Ep. 27: AI Tools for African American Genealogy and the 1870 Brick Wall

For millions of African American families, the search for ancestors hits a wall at 1870. Before that year, the federal census did not list enslaved people by name. They appeared only as ages and numbers in slave schedules, as property in estate inventories, as entries without identity. The 1870 census was the first time most formerly enslaved African Americans were documented by name in any federal record. That moment of visibility is where most family history research begins and, too often, where it stops.

This episode of Ancestors and Algorithms is dedicated to breaking through that wall using free artificial intelligence tools available to every researcher right now.

We follow a fictional but realistic research case centered on Louisa, a formerly enslaved woman in post-Civil War Georgia. Through her story, host and AI genealogist Brian demonstrates a complete multi-tool AI workflow that takes researchers from a named ancestor in the 1870 census back into Freedmen's Bureau records, labor contracts, marriage registrations, and ration registers from the years immediately following emancipation.

In this episode you will learn why searching Freedmen's Bureau records by full name often fails and what experienced African American genealogists do instead. You will learn how to use Perplexity AI to build a state-specific research strategy accounting for surname adoption patterns among formerly enslaved people. You will learn how to use Gemini through Google AI Studio to transcribe faded handwritten Reconstruction-era documents. And you will learn how to use Claude to compare multiple records simultaneously, spotting connections that are nearly impossible to catch one document at a time.

Every tool in this episode is available on a free tier. No paid subscriptions required.

Freedmen's Bureau records are not just genealogical sources. They are the first official acknowledgment that millions of people existed, had names, had families, and were making choices about their lives. AI can help researchers find those records faster. But the meaning of what is found belongs entirely to the families whose ancestors made those marks on paper.

The 1870 Brick Wall is not the end of the story. It is the beginning of a different kind of research.

Topics covered: African American genealogy, Freedmen's Bureau records, the 1870 brick wall, formerly enslaved ancestor research, surname adoption after emancipation, AI-assisted genealogy, free AI tools for family history, Reconstruction era records, labor contracts, marriage registrations, Perplexity AI, Gemini handwriting transcription, Claude document analysis, NotebookLM, and Black family history research in the American South.

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.

Transcript

She signed her name with an X, not because she couldn't write, we don't know if she could or couldn't, but because in 1866 in the state of Georgia, a woman named Louisa made a mark on a piece of paper that would become one of the most important documents ever created for the descendants of enslaved people in America, a labor contract. Witnessed by a freedman's bureau agent. Three lines of faded ink that said simply, she existed. She had a name. She chose. And for one listener, maybe you, that X might be the only thing standing between you and your great, great, great grandmother. Today, we're going to talk about how AI is helping researchers find those Xs and what happens when you do. Welcome to Ancestors and Algorithms where family history meets artificial intelligence. I'm your host, Brian, and I'm recently recovering from a cold, but today we're doing something I've been wanting to do since the very first episode of this podcast. This is episode 27, and we are dedicating it to African-American genealogy, specifically to one of the most challenging and emotionally profound research journeys any family historian can take. We're airing this episode in February, which is Black History Month, here in the United States, and I want to be clear from the start. I approach this episode with deep respect and genuine humility. I am not African-American. The research challenges we're talking about today don't affect my family tree the way they affect millions of others, but I am a genealogist, and genealogists believe, at our core, that every family story deserves to be found, told, and honored. So let's get to work. Today's case study follows a fictional researcher, we'll call her Maya, and her quest to find her ancestor Louisa. We're going to break down exactly how AI tools help Maya navigate one of genealogy's most well-known and heartbreaking obstacles. And along the way, I'll share the exact prompt she used, prompts you can copy, paste, and put to work today. Let's go! 

Maya had been staring at the 1880 census for three years. There was her ancestor, plain as day. Louisa Caldwell, 42 years old, listed as a farm laborer in Decatur County, Georgia. Living in her household were two daughters, Nettie, age 16, and Pearl, age 8, and a son named Amos, who was 19. Three years. That's how long Maya had been trying to find out where Louisa came from, who her parents were, where she was before 1880, and most importantly, what was her life like before emancipation. Because Louisa would have been approximately 27 years old when the Civil War ended in 1865. That means she spent the first 27 years of her life enslaved. Now here's the thing about African-American genealogy that every researcher in this space eventually learns. There is a place in the timeline where the trail doesn't just get harder, it nearly disappears. Genealogists call it the 1870 brick wall. Here's what that means. The 1870 federal census was the very first U. S. census to list formerly enslaved African-Americans by name. Before that, in 1850 and 1860, enslaved people appeared only on something called the slave schedule, where they were listed by age, sex, and sometimes a physical description, but not by name. They weren't people in the eyes of that record. They were property. So for millions of African American families, the research trail often stops cold in 1870. Or, if they're lucky, 1865, when records from a remarkable federal agency called the Freedmen's Bureau begin to appear. Maya had found Louisa in the 1870 census, living in the same county with older daughter. She had no husband listed. She had no husband listed. She had no husband listed. She was 32 years old with a listed birthplace of Georgia. And that's where the trail had stopped. No birth record. No marriage record. No record of any kind before 1870. But Maya had heard about the Freedmen's Bureau. She searched the index databases and found nothing under Louisa Caldwell. So she did what a lot of researchers do at that point. She assumed there was nothing to find. She was wrong. And a combination of AI tools would help her prove it. This is the case study of Louisa and the X that changed everything. So she knew there was something that happened to her. This is the case study. And that's the case study of the law. But where is it? It's not the case study. And the law is being the same. And this is the case study. And that's the case study. And this is why our golden rule matters now more than ever. AI is your research assistant, not your researcher. Because what these tools did was help Maya see what she'd been missing, not invent something that wasn't there. Maya's first mistake. And I say that gently because we've all been there. We're searching only for Louisa Caldwell in the Freedmen's Bureau databases. Here's the thing. After emancipation, surnames were often fluid and in transition. Some formerly enslaved people took the surname of their last enslaver. Some chose an entirely new name. Some kept a name they'd been using quietly within their community for years, a name that meant something to them, not to the people who owned them. And that surname might change again before the 1880 census pinned it down. So searching for Louisa Caldwell in 1866 records was, genealogically speaking, like looking for someone who might not have had that name yet. Maya knew she needed to rethink her strategy, and this is where she turned to perplexity. Now, why perplexity for this step? Because what Maya needed wasn't document analysis. She needed historical context. She needed to understand the landscape of post-Civil War Georgia, the record systems that existed, and the surname adoption patterns of that era. Perplexity is built for exactly this kind of cited, web-based research query. It pulls from current sources and gives you attribution, which matters a lot when you're building a research foundation. Here's the prompt Maya used. It's one I want you to write down or screenshot because you can adapt it for any state or time period. Quote, I am researching an African-American ancestor named Louisa who appears in the 1870 and 1880 federal censuses in Decatur County, Georgia. She is listed with the surname Caldwell by 1880. My research was stalled before 1870. Please help me understand what types of Freedmen's Bureau records existed for Georgia from 1865 to 1872. What are the common surname adoption patterns for formerly enslaved people in post-war Civil War Georgia? And what repositories or databases currently hold searchable Georgia Freedmen's Bureau records? End quote. That prompt does a few important things. It gives Perplexity the full context. Who you're looking for, where, and when. It asks three specific questions rather than one vague one. And it asks for repositories, which means you'll get actionable next steps, not just background information. What came back was a roadmap. Perplexity outlined that Georgia's Freedmen's Bureau records include labor contracts, ration registers, hospital records, marriage registers, letters, and complaints filed by free people against employers. It noted that these records are available in digitized form through the Smithsonian National Museum of African American History and Cultures search portal, through FamilySearch, and, critically, through Ancestors Free Freedmen's Bureau collection, which holds over 3.5 million index records. But here's the piece that changed Maya's entire approach. Perplexity noted that surname searches alone are often insufficient. Because of the fluidity of names in the 1865-1870 period, experienced researchers also searched by location, by age, and by first name only, looking for a Louisa in Decatur County area records, regardless of surname. Maya had been searching for the wrong thing. She immediately went back to the Ancestry Freedmen's Bureau collection and searched, first name Louisa, State Georgia, with no surname at all. She found 11 results, and one of them stopped her breath. The record was a labor contract from 1866, filed through the Freedmen's Bureau field office in Thomasville, Georgia, which is in Thomas County, just north of Decatur County. The document listed an employer, a plantation owner named J. W. Caldwell, and among the freed people who had signed the labor agreement, six in total, was listed a woman named Louisa, age approximately 28. With her, two children, listed only as girl, age 4, and boy, age 5. Maya stared at that. Louisa, 28, in 1866, that would make her approximately 32 in 1870, which matched. The children, a girl around four and a boy around five, could, with some math, align with Nettie and Amos in later censuses, and the employer's name, Caldwell. She wasn't born Louisa Caldwell. She chose to be Louisa Caldwell. Or, perhaps more accurately, she chose to keep the name, or was recorded with it, because that was the name she was known by in that community, on that land, in that moment. This is where I need to pause and acknowledge something. For many researchers tracing enslaved ancestors, finding the name of the enslaver is painful. It's a gut punch moment. The person who appears in those records as your ancestor's quote-unquote employer in 1866 was, a year earlier, legally allowed to own them. The surname your family has carried for generations may have come from that person. Genealogists who work in this space, particularly African-American genealogists, talk about this moment with extraordinary grace. Because it doesn't diminish Louisa's story. It deepens it. She survived. She stayed on that land. Maybe because she had no resources to leave. Maybe because her community was there. Maybe because it was simply the only home she knew. 

Now, 

Maya had a document, but she couldn't read it. The labor contract was handwritten in a 19th century script, faded and difficult to parse. She could make out Louisa and Caldwell and a date, but the rest of the text, the terms of the contract, any additional details about Louisa's family were illegible to untrained eye. This is where Gemini Access through Google AI Studio at aistudio. Google.com entered the picture. And I cannot stress enough, when it comes to handwritten historical document transcription, Google AI Studio's version of Gemini has become genuinely remarkable. This is not the Gemini app on your phone. This is AI Studio and it's free. Maya uploaded an image of the Freedmen's Bureau labor contract and used this prompt. Quote, I am uploading an image of a handwritten Freedmen's Bureau labor contract from Georgia approximately 1866. Please transcribe this document as completely and accurately as possible. Use unclear in brackets for any words you cannot read with confidence and place a question mark after any word you are less than 80% confident about. Pay special attention to names of all individuals listed, any ages or descriptions included, the name of the employer or landowner, the county or location, and any conditions or terms of the contract. This is a historical genealogy document and accuracy matters deeply to me. End quote. What came back was extraordinary. Gemini transcribed the full document, not perfectly, but well enough. It identified the employer as J. W. Caldwell, plantation owner, Thomas County, Georgia. It identified Louisa's two children and in the contract terms, it picked up a line that Maya had completely missed. 

That infant would have been born right at the cusp of emancipation. She doesn't appear in the 1870 census household. What happened to her? Now, Maya had a new mystery layered inside the original one. And this is where our golden rule came back into full force. AI is your research assistant, not your researcher. Gemini transcribed what was there. But Maya still needed to verify. She pulled up the actual document image, compared the transcription word by word, confirmed what Gemini had read correctly, noted the two unclear passages it flagged honestly, and made her own assessment. She wasn't going to stake a family history on an AI transcription alone. She used it as a first draft, not a final answer. 

With a verified transcription in hand, Maya turned to quad. Because now she has something specific to analyze, three documents, the 1866 labor contract, the 1870 census, and the 1880 census, and a set of questions about how they connected. Claude's strength is exactly this kind of multi-document comparative reasoning. She uploaded all three documents, or typed out the relevant information from each, and asked Claude to do something specific. Build a timeline of what could be confirmed versus what was inferred, and flag any discrepancies or gaps that needed further research. What Claude returned was a structured analysis that mapped each person across the three documents, noted that the unnamed infant from 1866 did not appear in either censuses, flagged the name Pearl in 1880, who was listed as H8, which would put her birth around 1872, after the labor contract, not before, and pointed out a possible gap. A child born around 1866, who should appear somewhere in 1870, but didn't. Claude also noted something that hadn't occurred to Maya, The 1870 census showed Louisa living with an additional person, an older woman named Ada, listed as 58 years old, with no stated relationship to the household. In many post-emancipation households, older women with no listed relationship were often mothers-in-law, mothers, or community elders who had been part of the same enslaved community. Ada might be Louisa's mother. Maya had come into this research session looking for one person. Now, she had threads to pull on for three. Let's talk about Ada. Using Claude's observation as a launching point, Maya went back to the Freedmen's Bureau records, this time searching for Ada in the same geographic area. She found a ration register from 1865 for the Thomasville area that listed an Ada, age approximately 45, marked as a widow with no children listed in the register. That age aligned if Ada was 45 in 1865, she'd be 58 in 1878, which is close enough to the 1870 census listing of 58 to be plausible. She couldn't prove Ada was Louisa's mother. Not yet. That would take more research. A fan club approach, looking at Fran's associates and neighbors across multiple records over multiple years to see if Ada and Louisa consistently appeared in the same orbit. But the hypothesis was sound. It was grounded in documented evidence, not assumption. This is what good genealogical research looks like, and it's also what good AI-assisted research looks like. You don't leap to conclusions. You build a case, brick by brick, and you're honest about which bricks are solid and which ones are still unverified. As for the unnamed infant from the labor contract, that trail went cold. Maya searched every available database and found no clear match. The child may have died in infancy. She may have been taken in by another family. She may appear in records not yet digitized. Maya documented the gap, noted it in her research log, and moved on. Because sometimes the most honest thing you can do as a genealogist is say, I looked, and I didn't find her. I will keep looking. But here's what Maya did find, and this is the part that made her cry. In the Freedmen's Bureau records, in the marriage registers, she found a record from 1866 of a couple formally registering their union. The Freedmen's Bureau helped formerly enslaved couples legalize marriages that slavery had made impossible to formally document. The couple in this record? Louisa, formerly of the Caldwell Plantation, and a man named Henry Freeman. Henry Freeman. A surname chosen, not inherited. The registration listed that Louisa and Henry had been living as husband and wife for approximately eight years, which would put their union at around 1858, well before emancipation. They had been a family, quietly, in secret, for nearly a decade before anyone recognized them as one. Henry is not in the 1870 or 1880 census. He may have died. He may have left. We don't know. But for a brief documented moment in 1866, he and Louisa stood in front of a Freedmen's Bureau agent and said, We are a family. Write it down. 

Maya uploaded everything she had found, the labor contract transcription, the ration register, the marriage registration, the census records, and her research notes into a Notebook LM notebook. She asked it to create a timeline of Louisa's documented life and to flag anywhere two documents seemed to contradict each other. What came back wasn't just a timeline. It was the outline of a life. A woman who had spent 27 years with no legal standing, who signed her name with an X, and who in the five years after emancipation appeared in at least three different sets of federal records. She had been witnessed. She had been documented. She had chosen her name, registered her marriage, and raised her children on the same land where she had once been enslaved. Not because she had nowhere else to go, necessarily, but because it was hers now. And AI helped Maya find her. All right, let's circle back for a moment and talk about what we actually did today because I want you to be able to replicate this. Here's the workflow in plain terms. We started with perplexity to build our historical and repository knowledge. Understanding what records exist, where they live, and what search strategies experienced researchers use for this era and geography. That's what gave Maya the insight to search by first name only, not by surname. We used Gemini via Google AI Studio. Again, that's aistudio. google.com to transcribe a handwritten document that would have taken hours to decode manually. The key was giving it context, document type, era, and what specific details mattered. We used Claude for multi-document comparisons, uploading multiple documents, and asking for a structured analysis of what connected, what conflicted, and what was still missing. That's where the insight about Ada came from. Not from a database, from a reasoning AI looking across three documents at once. And we used Notebook LM at the end to organize everything into a coherent timeline, turning a pile of research discoveries into a structured picture of a life. Every single tool I mentioned today has a free tier. Not one of them requires a paid subscription to do what we demonstrated. Here's your homework for this week. If you are researching African-American ancestors, your own family, or someone else's, I want you to try one thing. Go to the Ancestry Freedmen's Bureau collection, which is free to access with a free account. Instead of searching by full name, search by first name only with a state but no county, and see what comes back. You might be surprised what you find when you give the search room to breathe. And if you find a document you can't read, upload it to Google AI Studio. Gemini is there. Use it. I want to take a moment before the outro music to say something sincerely. African-American genealogy is not just genealogy. It is an act of restoration. It is the work of saying these people existed. Their lives mattered. Their names mattered. And even when the systems that surrounded them worked hard to make them invisible, there are still traces. Still X marks on labor contracts. Still names in ration registers. Still couples standing before a bureau agent and saying, write us down. If you are doing this research, whether it's your own family history or you're helping someone else find theirs, please know that the genealogy community honors this work. And AI used carefully and responsibly can be one more tool in the service of that restoration. Remember, AI is your research assistant, not your researcher. It can help you find the X, but understanding what that X means, that belongs to you and to the family whose story you're helping to tell. Now, before I sign off, if you want to take this research even deeper, I've been putting together some advanced prompting techniques and workflows specifically for African-American genealogy that go well beyond what we covered today. We're talking detailed prompts and case studies that walk you through things like how to trace the enslaver's family to work backward, how to use DNA results alongside Friedman's bureau findings, and how to build a full fan club analysis using AI assistance. I'll be sharing information about where to access all of that in our Facebook group and after the RootsTech 2026 conference May through 7th this year. But whether you ever see this or not, what you learned today gives you a real foundation to start. Head over to our Facebook group, Ancestors and Algorithms, AI for Genealogy, and tell me, have you tried searching Friedman's bureau records? What did you find? I read every post. Our group just reached over 1,500 genealogists willing and able to help with any questions. You can also reach me at ancestorsandai at gmail.com or visit ancestorsandai.com for more information. Next week, we're heading across the Atlantic to dig into Italian records. And I'm going to show you a disambiguation technique using CLODE that made me genuinely gasp the first time I saw it work. You will not want to miss it. Thank you so much for listening. If this episode moved you or helped you, please leave a review wherever you listen to podcasts. It helps other family historians find us. I'm your host, Brian. Thank you for getting through this with my raspy voice. And I will see you next week for another journey into the past powered by the future. Until then, happy researching.