RLP 379: Handwritten Text Recognition with Google AI Studio

Nicole (1s):
This is Research Like a Pro episode 379: Handwritten Text Recognition with Google AI Studio. Welcome to Research Like a Pro a Genealogy Podcast about taking your research to the next level, hosted by Nicole Dyer and Diana Elder accredited genealogy professional. Diana and Nicole are the mother-daughter team at FamilyLocket.com and the authors of Research Like a Pro A Genealogist Guide. With Robin Wirthlin they also co-authored the companion volume, Research Like a Pro with DNA. Join Diana and Nicole as they discuss how to stay organized, make progress in their research and solve difficult cases. Let’s go. solve difficult cases. Let’s go.

Nicole (42s):
Today’s episode is sponsored by Newspapers.com. Hello everybody and welcome to Research like a Pro.

Diana (48s):
Hi Nicole. How are you doing today?

Nicole (51s):
Great. I’m excited to talk about AI and transcription. How are you? What have you been doing?

Diana (56s):
Well, I’ve been working on the practicum for Angela McGee’s course for the Salt Lake Institute of Genealogy, which if people have not done this before, it’s a really neat course and it’s different every single year because Angela has five instructors come and present their cases and then the students have to go try to figure out the case. So I was privileged to get to present a case for this fall session and I presented my Thomas Beverly Royston, trying to find his parents and siblings case, which was really fun. And you know, in the course of preparing the practicum, my part of it, I discovered that there were a few records that were out there discovered by full text search that I didn’t have.

Diana (1m 44s):
And so I wanted to just discuss for a minute, what is reasonably exhaustive research now that we have Full Text Search with FamilySearch You know, I thought I had really done really good research on him, but turns out he went to a county I didn’t know about. And so now I have more information, which is fabulous, but, it does kind of make me wonder, you know, what else am I missing out there?

Nicole (2m 7s):
Yeah, it is tricky and I think at different times in our genealogy journeys we have a different idea of what that is. Right? Because you started this Royston research at the beginning of your professional genealogy career.

Diana (2m 21s):
This was the first research I did. This was like the very beginnings of my research. And so this was clear back in 2003 when we would go up to the library in Salt Lake City and look the Roystons up in books. And I became part of a Royston research group through YDNA and we shared information, but it was all done, old school, you know, microfilm and whatever was published in books, nothing was digitized yet. And so, you know, we felt like we had done really good exhaustive research, but then enter the era of FamilySearch and now things are just so much more accessible and finding things, you know, when they’re not listed in the index can be difficult.

Diana (3m 5s):
So if somebody is just listed in an inventory of an estate or if they are the signer of a will, they’re not gonna show up in the index. So unless you are reading every single county and every single record, you probably are gonna miss some people, which is what I’m finding.

Nicole (3m 22s):
Did you have any idea that maybe he was in a different county? Like was that a, a question you had? Like did he go to other counties or did you just assume he had always lived in that same place?

Diana (3m 32s):
Well, he moves around so much and so I had him in a land grant in 1815 in Bullock County. This is in Georgia, which was a few counties south of where he was originally up in Green County in Oglethorpe. And I had searched all the Bullock County records and couldn’t find anything else but Full Text Search found him in Montgomery County and in some Jefferson County records for court cases that were neighboring counties. And so, you know, there you go. I, I could have looked in all those county books, however, he’s not the main person of interest in any of those and in some of the records they are just a collection of loose papers.

Diana (4m 18s):
That’s one of those instances where you’d have to go through all 700 images to look for him. And so that’s, you know, that’s my question. We’re talking reasonably exhaustive research and not, you know, it’s like a needle in the haystack, would he be there or not? You know?

Nicole (4m 35s):
Yeah. So for reasonably exhaustive research, what is that then? So I was looking in genealogy standards just to refresh my memory of what that means. And it, it’s basically that you try to examine all the potentially relevant sources and the purpose of doing that is to make it less likely that eventually you’ll uncover a source that changes your conclusion on the research question. So the goal isn’t to find every single time your ancestor was mentioned in any record anywhere, right? Because like you said, they could be visiting a relative somewhere we never knew they went to and happened to be a witness on a deed there.

Diana (5m 16s):
Right.

Nicole (5m 17s):
And so if we’re looking for evidence about a research question that doesn’t have to do with that at all, do we need that to reach reasonably exhaustive research on a research question? No.

Diana (5m 29s):
So interesting to think about the ins and outs of that, isn’t it?

Nicole (5m 32s):
Yeah, I think we used to think that we were going to be able to find every single time our ancestor was mentioned on paper. I think that’s an idea that a newbie genealogist has. I know I used to think that, and now that I realize how many times people could be mentioned in court records and witnesses on deeds and that kind of thing, I don’t think that we necessarily will be able to do that. I think the goal sometimes is to find all of those FAN club associations so that we can find their, the siblings and you know, in-laws and all the relatives so that we can piece together their identity and their relationships.

Nicole (6m 15s):
But often we can piece together that information without getting every single mention of the person that exists. But with Full Text Search it is a good question now because it opens that opportunity to us to find more records and I think you’re, you’re hitting on an interesting question about it. Like do we need to find and organize and evaluate every single time they’re found or is this just an opportunity for increasing our research when we do have a brick wall and we can barely find them anywhere and now we can possibly find them on some different counties where we didn’t know they visited or lived?

Diana (6m 55s):
Right. And every project might be different, every ancestor, every situation might be different. But I like what you said about the research question and reasonably exhaustive research for that specific question. So usually we have something that we are specifically trying to do, you know, determine a set of parents and if we’ve already determined the parents who’ve had enough in evidence come together, then we wouldn’t need to seek out every single instance of them and some of these records to answer that question, right?

Nicole (7m 26s):
Yeah. And so here’s an example of something that wouldn’t meet the GPS for reasonably exhaustive research where we have the 1850 census and the 1860 census for a child who’s five and 15 and they’re in the household of this couple. So we decide that’s the parents and there’s no relationships listed in those censuses. And so if we try to write that up as a conclusion, it wouldn’t meet the GPS because we haven’t looked at an other sources to confirm or reject that hypothesis. And so we would need to look at probate records, deed, court tax, whatever else there is in that area. I’m thinking of all of the usual things we look at for Southern research, but we would have to check all the things that normally would be checked for that place and time to minimize the risk that our conclusion would be incorrect.

Diana (8m 18s):
That’s a great example. Absolutely agree with that. Well, it’s fun to think about. It’s something that we can continue to have a discussion on and I think that the whole community of genealogists could weigh in on this because it is something to think about. We’re very grateful for Full Text Search, but we do have to think about where we are with that exhaustive research reasonably or not.

Nicole (8m 43s):
Yeah. And I think that probably Full Text Search doesn’t really change what reasonably exhaustive research is, but it definitely changes our ability to find more records. And so now we are having to question like, how much more do I need to do because I could do it pretty easily.

Diana (9m 3s):
Yeah, well I do have an instance where Full Text Search did point to records that I did not previously have because of county boundary changes and just some really interesting situations with the counties and actually ended up changing one of my conclusions. So you know, it’s, it’s gonna be fun. I am excited to use it to further a lot of different ancestors and see if you can really discover more about their lives. Well, let’s do some announcements. Our next Research Like a Pro webinar series is coming up October 21st. This is a Tuesday at 11:00 AM Mountain Time and the title is Parents for Ellen Cecilia Scott: a 19th-Century Irish Immigrant DNA Case Study.

Diana (9m 52s):
Our presenter is Melanie Whitt and we’ll be talking all about this Ellen, who was an orphaned Irish American born in 1860 and she had conflicting historical records about her origins and parents, who were Mark Patrick Scott and Margaret Cox. Using autosomal DNA analysis, pedigree triangulation and Gephi network graphs with Ancestry DNA data, researchers identified her genetic networks and traced her maternal and paternal lines to specific US and Irish localities revealing the candidates for both parents despite the lack of official records. We are excited to learn from Melanie.

Diana (10m 32s):
She is one of our research team members here at Family Locket Genealogists and she has a Master’s degree and UK accreditation. She specializes in International Research and DNA Analysis and has solved many biological family identity cases for clients. Our next study group will be our DNA study group, which begins February, 2026. And if you’re interested in being a peer group leader, please apply. Application is on our website. And then be sure to join our newsletter, which comes out every Monday. We are looking forward to the Texas State Genealogical Conference, which is coming up soon on November 7th and 8th.

Diana (11m 16s):
It’s virtual. Nicole is presenting two classes, one on tax records and one on AI and genealogy. And I am presenting two classes, also one on crafting focused narratives and uncovering female ancestry. So we’re looking forward to that conference. It’s always fun to talk about our research and to learn from others. So we hope you’ll join us.

Nicole (11m 41s):
Today we get to talk about AI, artificial intelligence, and how it can help us with handwritten text recognition. And this has been an ongoing interest of mine over the last year or two as we have seen artificial intelligence get better and better at being able to read and transcribe correctly handwritten text from the past. And today we are going to talk about a 1791 South Carolina deed. And I just wanted to share something that we had learned from the Texas Institute of Genealogy Research where I coordinated the course about artificial intelligence over the summer. And one thing that I learned in preparing the lectures and working with the students that week is that the best tool for transcribing handwritten text right now is Gemini 2.5 Pro, but it only works the best when it’s accessed in its most pure form through the Google AI studio.

Nicole (12m 39s):
And I thought this was really interesting that when you access it there, the quality of the transcription is so much higher than when you use Gemini 2.5 Pro in the Gemini application, you know, when you go to the Gemini chat bot. And Google AI Studio is a web-based tool for developers to prototype and test Google’s Gemini AI models. But what I like about it is, is that it is also simple enough for people like myself and other enthusiasts with AI to explore and experiment without needing to write code or anything. So this simple interface has been the most accurate way to transcribe handwritten documents and why are, you know, why is it better this way than using it with the Gemini app or chat bot?

Nicole (13m 26s):
And what I found out in my research is that other people have seen that the system prompts that Google gives to the Gemini chatbot probably are what’s causing it to produce less accurate results just because they tell Gemini, format your responses this way or do this or that, and so it has all these extra instructions, the system prompts that we don’t see and we can’t change, and that are somehow interfering with the accuracy of the handwritten text recognition. And so when using Google AI Studio to access Gemini 2.5 Pro, that’s when I have found the most accurate results of transcribing this handwritten text from the past.

Nicole (14m 8s):
And so the deed that I used was on three pages of the deed book. So I took three screenshots and then pasted them one at a time into the same conversation and I just gave it the very simple prompt to transcribe and keep the line breaks, just meaning when the handwriting stops from one line and goes to the next line, that’s where I want the transcription to stop and go to the next line and Gemini understand it, understood that instruction perfectly.

Diana (14m 38s):
Well it’s really fun that you discovered Google AI Studio because I have been loving it. I’ve had been doing a lot of transcription and it’s my first choice for doing it now. I agree it’s so much better and I appreciate your explanation of how it works and why it is better than just going to Gemini 2.5 Pro. So let’s just talk a little bit about the example we’re going to share. And so the citation that you created for this goes like this, Anderson County, South Carolina, Mesne Conveyance, A: 367-369, James Gillison to Burgess Reeves, 24 June 1791;

Diana (15m 21s):
image, FamilySearch. Then we have the URL: accessed 21 July 2025. So this is a fairly, you know, straightforward deed. This is like you were saying on three pages and it has a lot of boilerplate language, which we recognize, we see that all the time, the same types of words over and over. And it was book A and then we have the specific pages. So it took up three pages, so let’s see what the transcription said. Are you gonna read that to us?

Nicole (15m 54s):
Yeah, I will. But first let’s talk about Mesne Conveyance. So I never knew how to pronounce that until you tried to pronounce it just now, and I looked it up. So it’s spelled Mesne like M-E-S-N-E, but it’s pronounced “mean” according to Google AI overview when I Google searched it and, and I had to also ask what does it mean because I wasn’t sure. Of course conveyance means like deeding property from one person to another, but why Mesne Conveyance? So Mesne Conveyance means/signifies intermediate or in the middle. So it’s a conveyance that is between the original grant of property from the state or the federal government to another person in the chain of title.

Nicole (16m 41s):
So it’s an intermediate conveyance.

Diana (16m 45s):
Oh, very interesting. I’m very glad to know how to say that.

Nicole (16m 48s):
I never do. And the funny thing is, is that I’ve looked at a lot of Mesne Conveyance books in South Carolina, but never really took the time to figure out how to pronounce it.

Diana (16m 58s):
There you go.

Nicole (16m 58s):
So that’s one of the themes here at Research Like a Pro is learning how to pronounce things.

Diana (17m 4s):
Well I, you know, I am learning French and so I look at everything that I don’t know how to pronounce and think of it immediately with a French pronunciation because that’s what comes to my mind. So there you go. Glad to know it was just simple mean. That’s funny.

Nicole (17m 22s):
Well, mesme does come from Anglo French and Latin, so close. Old French, M-E-I-E-N, meaning middle is where it comes from.

Diana (17m 34s):
Okay.

Nicole (17m 34s):
And of course we can all think of like medium and medium as other forms of the old Latin word. The Latin root is medias. Yes.

Diana (17m 42s):
That makes perfect sense.

Nicole (17m 45s):
Okie dokie. So the reason that I was looking at this deed in the first place is because it was part of my, my Sally Keaton research project and the goal was to figure out if Sally had any children. And all I knew is that she was married to William Reeves and that they didn’t live in South Carolina in 1830 when her father William Keaton died. So I found this Burgess Reeves who lived near William Keaton and I thought this Burgess Reeves was probably the father of Sal, Sally’s husband. And looking at his will, I found that it did list William Reeves as Burgess’s son in Burgess’s will. So then I did a little more research on Burgess and I was trying to find evidence of Burgess Reeves wife being Francis Malden because some of the evidence was naming pattern evidence and many of the people were named with the Malden name as like a middle name or even a first name.

Nicole (18m 41s):
But without strong evidence that Burgess married Francis Malden, I felt like I needed to beef that up. And I did find associations that helped beef it up, but nothing direct, no direct evidence, but I feel pretty confident in it with all the different indirect evidence that was found. Well this deed was found as part of that research and it’s a deed to Burgess Reeves in 1791. And at the time it was in Pendleton District, which became Anderson County eventually. And so I’m gonna go ahead and read this transcription of the deed and then after that we’ll talk about using AI to summarize it or make an abstract from it. So in order to compare the results of these abstracts, we need to hear the whole deed, so here we go.

Nicole (19m 27s):
“This Indenture, made the twenty-fourthDay of June, one thousand Seven hundred and ninety-one, and fifteenth Year of American Independence, Between James Gillison, of the State of South Carolina, Pendleton County, of the one Part, and Burgess Reeves, of the State and County aforesaid, of the other Part – Witnesseth, That the said James Gillison forand in Consideration of the Sum of Fifty-four Pounds Sterling to me in Hand paid the Receipt whereof is hereby fully acknowledged by the Said James Gillison, Hath bargained and sold, and by these Presents doth bargain and sell, make over and confirm, unto the Said Burgess Reeves his Heirs and Assigns for ever, All that Plantation or Tract of Land containing Two Hundred Acres, situate in Pendleton County, on the North Side of Great Rocky Creek, a tributary Stream of Savannah River, and granted to the above-named James Gillison the seventeenth of October, one thousand Seven hundred and eighty-five, by his Excellency William Moultrie Esquire, Governor and Commander in Chief in and over Said State – recorded in Book ZZZZ, Page 114, and hath such Shapes, and Marks, Buttings and Boundings, as will more fully appear by a Plat thereof annexed to Said Grant, To have and to hold Said Tract of Two Hundred Acres of Land with the Appurtenances thereunto belonging, clear of all Incumbrances whatsoever – To have and to hold the Said Tract of Land and Premises unto the Said Burgess Reeves, his Heirs, Executors, Administrators, or Assigns, against him the Said James Gillison, his Heirs, Executors, Administrators, or Assigns, do and will for ever warrant and defend the Title of the aforesaid Tract of Land unto him the Said Burgess Reeves, against the Claim or Claims of any Person or Persons whatsoever – In Witness whereof the Said James Gillison hath hereunto set his Hand, and affixed his Seal, the Day and Year first above written.

Nicole (21m 31s):
Signed, Sealed, and delivered in Presence of us Blake Mauldin, John Harris, Stephen Willet. James E.G Gillison Received the Day and Year within written, from the within-named Burgess Reeves, Fifty-four Pounds Sterling it being the full Consideration therein Specified. Blake Malden signature, James his IG Mark Gillison, South Carolina, Pendleton County – Personally appeared before me Blake Mauldin, and made Oath on the Holy Evangelists that he was present and Saw James Gillison Sign, Seal, and as his Act and Deed deliver to Burgess Reeves the within Instrument of Writing, for the Purposes therein mentioned – also saith on Oath that Gillison Signed the Receipt for the consideration Money – and that John Harris and Stephen Willis were Witnesses with himself.

Nicole (22m 25s):
Sworn to before Elijah Brownr, JP, Blake Malden. So this transcription was pretty easy to read through. There was only one part that was a little bit tricky and that, you know, I wasn’t quite sure what was going on with the transcription and that was at the signature section, but still I think it did a pretty good job with transcribing that section. It was just a little confusing because I told it to keep the same line breaks and doing that, it just had some issues with spacing. But what did you think?

Diana (22m 55s):
Right, I thought it did a really good job and this was a pretty clear image. I thought that the imaging was really good and the original writing was pretty clear and it wasn’t really scrunched together. The lines, there’s plenty of spaces in between lines. So anytime that we have got a pretty good image, I think AI does well. But I thought it did a nice job. Yeah, I didn’t see anything wrong. As you were reading through that and comparing to the original,

Nicole (23m 27s):
Right when I did my original fact check, the two possible errors that I found were that in the first instance of James Gillison’s, mark the EG was transcribed rather then the IG, although the mark wasn’t technically an I or an E, but it was like a mark that resembles an I with a dash through the center of the I. So it didn’t know how to make that mark. So it did an E and then in the first instance of James Gillison’s mark Gemini missed the mark notation below the IG. And so that was also tricky when I was reading the transcription to know what to say because it said it looked like it was going to say his mark, but then it just said his IG Gillison.

Nicole (24m 13s):
And so it was kind of funny, but those were the only errors. And this is remarkable because when, when I have tested using Claude and ChatGPT and other tools to transcribe handwritten deeds like this from the 1800s and the late 1700s, there are usually more errors. And granted, we, I did choose one that was clear to read, it had good handwriting and there weren’t issues with the handwriting, but at the same time the handwriting wasn’t perfect and there were some words that could have been misconstrued, you know, and would’ve, and if I did put this into another tool, I know that there would be errors just because there’s some things that are scrunched at the end of the the line and aren’t really that clear like Great Rocky Creek could be something else.

Nicole (25m 4s):
I could easily see that being transcribed as like Great Rooky Creek or Roki Creek. It, it’s not very clear there, but Gemini did an amazing job.

Diana (25m 13s):
I agree.

Nicole (25m 13s):
And another thing that was neat to see is that when I asked it to keep the same line breaks, it did a really good job with that. And a lot of the time words will break across the lines. And so the way that the clerk wrote it in the deed book, he would write like Se- and then finish the word 17th on the next line. And AI tools hate that they, they used to always get that wrong and they would try to just put the word together and that would cause errors and sometimes it would cause repeating of something or missing of something or just getting the word wrong. So it’s neat to see that that is even correct here. The one question I did have was like, why did it say recorded in book ZZZZ, but that was in the original, so I’m not sure why the, the clerk wrote that in there.

Nicole (26m 8s):
This is recorded in book A of Mesme Conveyance for the county. So I don’t know why they wrote it like that.

Diana (26m 16s):
Interesting. Well one of the things that we want to do after we have this great transcription is to get an abstract to put into our research log. You know, I like to have the transcription, but I also like to have an abstract so that I can quickly review all the really important pieces that come out of that. Because so much of a deed like this is boilerplate language, things that they just use all the time, the same type of wording. And so AI also works really, really well when we’re doing this. So you tried doing an abstract by Gemini with the AI, Google AI studio, and what does that say?

Nicole (26m 56s):
Okay, so what I did, and you can do this in the same conversation where you do the transcription, but I just started a new conversation and pasted in the transcription and said, create an abstract of this transcription and then I pasted it and the results were, An indenture, and then parentheses, land deed dated June 24th, 1791 in Pendleton County, South Carolina, James Gillison of Pendleton County sold a 200 acre tract of land to Burgess Reeves, also of Pendleton County for 54 pounds sterling. And then it has in parentheses the pound symbol for British pounds and then 54. And then it says the land is situated on the north side of Great Rocky Creek, a tributary of the Savannah River and was originally granted to James Gillison on October 17th, 1785 by Governor William Moultrie recorded in book ZZZZ, page 114, Gillison warranted the title against all claims and encumbrances.

Nicole (27m 55s):
The deed was signed, sealed, and delivered in the presence of Blake Malden, John Harris and Steven Willett, a receipt for the full consideration of 54 pound Sterling was also signed by James Gillison. Blake Malden later appeared before Justice of the Peace, Elijah Browne, to affirm an oath that he witnessed Gillison signing, sealing and delivery of the instrument and of the receipt. So that was so great and it helped me understand why it said book ZZ ZZ, those are the grant, the grant books.

Diana (28m 24s):
Right.

Nicole (28m 24s):
I have not looked, but now I’m curious if there is a ZZZZ book of grants.

Diana (28m 29s):
Right, right.

Nicole (28m 29s):
I wondered if that was just a placeholder because they didn’t have the book number or if there really was a book with four z four Zs.

Diana (28m 40s):
Who knows. That’s a good point. Well, I decided to put the same transcription into Claude. Claude is one of my favorite AI tools. And Claude, when I just asked it to do, do an abstract, actually organized it. So instead of a paragraph like Nicole you just read, I got something that was broken up into categories. And so first of all it gives it a title and it says Abstract of land deed, James Gillison to Burgess Reeves. And then it identifies document type is an indenture (land deed) and the date is June 24th, 1791, 15th year of American Independence, the location, Pendleton County, South Carolina.

Diana (29m 24s):
So those are all just basically bullet pointed at the top to give us those specifics. And then it has the bullet point parties and it lists grantor James Gillison of Pendleton County, South Carolina and grantee Burgess Reeves of Pendleton County, South Carolina. So it pulls those two names out for us. And then it has another bullet called transaction details and breaks this down. So consideration, it was 54 pound sterling and it has parentheses the the pound sign and the 54, the property is 200 acres plantation/tract, land location, north side of Great Rocky Creek, a tributary of Savannah River, Pendleton County, South Carolina.

Diana (30m 13s):
So we really easily can see specifically the land, the amount, the location. And then another bullet point is property history. So the original grant date, October 17th, 1785, original grantee James Gillison. Grantor of original grant, governor William Moultrie, Esquire, and then the recording reference the book ZZZZ, page 114. So that was nice that it pulled that out, that that was the history really made you see that if you hadn’t noticed that before. And then the legal provisions and other bullet point property conveyed with a full warranty of title, clear of all encumbrances includes all appurtenances.

Diana (30m 54s):
And then another bullet point witnesses, Blake Malden, John Harris, Steven Willett or slash Willis, and it says parentheses discrepancy in transcription. So noted that there was something not specific or just couldn’t tell exactly what that was. And then the final bullet point was acknowledgement by Justice of Peace, Elijah Browne, JP and Oath witness, Blake Malden testified to witnessing signatures and deed delivery. And then it adds, a note for genealogical research, James Gillison appears to have signed with his Mark (James, his EG Gillison and James, his IG Gillison mark) suggesting limited literacy.

Diana (31m 37s):
There’s a transcription discrepancy regarding one witness’s name Willett versus Willis. So it’s pulling out some of those things that maybe we need to be aware of as genealogists, which I thought was very interesting. So Claude tends to do a little bit more editorializing, you know, giving you some more information, pulling some things out. And so I find that I really have to go back and look and make sure that I agree with those, not just take it at its word because you know, it may, may have said some things that I don’t really agree with, but the basic facts are all there.

Nicole (32m 10s):
Oh, I love that. Yeah, that was great. It was really a useful thing, genealogically, because it gave you some things to check. And so while you were reading that at the end there, I checked Willett versus Willis and in the first signature of the witnesses section it says Steven Willett, but the T at the end of Willett was either written over with an S or the T was written over the S or I don’t know which one was written first, but you can see that it was both. And so there was some confusion about his name, spelling and what it was. And then at the very end, after the consideration was mentioned, there was another signature, or it wasn’t a signature, it was just the fact that Steven Willis was a witness.

Nicole (32m 56s):
It does say Willis that time. So it’s interesting that there were the two different spellings of his name with one of them being either Willes or Willette with a ET or ES. So that’s a good thing to note that we’re not really sure what his name was. It’s unclear and the transcription wasn’t wrong, it just was unclear in the original.

Diana (33m 17s):
Right. And that happens and it’s just something for us to be aware of because if we’re looking for that person, we’re not sure exactly how their name was spelled or what their name was, we probably need to look for both variations of that. So I noticed that in the basic abstract that Gemini did, that it didn’t pull that out, it didn’t say anything about the Willett versus Willis. So that was something extra that Claude added to its abstract.

Nicole (33m 51s):
Yeah, one thing that’s interesting also is that when I asked Gemini in Google AI Studio for this abstract, it gave me two responses and then I had to pick the one I liked better. And the one that the style that you shared from Claude was the other option that I had. And so it, it understood that there’s kind of two ways to do an abstract, the bulleted list format versus the paragraph format. And when I’m putting it into my research log, I prefer the paragraph format. It’s a little easier to put into a spreadsheet or a research log in Airtable. Whereas if you’re viewing it in a Word document or just wanting to read it there in the AI chat bot, it’s a little easier to understand it and take it all in if it’s broken up into headings and bulleted lists.

Diana (34m 41s):
Right! And nothing says you can’t have both, but you do have to decide how you’re going to get it into your research log or your research notes, whatever you are doing. Well now let’s have a word from our sponsor. Ever wondered what life was really like for your ancestors? With Newspapers.com you can explore the world they lived in through their eyes, in their time. Search over a billion newspaper pages dating from 1690 to 2025 and uncover the stories that shaped your family’s past from birth and wedding announcements to obituaries and community news. These pages hold the milestones and everyday moments that connect you to your roots, but it’s more than names and dates, it’s about standing in their shoes.

Diana (35m 24s):
Picture reading the local paper your great grandmother once held or stumbling on a forgotten article about your grandfather’s first business. These aren’t just clippings, they’re pieces of your family story. Newspapers.com turns curiosity into connection. Start your journey today at Newspapers.com because sometimes the past isn’t just history, it’s a story waiting to be discovered. Come make infinite discoveries today on Newspapers.com. Use promo code FamilyLocket for a 20% discount on your subscription. So I also wanted to share that when I am doing something like a complicated court case and using AI to transcribe it can be really helpful to just take all the different documents, like I did one yesterday and it had several different pieces of it, and I put those all into a Google Doc and it had the transcriptions after each transcription, then I would do the abstract and then the transcription and the abstract of of say, I think there were about six different records I was trying to understand.

Diana (36m 29s):
But then I took the all of that, you know, the entire transcription and asked Claude, I used Claude again to make sense of it and put it in chronological order and to understand, help me understand what was going on in this court case. And that was so helpful. That’s been one of my favorite uses of AI is to take these records that can have so many dates and things. You know, like in a court case, you’ve got the original appearing in the court, you know, for the person who’s the plaintiff, and then they have to go out and give notice to the defendant and then one of them comes back and then another one goes out. You know, there’s these things, these comings and goings that can be in a court case and then finally they appear and then there’s a trial and the dates can be just really tricky.

Diana (37m 16s):
And so I’ve really enjoyed having AI help me to make sense of those and bring it all together in a way that I can understand the case.

Nicole (37m 26s):
That’s so helpful. Well, it is exciting to see that we are finally reaching a point where AI transcriptions of handwritten text are highly accurate. And of course, this relies on clear images of legible text. If you use a poorly imaged or illegible document that’s hard for you to read, AI will struggle to read it as well. But if you do, try putting handwritten document into Google AI Studio and using Gemini 2.5 Pro, then compare the transcription results with the original and share with us how accurate it was. We’d love to hear how it goes for you. And one note about using Google AI Studio is that there are different models that you can select.

Nicole (38m 10s):
And right now the default model is not Gemini 2.5 Pro, it’s the new Nano Banana, which is like a Gemini 2.5 Flash Image preview. And so it might work just as well for transcription, I’m not sure. But I did try using it for a task that needed to have a really large context window, which is one of the great things about Gemini is that it has a 1 million token context window, which is like 1500 pages of text. And when I had Nano Banana selected by default on without my realizing it, I couldn’t do it because the context window is smaller for that model. So the way to change the model, on the right hand side of Gemini is to click on the gray box at the top right under run settings and it will by default be selected to Nano Banana. But when you click on that, you can change it to Gemini 2.5 Pro.

Nicole (39m 7s):
So after you select that model Gemini 2.5 Pro, then you’re all set to use the Google AI studio. And it’s great that it’s free. It does show you the prices that it would cost developers who are developing apps, but it doesn’t charge you that cost when you’re just using the AI studio to do simple transcription tasks and things like that. So if you do design like a web application and then you want to incorporate Gemini within that, then you would be paying for that. But using Google AI Studio in AI studio.google.com is free for practicing and playing with it and using it for simple uses.

Diana (39m 48s):
Thanks for taking us through that because that’s really helpful. It looks different. Google AI Studio, as you said, is for developers and it doesn’t look like some of the other chat bots because there’s all this extra information. It can look kind of confusing, but I like that it gives you so many options and I have tried it a bit for the images and it’s really fast and it’s kind of amazing. So very fun. The other thing I love about the image generation is that down in the bottom right hand corner, it has a little AI symbol to let people know that this was AI generated, so I didn’t have to go put that there myself. I love that it showed that. Actually, I think it’s showing the little Gemini icon, that little kind of a diamond.

Diana (40m 32s):
It’s what it’s showing so that you know it was created by Gemini, which is awesome. Well, time to wrap this episode up. We hope everybody listening has been inspired to go try out Google AI Studio and see how it works for your transcriptions of historical records. This is one of the things that can really help us as genealogists to do better research as we understand more about our records and transcribe them fully. And we don’t have to do it all on our own. We of course have to be the human in the loop, as we say, that makes sure everything is absolutely accurate. But with practice, we can learn how to use these tools to really help us make more progress.

Diana (41m 13s):
So thanks everyone for listening, and we will talk to you next time. Bye-bye.

Nicole (41m 17s):
Bye-bye. Thank you for listening. We hope that something you heard today will help you make progress in your research. If you want to learn more, purchase our books, Research Like a Pro and Research Like a Pro with DNA on Amazon.com and other booksellers. You can also register for our online courses or study groups of the same names. Learn more at FamilyLocket.com/services. To share your progress and ask questions, join our private Facebook group by sending us your book receipt or joining our courses to get updates in your email inbox each Monday, subscribe to our newsletter at FamilyLocket.com/newsletter. Please subscribe, rate and review our podcast. We read each review and are so thankful for them. We hope you’ll start now to Research Like a Pro.