(Google Gemini Notebook Infographic - Genealogy Truths and Certainty)
Here is my discussion with ChatGPT5:
A) Prompt: If you could tell us ONE truth we’re not ready to hear about genealogy research, what would it be?
OpenAI ChatGPT5 response:
If I could tell genealogists one truth they may not be ready to hear, it would be this:
But sometimes the records simply do not exist. Sometimes they existed but were destroyed. Sometimes an ancestor deliberately concealed something. Sometimes several records contradict one another. And sometimes the evidence will never allow us to distinguish between two reasonable possibilities.
The dangerous moment comes when we say:
That's especially tempting with brick-wall ancestors. We find a family that fits the age, location, naming pattern, DNA evidence and historical circumstances. Everything seems to line up. We build the connection into the tree, write the biography, and five years later we've forgotten that the original conclusion was only a hypothesis.
The mature genealogist learns to say:
That's good genealogy.
And perhaps the hardest part: an honest unknown is more valuable than a beautiful, completely documented-looking fiction.
B) Prompt: What is a second truth?
OpenAI ChatGPT5 response:
Here's the second truth, and this one may sting a little:
C) Prompt: How can I achieve certainty for a specific event for a person? How do I determine when they were born, died, buried, married, etc.? What metric should I use?
OpenAI ChatGPT5 response:
I would make one important distinction:
But for your day-to-day work in a large family tree, I think you need something more practical than simply "GPS: yes/no."
I'd use a 5-level "Genealogical Confidence" scale:
And I would add a sixth status:
Apply the scale to each fact separately
This is the part I think could really improve your RootsMagic research.
Don't ask:
Now you've got something much more meaningful than a tree full of green checkmarks.
How do you decide the confidence number? I'd use four questions.
1. How close is the evidence to the event?
A record created at or near the event by someone with firsthand knowledge generally deserves more weight than a later recollection or an unsourced compiled tree. FamilySearch's guidance similarly emphasizes considering who created the record, when it was created, who supplied the information, and whether it is original, derivative, or authored. FamilySearch
For example:
2. How many independent pieces of evidence agree?
One excellent record can be powerful. But independent agreement is gold.
Suppose you have:
The key word is independent. Ten online trees repeating the same incorrect date aren't ten pieces of evidence. They're one piece of evidence wearing ten different hats.
3. Is there contradictory evidence?
This is where serious genealogy separates itself from "I found a record, so I added it."
Suppose:
I'd ask:
And here's my favorite metric:
If you want one number for your own research management, I'd use:
Not as literal mathematics, but as a disciplined way of thinking.
Imagine you are evaluating:
Evidence supporting:
You might conclude:
Birth year 1839 — Confidence 3 (Probable).
Evidence supports 1839, but conflicting later evidence for 1842 prevents treating the date as established.
And:
Biological parents — Confidence 0 (Unknown).
The DNA and circumstantial evidence make the proposed Lanfear connection plausible, but the available evidence does not yet establish the parentage.
That is a much more honest tree.
One other thing I'd change
I wouldn't give an ancestor an overall "confidence score" such as 87%. That creates a false precision.
Instead, I'd use words + evidence:
The GPS itself is deliberately a process for reaching credible conclusions rather than a numerical scoring system. National Genealogical Society
So, for your genealogy, I'd adopt this rule:
A birth date copied from a contemporary civil registration may need little additional work.
The supposed parents of a colonial-era brick-wall ancestor may require a full proof argument.
And sometimes the correct conclusion really is:
That's a finished research conclusion.
Links to my blog posts about using Artificial Intelligence are on my Randy's AI and Genealogy page. Links to AI information and articles about Artificial Intelligence in Genealogy by other genealogists are on my AI and Genealogy Compendium page.
A) Prompt: If you could tell us ONE truth we’re not ready to hear about genealogy research, what would it be?
OpenAI ChatGPT5 response:
If I could tell genealogists one truth they may not be ready to hear, it would be this:
"You will probably never know the whole truth about your ancestors—and your job is not to fill the gaps with the most plausible story. Your job is to become comfortable with the gaps."Genealogy trains us to expect an answer. A birth date should lead to parents; parents should lead to grandparents; a census should tell us where someone lived; DNA should eventually break the brick wall.
But sometimes the records simply do not exist. Sometimes they existed but were destroyed. Sometimes an ancestor deliberately concealed something. Sometimes several records contradict one another. And sometimes the evidence will never allow us to distinguish between two reasonable possibilities.
The dangerous moment comes when we say:
"This is probably what happened."—and gradually begin treating probably as proven.
That's especially tempting with brick-wall ancestors. We find a family that fits the age, location, naming pattern, DNA evidence and historical circumstances. Everything seems to line up. We build the connection into the tree, write the biography, and five years later we've forgotten that the original conclusion was only a hypothesis.
The mature genealogist learns to say:
"I don't know."And even better:
"Here is what I know, here is what I suspect, here is the evidence for it, here is the evidence against it, and here is what would have to be discovered to change my mind."That isn't failure.
That's good genealogy.
And perhaps the hardest part: an honest unknown is more valuable than a beautiful, completely documented-looking fiction.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
B) Prompt: What is a second truth?
OpenAI ChatGPT5 response:
Here's the second truth, and this one may sting a little:
"Most of the genealogy research you do will never actually change your family tree."
We spend enormous amounts of time searching for the elusive record—the marriage record, the missing census, the maiden name, the parents of the brick-wall ancestor. And sometimes we find it. But much of genealogy is actually confirming what we already know, eliminating possibilities, documenting existing facts, and discovering that the record we desperately want simply isn't there.
That doesn't mean the research was wasted.
In fact, some of the most valuable genealogy work produces negative results:
And here's the uncomfortable part:
I think a better measure is certainty gained.
You might spend three hours researching one person and add nobody to your tree—but eliminate four incorrect possibilities and establish exactly what records have been searched. That's progress.
And for someone with a huge tree, there's an even bigger lesson:
That doesn't mean the research was wasted.
In fact, some of the most valuable genealogy work produces negative results:
- The record isn't in the county where you expected it.
- The supposed parents aren't supported by the evidence.
- The family wasn't living where the online tree says they were.
- The DNA match doesn't actually fit the proposed relationship.
- The "obvious" person turns out to be the wrong person.
And here's the uncomfortable part:
A carefully documented conclusion that you cannot prove something may be a greater accomplishment than finding a new ancestor.Genealogists sometimes measure progress by names added.
I think a better measure is certainty gained.
You might spend three hours researching one person and add nobody to your tree—but eliminate four incorrect possibilities and establish exactly what records have been searched. That's progress.
And for someone with a huge tree, there's an even bigger lesson:
The goal isn't to make your family tree bigger. It's to make the tree you already have more trustworthy.That's a truth worth putting on a genealogy conference T-shirt.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
C) Prompt: How can I achieve certainty for a specific event for a person? How do I determine when they were born, died, buried, married, etc.? What metric should I use?
OpenAI ChatGPT5 response:
I would make one important distinction:
"You cannot achieve mathematical certainty in genealogy. You can achieve a defensible level of confidence."The traditional framework for that is the Genealogical Proof Standard (GPS). It calls for reasonably exhaustive research, complete citations, analysis and correlation of evidence, resolution of conflicts, and a written conclusion. National Genealogical Society
But for your day-to-day work in a large family tree, I think you need something more practical than simply "GPS: yes/no."
I'd use a 5-level "Genealogical Confidence" scale:
And I would add a sixth status:
0 — Unknown.That's important. Unknown is not the same as probably X.
Apply the scale to each fact separately
This is the part I think could really improve your RootsMagic research.
Don't ask:
"How certain am I about Devier James Lamphier Smith?"Ask:
"How certain am I about each fact concerning Devier James Lamphier Smith?"For example:
How do you decide the confidence number? I'd use four questions.
1. How close is the evidence to the event?
A record created at or near the event by someone with firsthand knowledge generally deserves more weight than a later recollection or an unsourced compiled tree. FamilySearch's guidance similarly emphasizes considering who created the record, when it was created, who supplied the information, and whether it is original, derivative, or authored. FamilySearch
For example:
- Birth
- Contemporary birth record
- Baptismal record close to birth
- Family Bible created contemporaneously
- Marriage record giving age/birthplace
- Census
- Death certificate
- Obituary
- Later family history
- Online family tree
2. How many independent pieces of evidence agree?
One excellent record can be powerful. But independent agreement is gold.
Suppose you have:
- 1839 birth record
- 1850 census giving age 11
- 1860 census giving age 21
- 1861 marriage record giving age 22
- 1870 census giving age 30
- 1875 census giving age 35
- 1880 census giving age 41
- 1885 census giving age 46
- 1889 newspaper article stating 1839 birth
- obituary giving the same birthplace
The key word is independent. Ten online trees repeating the same incorrect date aren't ten pieces of evidence. They're one piece of evidence wearing ten different hats.
3. Is there contradictory evidence?
This is where serious genealogy separates itself from "I found a record, so I added it."
Suppose:
- 1850 census: age 11 → about 1839
- 1860 census: age 21 → about 1839
- 1870 census: age 30 → about 1840
- 1880 census: age 41 → about 1839
- 1889 article: says 1839
- 1889 Bible entry: says 1842
- 1894 Obituary: says 1842
- Death record: says 1842
I'd ask:
- Why do the Bible, obituary and death record disagree with the census data?
- Who provided the information? When? Was that person actually present at the birth? Could the age have been calculated incorrectly?
- The GPS specifically requires contradictory evidence to be addressed rather than ignored. NGS Monthly
4. Could another person fit the evidence?
This is the question I would add to your normal research routine:
If you have Devier Smith born about 1839 in Henderson, New York, ask:
This is the question I would add to your normal research routine:
"What evidence would convince me that I have the wrong person?"That's enormously useful.
If you have Devier Smith born about 1839 in Henderson, New York, ask:
- Are there other Devier Smiths of approximately the same age in Henderson or neighboring communities?
- If there are none, your identification gets stronger.
- If there are three, your burden of proof gets much higher.
And here's my favorite metric:
If you want one number for your own research management, I'd use:
Confidence = Evidence − Contradiction − Alternatives
Not as literal mathematics, but as a disciplined way of thinking.
Imagine you are evaluating:
"Devier James Lamphier Smith was born 7 May 1839 in Henderson, Jefferson County, New York."You might write:
Evidence supporting:
- Later records give 1839.
- The reported birthplace is consistent.
- His adoption circumstances fit the known timeline.
- DNA evidence potentially connects him to the Lanfear family.
- Some later records suggest 1842.
- No discovered birth record provides a birth year or establishes his parents.
- He could have been born in 1839.
- He could have been born in 1842.
- He could belong to a different Lanfear family.
- His biological parentage could remain unidentified.
You might conclude:
Birth year 1839 — Confidence 3 (Probable).
Evidence supports 1839, but conflicting later evidence for 1842 prevents treating the date as established.
And:
Biological parents — Confidence 0 (Unknown).
The DNA and circumstantial evidence make the proposed Lanfear connection plausible, but the available evidence does not yet establish the parentage.
That is a much more honest tree.
One other thing I'd change
I wouldn't give an ancestor an overall "confidence score" such as 87%. That creates a false precision.
Instead, I'd use words + evidence:
- Proven
- Highly probable
- Probable
- Possible
- Speculative
- Unknown
The GPS itself is deliberately a process for reaching credible conclusions rather than a numerical scoring system. National Genealogical Society
So, for your genealogy, I'd adopt this rule:
Every important fact gets its own confidence assessment, and every confidence assessment is supported by evidence.That gives you something extremely useful when you're dealing with 75,000+ people: you don't have to prove every fact to the same standard.
A birth date copied from a contemporary civil registration may need little additional work.
The supposed parents of a colonial-era brick-wall ancestor may require a full proof argument.
And sometimes the correct conclusion really is:
"Parents unknown after reasonably exhaustive research."That's not an unfinished genealogy.
That's a finished research conclusion.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The Google Gemini Notebook Video Overview of this story is on my YouTube channel, and below:
These concepts are not original or unknown. The Genealogical Proof Standard helps us judge certainty. In addition, Yvette Hoitink created a "Level Up" metric system to judge certainty (see Six Levels of Ancestral Profiles – Level-up Challenge!).
Steve Little has embodied many of these ideas in his Genealogy Research Assistant AI prompt -- see https://chatgpt.com/g/g-69701d25d61c819192c2db4589b366d9-genealogical-research-assistant.
==============================================
The URL for this post is: https://www.geneamusings.com/2026/04/using-steve-littles-ai-genealogy.html
Copyright (c) 2026, Randall J. Seaver
Please comment on this post on the website by clicking the URL above and then the "Comments" link at the bottom of each post. Share it on X, Facebook, or Pinterest using the icons below. Or contact me by email at randy.seaver@gmail.com. Note that all comments are moderated, and may not appear immediately.
Subscribe to receive a free daily email from Genea-Musings using www.Blogtrottr.com.
No comments:
Post a Comment