asked it to list every place name and the letters it wasn't sure of, because one misread letter can move you fifty miles. One caution: on the free tier, people at Google may review what you upload. A church record from 1874 is fine. A living cousin's birth certificate isn't.

The click-by-click setup for all four is in Episode 59 at the link in the first comment.

Which of these have you tried on German records, and how did it go?

Every record says "Germany." Not one of them names the village.If you've got an immigrant ancestor like that, this week'...
10/13/2026

Every record says "Germany." Not one of them names the village.

If you've got an immigrant ancestor like that, this week's episode of Ancestors and Algorithms is for you.

Friedrich "Fred" Eckhoff was a cooper in Baltimore. The 1870 census says he was born in Hanover. The 1880 census says Prussia. His death certificate and his obituary both say Germany. Seven documents, one family rumor ("somewhere near Hamburg"), and no village anywhere. Friedrich is a composite, but every record in his story is the kind you'll find for your own immigrant.

Here's what I learned working his case: the problem usually isn't the records. It's the question we keep asking them.

In Episode 59, "The Right Question: Find Your German Village," I walk you through it one click at a time with three free AI tools:

• ChatGPT interviews you and turns a vague wish into a written Question Card
• Perplexity tests every record you hold: could that form, in that place and year, even hold a town of birth?
• Gemini in Google AI Studio reads old German handwriting and flags the letters it isn't sure of

If you've never opened a chatbot before, you can follow every step. That was the whole idea.

Listen free at link in the first comment, or wherever you get your podcasts.

Address

Salt Lake City, UT

Website

Alerts

Be the first to know and let us send you an email when Ancestors and Algorithms posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Practice

Send a message to Ancestors and Algorithms:

Shortcuts

Share

Category

Other Genealogists in Salt Lake City

Show All