Episodes

  • Ep. 64: The Memory Audit | What Your AI Thinks You Proved
    17 Nov 2026

    Your AI chatbot keeps notes on your family research and drops the question marks. In one audit, 4 remembered lines didn't hold up.

    As of November 2026, memory is on by default in Claude, including the free plan, and ChatGPT and Gemini keep their own versions. This episode shows where those notes live and how to grade every line the way a genealogist grades evidence: with a confidence word and a reason.

    The case: a composite family built from real Maine record types. An AI's memory said Joseph Thurlow of St. George, Maine (died 1911) was the son of Elias Thurlow and Lydia Pease, and that he served in the Civil War. Lydia came from an online tree with no sources. The 1880 census and his 1911 death record both name his mother as Mercy. The Civil War service started as a question someone asked once.

    What you'll learn:
    - Where to find and edit AI memory in Claude, ChatGPT and Gemini (menus as of November 2026)
    - How to audit AI memory against your genealogy records, line by line
    - How to grade each claim Certainly, Probably, Possibly, Not supported or Contradicted
    - How to write "proof lines" and store them in Claude Projects and memory
    - A trap test: one question in a new chat that shows whether your fix worked
    - Three copy-paste prompts for genealogists

    FAQ

    Does my AI chatbot remember my genealogy research?
    Yes, if memory is on. As of November 2026, Claude's memory is on by default, including the free plan. ChatGPT and Gemini have their own memory features on many accounts.

    Where do I see what Claude remembers about me?
    In Claude, open Settings, then Your Claude, then Memory, and click Manage. Since August 25, 2026, you can edit or delete each memory topic one at a time.

    Where do I see what ChatGPT remembers?
    Open Settings, then Personalization, and look for the Memory summary with a Manage button. Some accounts show an older list called saved memories.

    Can I trust an AI's memory of my family tree?
    Not without checking it. In this episode's audit, nothing in the AI's memory earned Certainly, and an unsourced tree's guess had been stored as fact.

    What is a proof line in genealogy?
    A proof line is one claim, one confidence word and the records behind it, for example: Mother: Mercy Thurlow, maiden name unknown. PROBABLY. Basis: the 1880 census and the 1911 death record.

    How do I test whether my AI's memory is accurate?
    Open a new chat and ask a question that assumes the old, wrong claim is true. In our Claude test, the original notes produced research plans for the wrong mother every time. With proof lines in place, every run corrected the question first.

    The honest outcome: we ended with fewer facts than we started. Elias is probably Joseph's father and Mercy probably his mother, but Mercy's maiden name is still unknown. The next record to find is a marriage for Elias and Mercy in the St. George town records, dated before 1846.

    For Australian and UK researchers: grade what you know the same way, then check the NSW Registry of Births, Deaths and Marriages free historical index, family notices in Trove, the Society of Genealogists' Standards and Good Practice page, and the GRO's free online birth indexes for England and Wales.

    Patreon members get the Companion Guide for this episode. It walks through the full memory audit in Claude, ChatGPT and Gemini, with advanced prompts for turning a whole Project's notes into proof lines. Learn more at ancestorsandai.com.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    35 mins
  • Ep. 63: Burned County Research - Find Surviving Records with AI
    10 Nov 2026

    A courthouse fire destroys the county's copy, not every copy. From 1853, Virginia counties sent the state a yearly copy of their vital registers.

    In this composite case from Buckingham County, Virginia, where the courthouse burned in 1869, every online tree named Nathan Gilliam as the father of Joel Gilliam, born about 1833. None of them cited a record. The state's surviving copy of the December 1856 marriage register named Joel's real parents: Archer and Frances Gilliam.

    What you'll learn:
    - How to research a burned county by asking "where did the second copy go?" before you search for a name
    - A copy-paste Perplexity prompt for a two-copy source survey, with every citation checked
    - How to build a Record Survival Map in Claude, as an artifact, from checked facts only
    - Why a FamilySearch Research Wiki start date can describe the county's copy only
    - How to test one record against the 1850 census and tax lists with the Genealogical Proof Standard
    - Where to find the state copies that survive for Virginia's lost records localities

    FAQ

    Are all records lost when a courthouse burns?
    No. A fire destroys the copy kept in the courthouse. Copies sent elsewhere, such as state registers, federal census copies, tax lists, and exhibits in another county's lawsuits, can survive.

    When did the Buckingham County, Virginia courthouse burn?
    In 1869. The Library of Virginia lists Buckingham among its lost records localities.

    Does Virginia have state copies of birth, marriage, and death registers?
    Yes. From 1853 to 1896, Virginia counties sent a yearly copy of their registers to the Auditor of Public Accounts. For Buckingham, the state's marriage registers survive from 1854, and its birth and death registers from 1853, with 1863, 1869, and 1870 missing.

    Do Virginia marriage registers from the 1850s name the parents?
    Yes. Virginia marriage registers from 1853 on record the names of the parents. That column named Joel Gilliam's parents in this case.

    Where can I search Virginia's state copies of marriage registers?
    On FamilySearch, in the free collection "Virginia, Bureau of Vital Statistics, County Marriage Registers, 1853-1935." Some images can be viewed only at a FamilySearch Center or affiliate library.

    Can AI help with burned county genealogy?
    Yes, as a research assistant. Perplexity can survey which copies of each record survive, with a source beside each claim. Claude can turn checked facts into a Record Survival Map. Check every citation yourself.

    Can Claude's free plan make artifacts?
    As of November 2026, yes. Claude's free plan can build and save artifacts with the Create and edit files setting turned on. Sharing one by link takes a Pro plan or higher.

    For Australian and UK researchers: the NSW Registry of Births, Deaths and Marriages holds copies of church registers from 1788 to 1856, and bishops' transcripts are yearly copies of parish registers in England and Wales, many at County Record Offices.

    The outcome: a full breakthrough. The marriage register, the 1850 census, and the tax lists point to Archer and Frances as Joel's parents. Still open: whether Nathan was Archer's brother, and Frances's maiden name.

    The Companion Guide adds the full Record Survival Map workflow, advanced prompts, and a GPS Research Checklist for burned counties. Find it at ancestorsandai.com.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    34 mins
  • Ep. 62: The FAN Club Hiding in Your Old AI Chats
    3 Nov 2026

    When an ancestor vanishes after the 1860 census, his neighbors can show you where the family went. In this episode, 3 old AI chats, 1 FAN club and a probate file solve a "disappearance" that never happened.

    Charles Spafford farmed in Columbia County, Wisconsin in 1860, and then seemed to disappear. My old chats in Claude and ChatGPT held competing theories, from the California gold fields to a Kansas man with a new wife. This composite teaching case shows how to pull those scattered chats into one AI Project, harvest them into a single research log, and then trace the family through its friends, associates and neighbors (the FAN club method).

    This episode is for you if:
    - Your genealogy research is scattered across dozens of old AI chats
    - You want to learn Claude Projects, ChatGPT Projects or Gemini notebooks for family history
    - You have an ancestor who disappears between the 1860 and 1870 census
    - You want to use FAN club research, deed indexes and probate records to break a brick wall
    - Your ancestor's surname may have been misread by an indexer

    FAQ

    What is the FAN club method in genealogy?
    FAN stands for friends, associates and neighbors. When records about your ancestor run out, you research the people around them, such as deed witnesses and census neighbors, because families often moved with the same group.

    Can AI remember my genealogy research from one chat to the next?
    Not by default. In Claude, chats inside the same Project do not share context unless the information is added to the Project's knowledge files, so the episode shows how to harvest old chats into one research log file.

    Do Claude, ChatGPT and Gemini offer Projects on the free plan?
    As of October 2026, all three offer a version on the free plan. Claude and ChatGPT call it a Project, Gemini calls it a notebook, and Claude's free plan allows up to five Projects.

    Why can't I find my ancestor in the next census?
    The family may have moved, or an indexer may have misread the name. In this case the widow was indexed as "Stafford" instead of "Spafford" in the 1870 census, which hid her from every Spafford search.

    How do you find out if an ancestor died before death certificates existed?
    Check land and probate records. Wisconsin did not begin statewide death registration until 1907, and Columbia County's death records generally start in 1877, so a deed and a probate file proved this 1865 death.

    Which ranks higher in FAN research, a census neighbor or a deed witness?
    A person who appears in two records with your ancestor outranks a person in one. A deed witness who also lived next door is stronger evidence of a real connection than a name on the same census page.

    The honest outcome: AI found no new record. It organized what I already had, and the answer was a lead I had ignored in three separate chats. The new Gemini Notebook shortcut hung and delivered my log two days late, the Kansas Charles was the wrong man, and Harriet's parents and her daughter Emma after 1870 are still open questions.

    Records and places covered: Columbia County WI, Freeborn County MN, 1860 and 1870 US census, grantor and grantee deed indexes, county probate files, administratrix deeds.

    For Australian and UK listeners: the same neighbor method works with Trove, Public Record Office Victoria, County Record Offices and FreeCEN.

    Want the deeper workflow? The Companion Guide has the full case-file setup for Claude, ChatGPT and Gemini, an advanced harvest prompt for long chats, a deed and probate checklist and a GPS Research Checklist. Find it at ancestorsandai.com.

    AI is your research assistant, not your researcher.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    34 mins
  • Ep. 61: The Informant Test - Death Certificates and AI
    27 Oct 2026

    Who told the clerk your ancestor's parents? On this 1938 death certificate, the informant was born 24 years after the birth she reported.

    The parents' names on a death certificate come from an informant, often a child or grandchild who wasn't there. In Episode 61, Brian works a composite Missouri case built from real record types. Ida Belle Corwin died in Holden, Johnson County, in 1938. Her certificate names her mother as Martha Dowell. The informant was Ida's daughter, born in 1898. With Claude, ChatGPT and tested prompts, Brian profiles the informant, builds a test plan, counts witnesses instead of documents, and follows a Civil War pension file to the woman the family forgot.

    This episode is for you if you're searching for: How to check the parents named on a death certificate Who the informant is on a death certificate Stepmother listed as mother on a death certificate Civil War pension file genealogy and the 1898 pension questionnaire Using Claude and ChatGPT for genealogy and the Genealogical Proof Standard

    FREQUENTLY ASKED QUESTIONS

    Can a death certificate name the wrong mother? Yes. The parents' names usually came from an informant who wasn't present at the birth. A stepmother who raised a child can be reported as the child's mother.

    Who is the informant on a death certificate? The informant is the person who answered the questions about the person who died. The informant usually supplied the birth date, birthplace and parents' names. The physician supplied the cause of death.

    Is a death certificate direct evidence of a parent's name? Yes, because it answers the question outright. But "direct" describes how a record answers, not whether the answer is true. The parents' names are usually secondary information.

    How can AI help test a death certificate? Claude dates every fact on the certificate against the informant's own life and predicts the informant's most likely error. ChatGPT builds a test plan of earlier records made by other people.

    What is the 1898 Civil War pension questionnaire? In January 1898 the Bureau of Pensions sent Union pensioners a questionnaire asking for their wife's maiden name, any former wife and the date and place of her death or divorce, and the names and birth dates of their living children. It is found in the pension file, not in the index.

    Where can I find Missouri death certificates for free? The Missouri State Archives offers free images of 1910 to 1975 death certificates on Missouri Digital Heritage, as of October 2026.

    Do two records that agree prove a fact? Not always. If the same informant supplied both, they count as one witness. In this case the death certificate and the obituary came from the same family memory.

    THE HONEST OUTCOME This is an unexpected discovery, but not a clean one. The first two versions of Brian's Claude prompt missed the stepmother, and the third worked once, then failed on a rerun. ChatGPT's test plan never suggested the pension file that finally answered the question. And the birth mother's own parents, birthplace and grave are still unknown.

    Australian and UK listeners: the same test works on ScotlandsPeople death registers from 1855, which usually name both parents and the informant, and on NSW and Queensland BDM death certificates.

    The Episode 61 Companion Guide for Patreon members includes an informant profile for marriage and census records, a test plan template, a witness count worksheet and a GPS Research Checklist for parents named on death certificates. Learn more at ancestorsandai.com. The free episode stands on its own.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    33 mins
  • Ep. 60: The Leeds Method and AI - Finding an Unknown Father
    20 Oct 2026

    How do you find an unknown father among more than 2,000 DNA matches? Sort 20 of them into four grandparent groups with the Leeds Method and AI.

    In this composite case, Lorraine was born in Brisbane, Queensland, in 1944. Her birth registration left her father's name blank, and her mother told her only one thing: he was a Yank. This episode walks through every step of her search, from match list to a short list of candidates.

    In this episode you'll learn:
    - How the Leeds Method sorts AncestryDNA matches between 90 and 400 cM by shared matches
    - Why to replace every match's name with a code before using AI
    - A five-rule Claude prompt that sorts matches into clusters and flags bridge matches
    - The shuffle test: rerun the AI sort with the rows reversed and keep only what holds
    - How Gemini Notebook names each cluster from your own tree notes, with citations
    - How Perplexity and DNA Painter's free Shared cM Project tool test a relationship
    - What Ancestry's Regions by Grandparent (announced October 9, 2026) can and can't tell you
    - What target testing is, and how to approach possible relatives gently

    The honest outcome: a partial answer. The DNA made one candidate very unlikely, and two brothers remain. Paper records may not settle it, because the 1973 records fire destroyed an estimated 80 percent of Army personnel files for soldiers discharged between 1912 and 1960. Target testing is the next step.

    FREQUENTLY ASKED QUESTIONS

    What is the Leeds Method?
    The Leeds Method, created by Dana Leeds in 2018, sorts DNA matches sharing 90 to 400 cM into groups using shared matches, so the groups can point to your four grandparent lines.

    Can AI sort DNA matches with the Leeds Method?
    Yes, if you give it a coded table of shared matches and written rules. In this episode, Claude sorted 20 coded matches into four groups, set aside one bridge match, and refused to place two matches it had no evidence for.

    What is a bridge match?
    A bridge match links into two of your groups at once. It can descend from the couple whose marriage joined those two families, so study it instead of merging the groups.

    What is the shuffle test for AI DNA sorting?
    Run the same sorting prompt again in a fresh chat with the rows in reverse order. Keep only the placements that land in the same group both times.

    Should I put my DNA matches' names into ChatGPT or Claude?
    This show's rule is no. Give each match a code like M01 and keep the key on paper. The AI only needs the links between matches, not the names of living people.

    Does 297 cM fit a half niece or a first cousin once removed?
    In the Shared cM Project 4.0, a first cousin once removed ranges from 102 to 980 cM, so 297 cM fits. A half niece ranges from 492 to 1,315 cM, so 297 cM falls below that range.

    For Australian and UK researchers: the National Archives of Australia's fact sheet on United States forces in Queensland, Trove for wartime newspapers, and the GRO birth indexes and FreeBMD for England and Wales.

    The Companion Guide includes the full shuffle-test workflow, an advanced prompt for working a bridge match, a gentle first-message template for target testing, and a GPS Research Checklist for unknown parentage. Find it at ancestorsandai.com.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    36 mins
  • Ep. 59: Find Your German Ancestor's Village - Research Questions & AI
    13 Oct 2026

    Four records said Hanover, Prussia or Germany, and not one named his village. Here's how to find a German immigrant's village of origin with AI, step by step.

    When every record says "Germany," the problem is usually the research question, not the records. In this episode, we follow Friedrich Eckhoff, a Lutheran cooper who came from Bremen to Baltimore in 1869 (a composite case built on real record types). You'll learn the method that finally led to his village, one click at a time, with three free AI tools.

    This episode is for you if:
    • You're trying to find your German ancestor's village or hometown
    • Your immigrant's records say only Germany, Prussia or Hanover
    • You found several towns with the same name and don't know which is yours
    • You want to use ChatGPT, Perplexity and Gemini in Google AI Studio for genealogy without trusting them blindly

    FAQ

    What makes a good genealogy research question?
    A good research question names one person with identifiers such as age, place and occupation, asks for one fact, and points to a record that could contain that fact.

    Why does a census say Prussia instead of my ancestor's village?
    The 1870 U.S. census told enumerators to record the German state, such as Prussia or Bavaria, not a town. Prussia annexed the Kingdom of Hanover in 1866, so one man can appear as "Hanover" in 1870 and "Prussia" in 1880.

    Which records are most likely to name a German immigrant's village?
    The church registers of the immigrant's own congregation in America, especially marriage entries, are often the best chance. U.S. censuses, naturalizations before 1906 and 1860s arrival lists usually give only a state or country.

    Do the Bremen passenger lists survive?
    Mostly not. From 1875, Bremen kept only the three most recent years of lists, and it reversed that policy in 1907. Survivors are mainly from 1907 and 1908, 1913 and 1914, and 1920 through 1939.

    How do I tell which same-named German village is the right one?
    Search Meyers Gazetteer (meyersgaz.org), filter by region, check each place's parish church, then use witnesses and sponsors from family records. In this episode, a wedding witness's village narrowed eight Neuenkirchens in the province of Hanover to one.

    Can AI find my ancestor's German village for me?
    Not on its own. When asked directly, ChatGPT listed two to four Neuenkirchens and missed the right one both times. AI helped most by interviewing for a better question, sorting which records could answer it, and transcribing old German handwriting, with every claim checked against its source.

    The honest outcome: an 1846 baptism names Friedrich, son of Hinrich Eckhoff, in the one Neuenkirchen in his wedding witness's district. It's strong evidence but not airtight, because at least one other Neuenkirchen has no church book images online.

    Australian and UK listeners: the same approach works with the NSW assisted immigrant lists, which often record a native place, and the 1851 census, the first to record parish of birth, with free transcriptions on FreeCEN.

    AI is your research assistant, not your researcher.

    Want to go deeper? The Companion Guide at ancestorsandai.com includes the full Question Card workflow, a same-named village sorting prompt and a GPS Research Checklist for finding a place of origin.

    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    29 mins
  • Ep. 58: Same Name, Two Men - Untangle a Merged Family Tree With AI
    6 Oct 2026

    Two men named Giovanni Farruggia registered for the World War I draft in Tampa on the same day in 1917. An online family tree fused them into one ancestor.

    Identity separation for genealogists: how to tell two same-name ancestors apart with ChatGPT, Claude and Gemini Notebook, using a Sicilian case from Ybor City, Tampa.

    This episode is for you if:
    You attached a record because the name matched and something feels off.
    Your ancestor seems to live in two houses in the same year.
    You suspect a FamilySearch or Ancestry tree merged two people.
    Your Italian, Sicilian or immigrant family reused the same names every generation.

    What you'll learn:
    The four-step identity separation method: plan the identifiers, gather every record of the name, count the people, then trace every tree fact to its owner.
    Why birthplace, occupation and arrival year are weak identifiers for cousins from one village.
    The collision test: two records that can't describe one man on one day prove two men.
    How a census citizenship column led to a naturalization petition that settled the case.
    Click-by-click setup: a ChatGPT temporary chat set to Unpersonalized, a Claude incognito chat, and Gemini Notebook (formerly NotebookLM) with one record per source.

    Questions this episode answers:

    How do I tell if two ancestors with the same name are the same person?
    Count before you claim. Look for collisions: two records made on the same day or in the same census that list different birth dates, addresses, wives or households. One collision proves two people. No collision proves nothing.

    Can AI help separate two people with the same name in a family tree?
    Yes, if you make it show its work. A blind Claude chat sorted ten records into two men and named its own weakest link. Gemini Notebook traced each tree fact to the man it came from, with clickable citations.

    Is a birthplace enough to identify an ancestor?
    No. Historians say Santo Stefano Quisquina supplied more than sixty percent of Tampa's Italians, so two cousins could share a village, a trade and an arrival year.

    How do I know if an online family tree merged two people?
    Trace every fact back to its source. Here the birth date and parents came from a grocer born in 1893. The wife and children came from a cigar maker born in 1895.

    What census column can point to naturalization records?
    Citizenship. The cigar maker was an alien in 1920 and naturalized by 1930, which led to his 1928 petition with his exact birth date, address, wife and children.

    The honest outcome: the tree's one Giovanni was two men. Rosalia's husband was identified. His father is probable and his mother is still unproved. Both AI tools made the same slip, reading "son of the head" as naming both parents. The family is a composite built from real Tampa record types.

    For Australian and UK listeners: the same method works with First World War service records at the National Archives of Australia, Trove, the 1939 Register for England and Wales, and FreeBMD.

    The Companion Guide adds advanced prompts for a side-by-side identity table and for untangling a tree merged more than once. Find it at ancestorsandai.com.


    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    34 mins
  • Ep. 57: Find Her Maiden Name: Proving Parents with Indirect Evidence and AI
    29 Sep 2026

    No record names her parents, and you can still prove them. This episode shows how, step by step, with 5 ordinary records and 3 free AI tools.

    Every family tree has a woman like Nancy Brockway: a first name, a married surname, and nothing before that. Our case is a composite family built from real 1850s New York record types. Nancy lived in New Berlin, Chenango County, and was born about 1816 in Otsego County. There's no marriage record and no will. There is a Hannah Pettibone, 67, in her 1850 household, and an 1847 deed that puts Nancy's name on a dead man's farm.

    WHAT YOU'LL LEARN
    - How to find a woman's maiden name with no marriage record
    - Why most New York marriages before 1847 left no government record
    - What the 1855 New York State census tells you that the 1850 census doesn't
    - How to search old deeds with FamilySearch Full-Text Search
    - How to transcribe a handwritten deed with Gemini in Google AI Studio
    - How to run a blind read in a Claude incognito chat, so AI can't confirm your bias
    - The knockout test: remove one record at a time to see what your case rests on

    FREQUENTLY ASKED QUESTIONS

    Can you prove a parent-child relationship when no record states it?
    Yes. Indirect evidence, several records that together leave one reasonable explanation, is fully acceptable under the Genealogical Proof Standard.

    Why can't I find a marriage record for my New York ancestor before 1847?
    According to the FamilySearch Research Wiki, no county, city, town or village government in New York recorded marriages before 1847. Statewide registration began in 1880 and 1881.

    How can a deed reveal a woman's maiden name?
    When a man died owning land and no one went to court, his heirs often settled it by deed. A married daughter may appear beside her husband with words like "in right of the said Nancy," meaning the share was hers, not his.

    What is a blind read with AI?
    You give a chatbot your records but not your theory, in a fresh incognito or temporary chat, and ask for every explanation the records allow. In this episode, Claude raised one I hadn't considered: stepmother.

    What is the knockout test in genealogy?
    Remove one record at a time and ask whether your conclusion still stands. Split your conclusion into single claims first. Claude overreached when it tested a two-part claim.

    Can I trust ChatGPT to tell me which genealogy records exist?
    Use it to brainstorm, then check every record type in the FamilySearch Research Wiki. In this episode, ChatGPT pointed to early county marriage records that don't exist for 1830s New York.

    THE HONEST RESULT
    A partial answer. The evidence best supports Nancy as a daughter of Stephen Pettibone. Whether Hannah was her mother or her stepmother is still open, and I name the record that could settle it.

    Australian and UK listeners: the same approach works with the NSW Historical Land Records Viewer and the National Library of Wales's wills proved before 1858.

    Patreon members: this episode's Companion Guide includes the full blind read and knockout workflow, plus advanced prompts for turning your results into a written proof summary. Find it, and every prompt from this episode, at ancestorsandai.com.


    Connect with Ancestors and Algorithms:

    📧 Email: ancestorsandai@gmail.com
    🌐 Website: https://ancestorsandai.com/
    📘 Facebook Group: Ancestors and Algorithms: AI for Genealogy - www.facebook.com/groups/ancestorsandalgorithms/

    Golden Rule Reminder: AI is your research assistant, not your researcher.

    Join our Facebook group to share your AI genealogy breakthroughs, ask questions, and connect with fellow family historians who are embracing the future of genealogy research!

    New episodes every Tuesday. Subscribe so you never miss the latest AI tools and techniques for family history research.




    Show more Show less
    32 mins