Ep. 30: ChatGPT, Perplexity, Claude & NotebookLM Meet the Genealogical Proof Standard

[Music] What if I told you that the most rigorous research standard in all of professional genealogy, the one certified genealogists stake their reputation on, actually has a natural specific role for each of the major AI tools you probably already have open on your computer? Not a vague AI can help a genealogy kind of role, a specific one. Chatupetit does this, perplexity does that, Claude does this other thing, and notebook LM pulls it all together at the end. That's what I figured out, sitting in my office after coming home from RootsTech 2026 with multiple pages of notes, and one question nobody had answered for me yet. How do these four AI tools work together, step by step, inside of the genealogical proof standard? Today is the episode where we answer that. We're going to go through all five GPS elements one at a time with a real ancestor story running through every step. Copy paste prompts you can use today. All four tools on their free tiers. This one might just change how you research. Let's go. Welcome to Ancestors and Algorithms, where family history meets artificial intelligence. I'm your host Brian, and today we're doing something I've been working toward for 30 episodes. Today is the GPS episode. I know that phrase, the genealogical proof standard, might be making some of you a little nervous. You might be thinking this is about to get very academic, that it's only relevant to people chasing a professional certification, that it's not for someone who just downloaded Chatubit last Tuesday. I want you to put all of that aside before we even get started. The GPS is not a gatekeeping mechanism. It is a framework for doing research that actually means something. Research where your conclusions hold up. Research where your sources are documented. Research that your grandchildren can actually use someday. And here's the thing I want you to hear. For AI tools, all free, make every single step of the GPS more manageable for everyone from the professional genealogist to the beginner who just got their first DNA kit. Let's dig in. Let me tell you what happened at RootsTech 2026. Early this month in March, I made the trip to Salt Lake City for my second in-person RootsTech. And if you've never been, imagine 30,000 genealogists from all over the world filling up a convention center. Australia, the UK, Canada, Japan. People who have dedicated decades to family history research, to methodology, to preserving stories that would otherwise disappear. On the first day, I attended a session from the Coalition for Responsible AI and Genealogy. If you haven't heard of this group, they're worth knowing about. They include some of the most respected names in our field and they've developed five guiding principles for responsible AI use in genealogy, accuracy, disclosure, privacy, education, and compliance. Those principles were everywhere at RootsTech this year. They've been adopted by the National Geneological Society and endorsed by organizations across the community. But the conversation that wouldn't leave my head the one I kept coming back to in my drive home was around a question I kept hearing as sessions and between sessions and at lunch. How do AI tools and the genealogical proof standard actually work together? I heard broad strokes. I heard philosophical discussion. I didn't hear anyone sit down and map it out. Element by element. Tool by tool. With the prompts written now. So that is what today is. Let me introduce the GPS for anyone who isn't familiar with it. The genealogical proof standard is a five element framework developed by the board for certification of genealogists, which you can find at bcgcertification.org. Here are the five elements. Element one reasonably exhaustive research. Element two complete and accurate source citations. Element three thorough analysis and correlation of evidence. Element four resolution of conflicting evidence. And element five a soundly reason coherently written conclusion. Five elements. And as I set with these after RootsTech, something became very clear to me. Each element has a different AI tool that fits it best. Not because any one tool is always superior. But because each tool has genuine strengths and matching those strengths to the GPS is how you get results that actually hold up. Before we go any further, I want to address something directly because this podcast values accuracy over everything else. If you are pursuing certification through the board for certification of genealogists, you need to know that bcg has specific guidance about AI and certification portfolios. Their position issued in 2024 and reaffirmed in the 2025 application guide is clear. Work submitted in a CG portfolio must be the applicant's work. AI generated analysis and AI generated writing are not acceptable for portfolio submission. Search tools and grammar checkers are genuinely permitted. But the analysis and the writing must be yours. This is not arbitrary. Certifications exist to demonstrate your personal research competence and an AI cannot do that on your behalf. I mention this not to scare anyone away from AI tools, but because accurate information matters. If you're on a certification path, you deserve to know exactly where the lines are. For everyone else, which is the vast majority of us doing family history research outside of a certification context, AI tools are fully available and, when used responsibly, genuinely powerful for reaching GPS quality research. The GPS doesn't restrict what tools you use. It sets a standard for how credible your conclusions are. And four AI tools can help you reach that standard in ways I'm going to show you today. Now, I want us to have a concrete example running through this entire episode. So I'm going to introduce you to an ancestor all called Thomas Alvin Whitfield. Thomas is a teaching composite I built from the kinds of situations our community runs into all the time. He was born around 1852 somewhere in Ohio. He shows up in Kishakten County in the 1880 census. He's there again in 1900 and then he disappears. What makes Thomas an interesting teaching example is that his name is spelled three different ways across four documents and his father's birthplace changes between 1880 and 1900. The 1880 census says Pennsylvania, the 1900 census says Virginia. Those two things cannot both be right. And figuring out why that conflict exists is the kind of problem that the GPS was designed to solve and that AI tools can help you tackle. Our research question for Thomas today, who were his parents and where was he actually born? One more note before we dive in. Today's four tools all have free tears that work perfectly for everything I'm going to show you. Chatchubt is at chatchubt.com. Perplexity is at perplexity.ai and if you want to go even further, their comment browser is a free download that brings perplexity intelligence right into your web browsing. Claude is at clad.ai and notebook.lm is at notebook.lm.google.com. Not a single paid subscription required for today's techniques. Let's go. The GPS defines element one as a thorough search of all sources that might reasonably be expected to provide relevant evidence. The keyword is reasonably. This doesn't mean searching every archive in the world. It means looking everywhere that a careful informed researcher would think to look. Here's the problem most of us run into. We know the big platforms, ancestry, family search, my heritage, find my past. We know the obvious record types, census, vital records, military, but the gap between what we typically search and what a genuinely exhaustive search looks like can be enormous. County tax records, church membership roles, local newspaper archives, fraternal organization membership lists, county histories with biographical sketches, how many of those are in your regular workflow. This is where we start with chat you BT. And if you're relatively new to AI, tools, this is a great place to begin, because Chatubt is probably the tool you've heard the most about. Its strength, the thing it generally does better than most of its competitors, is creative brainstorming and research strategy generation. You give it a situation and it gives you back possibilities you hadn't considered. Here's the first prompt I want you to use today. Take your ancestors basic information and put it into Chatubt exactly like this. Quote, I'm researching Thomas Alvin Whitfield born approximately 1852 in Ohio. I believe he settled in Koshakten County, Ohio around 1875. My research question is, where and when was Thomas born and who were his parents? Please generate 15 research strategies I should pursue, ranging from obvious approaches I may have already tried to creative strategies I might have overlooked. For each strategy, tell me what record type to search, what I might find there and why it's relevant to my research question. When I ran this for Thomas, Chatubt came back with 15 strategies organized across four categories. Some were things I'd already tried, but five of them I hadn't considered. A county history published in 1881 that included biographical sketches of prominent residents. Ohio tax duplicate records going back to the 1840s, land entry records in the General Land Office Federal Database. For eternal organization, lodge records, specifically the independent order of odd fellows, which was well established in Koshakten County in 1870s. And most interestingly, birth and death records from Ohio townships separate from the county registrations that some counties kept independently before standardized vital registration. None of those came from my own insights. All of them came from a 30 second conversation with Chatubt. But now here's the problem and this is where a lot of researchers stop too soon. Chatubt told me what to look for. It didn't tell me where those records actually are today, whether they're digitized or what search strategies work best for finding them. That question, where do I actually go to find this stuff is where we hand off to perplexity. Perplexity is built differently than Chatubt or Claude. It functions as an AI-powered search engine, which means it's actively searches the web and gives you answers with citations attached. For genealogy, that is powerful. When I ask perplexity a research question, it comes back with links to actual archives, actual databases, actual finding aids. So after Chatubt generated my research plan for Thomas, I took the most promising strategies and brought them to perplexity. I asked, "Where are Ohio tax duplicate records held? And are any of them available online for Cochakton County in the 1840s through 1870s? What is the best way to access county history publications from Ohio in the 1880s for genealogy research?" What came back were direct links to the Ohio History Connections Digital Repository, a note that Cochakton County's tax duplicates were partially digitized through the county auditor's office and a link to the internet archive where several Ohio County histories had been scanned. Those are actionable answers with sources behind them. Now, if you want to take perplexity even further, this is a great time to mention Comet. Comet, C-O-M-E-T, is perplexity's free browser, a download for Mac, Windows, and Android, and of two days ago iPhone. Comet makes Comet remarkable for genealogies that it brings perplexity's research intelligence into your actual browsing. So when you're on a county archive website trying to figure out what collections they hold, you can highlight text on that page and ask Comet to explain it, cross-reference it with other sources, or find related collections. It turns passive browsing into active research. Using Chatsubt and perplexity, we now have a research plan for Thomas that goes well beyond the obvious databases, and we know exactly where to find each record type. That is GPS Element 1 taking shape. We haven't found our evidence yet. We've built the roadmap. Our Golden Rule is doing exactly what it's supposed to do here. AI is your research assistant, not your researcher. That's UBT and perplexity map the territory. You still have to travel it. Before we move into the analysis, let me take 60 seconds on Element 2 because it applies to every record we find from here on out. The GPS requires that every statement of fact has a complete and accurate source citation. The gold standard reference for genealogical citations is Elizabeth Sean Mills' evidence explained now in its fourth edition. If you don't own this book, it belongs in your toolkit. Here's how Cloud helps with Element 2. If you have the key details of a source, you can describe it to Cloud and ask it to format a citation in evidence-explained style. You describe the document type, the repository, the platform you access it on, the specific identifying details like page numbers or image numbers, and Cloud generates a proper instruction citation for you to verify. I need to say that again to verify. Never paste a Cloud-generated citation into your researcher report without checking every single element against the actual record. AI occasionally inverts numbers, adds details that aren't there, or omits required elements. Cloud gives you the structure. You confirm the substance. And here's something the professional genealogy community is increasingly aligned on. When AI assists you with any part of your research process, transparency is the right path. A brief note in your methodology, something like "Document Analysis and Research Strategy" assisted by AI tools with all source information verified against original records, "is honest and appropriate." It doesn't diminish your work. It makes your process transparent, which is one of the five guiding principles from the Coalition for Responsible AI and Genealogy. Now we get to the part of GPS work that most researchers find the most challenging. And this is where Cloud becomes the most important tool in today's workflow. Let me tell you where we are with Thomas. After following the research plan from Chat's UBT and using perplexity to locate the records, I found him in the 1880 census listed as "TOS Whitfield," age 28, birthplace Ohio, father's birthplace, Pennsylvania. I found him in the 1900 census as Thomas Whitfield, age 49, birthplace, Ohio, father's birthplace, Virginia. And I found an 1887 land deed in Cushuctin County naming "T.A. Whitfield," formerly of Licking County. Three records, multiple conflicts. His name appears three different ways. The land deed places him in neighboring Licking County before Cushuctin, which was new information. And most significantly that father's birthplace, Pennsylvania in 1880, Virginia in 1900. Those two things cannot both be correct, or there needs to be a very specific explanation for why they differ. For AI, I would read through each document, take notes, and try to hold all this in my head. That works. It's just slow, and it's easy to miss things when you have multiple documents and dozens of data points. Here's what I do now. I describe each document's information in clear text and give it to Claude with this prompt. This is the second prompt I want you to copy today. Quote, "I'm researching a person I believe is the same Thomas Alvin Whitfield across three documents. Here's what each document says. Document 1, 1880 US Census, Cushuctin County, Ohio, named THOS Whitfield, age 28, birthplace, Ohio, father's birthplace, Pennsylvania, mother's birthplace, Ohio, occupation, farmer. To 1900 US Census, Cushuctin County, Ohio, named Thomas Whitfield, age 49, birthplace, Ohio, father's birthplace, Virginia, mother's birthplace, Ohio, wife, Martha, three children and household. Document 3, 1887 Landied, Cushuctin County, Ohio, document references TA Whitfield, formerly of Licking County as a grantor. Please create a comparison table showing what each document says about full name, age or calculated birth year, birthplace, father's birthplace, mother's birthplace, county of residence, and any other notable details. identify every disability. discrepancy between the documents. For each discrepancy, suggests at least two specific explanations I should investigate." What Claude produces is a clean, organized comparison table in about 15 seconds. It catches the discrepancy's already saw, and it flagged one I had half noticed but not fully thought through. The calculated birth year from the 1880 census, 1852 and from the 1900 census, 1851, differed by one year. That single year gap is common in census records and often means nothing. But Claude flagged it because it might indicate the 1900 informant was someone other than Thomas himself, which matters when we're evaluating how reliable that informant was about Thomas' father's birthplace. For the Pennsylvania Virginia conflict, Claude suggested three possible explanations. First, the informants for the two census entries were different people with different knowledge. Thomas himself in 1880 and possibly his wife Martha in 1900, who may not have known her father-in-law's exact birthplace. Second, parts of what was Virginia before 1863 became West Virginia after the Civil War, and a father born in the border region might have his birthplace recorded either way depending on who answered and when. Third, a census enumerator mishearing what was said. Now, do any of these prove anything? No. Their research directions. And that is exactly what AI should do in GPS element three. Help you see what needs investigation, not decide what the answer is. Here's where our story takes a turn and where perplexity comes back in for something it does better than any other tool in today's workflow. That second explanation Claude raised about the Virginia West Virginia boundary needs historical context. I needed to know whether this was actually plausible, whether there were counties in that border region where this kind of confusion could genuinely arise. And I needed that answer with actual historical sources behind it, not just AI confidence. So I went back to perplexity and asked, quote, "When did West Virginia become a state in which counties in the Virginia West Virginia border region were more commonly recorded inconsistently in genealogical records after 1863? What does genealogical guidance say about interpreting Virginia versus West Virginia birthplace listings and post-Civil War census records?" What perplexity came back with stop me cold? It confirmed that West Virginia was admitted to the Union on June 20, 1863, card from Virginia's Northwestern counties. It cited a specific genealogical guidance resource pointing out that the individuals born before 1863 and what became West Virginia would legitimately have two different correct answers for their birthplace depending on how the question was framed. A father born in, say, Dodger's County before 1863 could truthfully say Virginia for where he was born, and by 1900 that same county was firmly established as West Virginia. Family members who grew up after 1863 might not even know the nuance, and a census enumerator recording from memory what a daughter-in-law said could easily write Virginia when the more precise modern answer would have been West Virginia. And perplexity gave me the sources. A link to the West Virginia encyclopedia on the state hood question. A link to a Gen Web guidance page on interpreting Virginia versus West Virginia records. Actual citations, not just confident assertions. That is the GPS in action. Not one tool, a workflow. Chat GBT generated the research strategy. Perplexity found the repositories. Plot analyzed the documents and identified the conflict. Perplexity came back in to provide the cited historical context needed to evaluate one of Claude's suggested explanations. And now I had a specific, defensible theory for the Pennsylvania Virginia discrepancy. Thomas's father was likely born what was then Virginia but became West Virginia in 1863, and the two census informants gave different versions of the same true answer. That theory needed one more thing to be confirmed. A record documenting Thomas's father directly, ideally a death certificate or church record listing his birthplace. I found it. A county death registry entry for a man named Eli Whitfield listed as Thomas's father in the same land deed, showed a birthplace of Dodger's County, Virginia. The date of that record was 1878 before Dodger's County had been West Virginia for 15 years. That single record resolved the conflict. Eli Whitfield was born in what was Virginia, died in what was Ohio, and his birthplace was recorded by two different people, 22 years apart, using two different correct answers for the same county. GPS element for resolution of conflicting evidence complete. But I want to be very clear about who resolved that conflict. I did. Claude gave me the framework for thinking about it. Perplexity gave me the historical context and the sources. But the judgment call, deciding which explanation was more credible and what additional information would confirm it, that was mine. The GPS requires your judgment. These tools sharpen it. They don't replace it. That's our golden rule in action at every step. Let me tell you something about element five that surprised me when I started working with notebook L.M. The GPS requires that your conclusion be soundly reasoned, meaning the logic follows from the evidence, and coherently written, meaning it's organized, clear, and documented away. Another researcher could follow and evaluate. This is where genealogy research becomes genealogical writing. And it is the step that most family historians, including me for most of my career, never actually take. We do the research. We accumulate the documents. And then we move on to the next ancestor without ever sitting down and writing out what we actually concluded and why. That's a real problem. Not because someone is grading us, but because if you haven't written it down, you haven't really finished the proof. And if something happens to you, or if you walk away from the research for two years and come back, all of that analysis lives only in your head. It doesn't serve your family. It doesn't serve future researchers. It disappears. Element five is what turns research into a family history. And notebook L.M. is the best tool I've found for the helping with this step, specifically because of one fundamental thing that it makes different from the other three tools we've used today. notebook L.M. is not a chatbot in the traditional sense. Chatupy T, Claude, and perplexity all have broad knowledge from their training data and bring that knowledge to your questions. notebook L.M. works differently. You upload your own research materials and notebook L.M. becomes an expert specifically on your sources. It reads what you've given it. It answers questions grounded in those materials. And because it stays grounded in your actual uploads, it almost never invents information that isn't in your sources. That extremely low hallucination rate is exactly why notebook L.M. is the right tool for element five. You're building a proof summary from your actual evidence, not from AI imagination. Here's how I use it for Thomas Whitfield. Go to notebooklm.google.com and create a new notebook. Name it something specific like Thomas A. Whitfield research. Then upload your relevant materials, the text of the census records, the land deed, the county death registry entry we found for Eli Whitfield, the comparison table Claude produced, any research notes you've taken. notebooklm accepts PDFs, text documents, Google docs and pasted text. If you have scanned images, convert them to text first. Once your sources are loaded, ask notebooklm this question. This is the third copy paste prompt for today. Quote, based on the sources I've uploaded to this notebook, please help me build a proof summary for this research question. Where was Thomas Alvin Whitfield of Koshukton County, Ohio born and who were his parents? Please structure your response in three parts. Part 1 list every piece of evidence in my uploaded sources that is relevant to this research question. For every piece of evidence, note the source document and what it specifically says. Part 2 identify any information in my sources that conflicts with other information and describe how each conflict was addressed in my research notes. Part 3 read a 2-3 paragraph proof summary in the third person. State the research question. Summarize the supporting evidence, address the conflicting evidence, and explain how hit was resolved and state the conclusion. After completing Part 3, flag any statements you made that you are less certain about based on the uploaded sources, in quote. I want to be completely clear about what you do with that output. No book Elim's draft proof summary is a starting point, not a finished document. Every statement needs your critical eye. Does this accurately represent a record? Does the conclusion actually follow from the evidence? Is every source that cited actually saying what it claims? The blank page problem is real. Most of us know roughly what we want to say in a proof summary. We just don't know how to start. No book Elim gives you something to react to, something to shape into your own written conclusion. The writing is still yours. The reasoning is still yours. The AI organized and drafted so that you can do that shaping faster and with more confidence that nothing was missed. For Thomas, no book Elim's draft was three paragraphs. It accurately summarized a three document evidence set, correctly flagged the Pennsylvania Virginia conflict, and in part two it found my uploaded research notes about the West Virginia statehood context and incorporated that explanation clearly. The part three draft was the best starting point I've ever had for a proof summary. I revised two sentences, corrected one slightly imprecise source reference, and the conclusion was done. Thomas Alvin Whitfield was born in Ohio circa 1852, the son of Eli Whitfield, who was born in what is now Dodgers County, West Virginia, prior to that states formation in 1863. The apparent conflict between the 1880 census recording Eli's birthplace as Pennsylvania in the 1900 census recording it as Virginia is resolved by the 1878 Cushuckton County death registry, which identifies Eli Whitfield with a birthplace of Dodgers County, Virginia, consistent with the pre-statehood recording convention for that region. That conclusion took 11 months of research and four AI tools to reach, but the reasoning is mine. The verification is mine. The sources are documented, and the conclusion is written in a way another researcher can evaluate, challenge, and build on. That is the GPS. That is what we're building toward. By the way, notebook LM has one other feature I want to mention for anyone who learns differently. After you've built your notebook, you can ask it to generate an audio overview of your research. This is a podcast style conversation between two AI voices that summarizes your uploaded sources. It's not a substitute for the written proof summary, but I've found it genuinely useful as a perspective check. When I heard an AI summarize my Thomas Whitfield research and audio form, I caught a gap in my own reasoning that I skipped right over when reading my notes. Fresh ears on your own evidence. Let's pull this together fast because I want you to walking away with a clear picture exactly which tool does what. Chatsubt is your research strategy brainstormer. It's the most widely used AI chatbot in the world for a reason. Give it your ancestor situation and your research question, and ask it to generate 15 research strategies you might not have considered. It will surprise you with what it comes up with. That output becomes the research plan you execute perplexity is your research intelligence tool with cited sources. After chat GPT tells you what to look for perplexity tells you where those records actually live with links to the actual archives and databases. And when you need historical context to evaluate a conflict in your evidence perplexity provides that context with the sources behind it. If you want to go deeper, comment is perplexity's free browser, available as a download that brings the same intelligence into your actual web browsing so you can highlight text on an archive website and get instant research context. Claude is your document detective. Once you have records in hand, Claude does the analytical heavy lifting. The comparison table approach I showed you today, describing each document in text and asking Claude to identify every discrepancy and suggest explanations is one of the most practical techniques in this entire episode. Use it every time you have multiple documents about the same ancestor. And notebook LM is your written conclusion partner. Upload everything you've gathered, ask it to draft a proof summary grounded in your sources, and then revise that draft in your own voice. The blank page problem goes away. All four tools, all five GPS elements, all on free tiers. Here's your homework for this week. Pick one ancestor where you have at least three documents that don't quite agree with each other. Open Claude on the free tier. Run the comparison table prompt I gave you today, describing what each document says. Then take one of the discrepancy Claude identifies and take it to perplexity. Ask perplexity for the historical context that might explain it. Write two sentences in your own words explaining what you found and why you believe one source over another. That right there, running the comparison and writing your own explanation is GPS elements three and four in action. And it took 20 minutes. If you want to go further this week, open a fresh chat chat you BT conversation and ask it for 15 research strategies on your biggest brick wall ancestor. Then bring the most interesting strategy to perplexity and see where the records actually live. And then come share what you found. Head over to AncestorsandAI.com. You'll find a direct link to our private Facebook community right there on the homepage. Post your experience. Tell us what discrepancies Claude caught that you'd missed. Tell us what perplexity found that sent you in a completely new direction. This community is how all of us get better together. Thank you so much for listening to Ancestorsand algorithms. If this episode helped you see the GPS or any one of these four tools a little differently, please leave a review wherever you listen to podcasts. And if you know a fellow genealogist or family history researcher who has been intimidated by professional standards, share this episode with him today. This is the kind of content that can change how someone approaches their research entirely. That's the best way you can help our community grow. Now, if you want to go deeper into GPS aligned AI workflows, I'll put together a companion guide for this episode at AncestorsandAI.com. It includes 12 advanced prompts covering all four tools in depth, advanced chat chat BT research strategy techniques, perplexity workflows for locating obscure repositories, Claude multi-document analysis at a more sophisticated level and a complete notebook LM proof argument pipeline from research log to finish document. Everything in today's free episode gives you a complete and actionable toolkit. The companion guide is for researchers who want to go further. For everything else, including every episode, our private community and the research lab, head over to AncestorsandAI.com. One stop for everything. Next week, we're heading to the Kansas homestead records of the 1870s for a case study about a land claim that appears to have simply vanished from a family's history. I'm going to show you how perplexity and Claude working together unlocked a 50-year mystery in an afternoon. You won't want to miss that one. I'm your host Brian and I will see you next week for another journey into the past powered by the future. Until then, happy researching!