RLP 337: Handwritten Text Recognition by Claude 3.5 Sonnet

Nicole (1s):
This is Research Like a Pro episode 337 Handwritten Text Recognition by Claude 3.5 Sonnet. 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.

Nicole (41s):
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. Today’s episode is sponsored by Newspapers.com. Hello to all of our listeners at Research Like a Pro, and hello to you, Mom.

Diana (54s):
Hello Nicole. How are you doing today?

Nicole (56s):
Really well. I’m just cruising along on adding documents to my research log for my kinship determination project and I’m really just happy with all of the things I’ve been able to get and recently just putting in a bunch of records that a researcher got from me in Texas at the State Library and archives. So that was just exciting to receive those and to have them for my project. What about you?

Diana (1m 23s):
Wow, that’s really great. Reading reports. These are all reports that our recent Research Like a Pro study group have completed and it’s been so fun to travel to different locations virtually and learn about their research. I’ve read reports that have been in Ireland and Southern United States and New England, all sorts of different fun places and it’s been fascinating. I love reading about other people’s research and it’s been really fun to see the reports where someone had never written a report before and they come up with a great report just following our process. So it’s always really fun at the end of the study group to see, see the progress.

Nicole (2m 6s):
Yes, it is fun and I always learn from other people and the research that they’re doing, getting ideas from their strategies. I read a really good report yesterday that had some really difficult research identifying candidates to be the individual named in the hypothesized father’s will and then tracing them forward looking for patterns, connections back to the family. And it was really neat to see kind of a conclusion emerge from the indirect evidence.

Diana (2m 33s):
Right, right. I’ve read several where there are multiple people of the same name and that they’re going to have to continue on with subsequent phases because as they started doing their research, they discovered all these people with the same kind of common name and which one is theirs. So you know, that can seem like an overwhelming problem to tackle and that’s why you need a really good process to discover that. Well, let’s do some announcements. We’re excited about our Research Like a Pro Webinar Series that we’ll be starting for 2025 on January 18th at 11:00 AM Mountain Time.

Diana (3m 13s):
This will be presented by Susan McKee, who is one of our past study group members. The title is Using Irish Naming Convention to Discover the Family of Thomas Delaney in the Mid to Late 1800s in Ireland. And so this will be all about Thomas Delaney, who was the father of Catherine “Kate” Delaney who married Henry Burge in 1878 in Dublin, Ireland. And Catherine was born about 1860 in Queen’s County (present day County Laois) in Ireland and died in 1909 in Dublin. So to learn more about Thomas evidence was gathered from researching the life and family of his daughter Kate and her husband Henry.

Diana (3m 60s):
This evidence identified a specific geographical location to focus the search for records for Thomas. This case study shows how Irish naming convention can be used as a powerful tool to analyze and predict family names and aid in the search for family units in Irish records in the mid to late 1800s. So topics covered will be Ireland, Queen’s County, Mid to Late 1800s, Irish naming Convention, Catholic Parish Registers, cCivil Registration, 1901 Census of Ireland. Well that will be exciting. That will be something a little bit different from the research that I do and I’m looking forward to learning from Susan. Our next Research Like a Pro with DNA study group begins February, 2025 and if you have thought about putting DNA into your process, maybe this is the time to learn.

Diana (4m 45s):
We do have our peer group leader application on our website, so if you’ve worked through the process in a past study group or an e-course or even through our book, we invite you to apply, join our newsletter for coupons and our latest news and we’re excited about the upcoming conferences for 2025. We’ll have RootsTech, which is March 6th through 8th and the National Genealogical Society Conference, which will be May 23rd to 26th. Both are in person, Nicole and I will be there and we would love to catch up with any of you.

Nicole (5m 20s):
Yeah, we really would. Well, today we’re talking about AI and transcription and specifically back in June I had tried out using Claude 3.5 Sonnet to help with transcribing a deed. This was an exciting thing to do and what I’ve noticed is that over time handwritten text recognition by large language models has become much better and especially with the release of ChatGPT-4o and Claude 3.5 Sonnet.

Nicole (6m 13s):
And so back when I wrote this originally in June of 2024, Claude 3.5 Sonnet was thought to be the smartest large language model available. And now we are recording this in November and the best models according to the Chatbot Arena LLM leaderboard are Gemini, which is available in the Google AI studio and the latest ChatGPT-4o model released in September. And the best large language model is constantly changing. But these latest large language models like Claude ChatGPT-4o are some of the most advanced large language models. So when this came out, many were talking about its ability, ability to transcribe handwritten documents. So I thought I would give it a try and test out Claude 3.5 Sonnet. And this is just one of my favorite tasks that I really hope large language models get really good at doing And handwritten text recognition can be time consuming when we have to transcribe a lot of documents on our own time.

Diana (6m 55s):
And often there are a series of deeds and it’s just can be tedious because a lot of the language is repeated over and over, but you still have to go through it and transcribe it to really get the full picture. So if the large language models can help us do that, it’ll be such a wonderful boon to our research. So I think this is so fun that you were doing a test and I like doing this type of a test also with the different AI models just to see which one performs the best. So what did you use to test Claude 3.5 Sonnet? Well,

Nicole (7m 26s):
I created a manual transcription myself of the first like page of an 1829 North Carolina deed and it’s from Lincoln County Volume 34, page 546, Joseph C. Pearson to Archibald Dyer on January 14th, 1829. And I got this image of the deed from FamilySearch from a digitized microfilm. I only provided this one page because often FamilySearch microfilms will have two pages digitized in the deed book. So imagine like the full spread of a book that’s open, the left page and the right page.

Nicole (8m 7s):
I only provided one of those sides So I cropped it down so that it would be really simple for the language model to read. And then I provided Claude 3.5 Sonnet with the image and the following prompt, you are an expert genealogist and transcriptionist. Transcribe this deed and put anything you can’t read in square brackets with question marks, or your best guess. Preserve the original line lengths in the document.

Diana (8m 36s):
Oh, that’s such a good prompt. I like that you had it tell you where it couldn’t read it and really show that with the brackets. That’s a great idea for anyone who wants to try doing this. And I agree with preserving the original lines, that’s important too because it’s so easy to check it. So how did Claude do with this transcription task?

Nicole (8m 57s):
Well, the result was very accurate. I was very impressed and the main mistake that was made was changing the grantor’s surname from Peerson to person P-E-R-S-O-N instead of P-E-E-R-S-O-N. And I noticed that the surname was spelled a few different ways even within the deed. I think at one time it was P-E-A-R-S-O-N and another time it was PEE, but then Claude changed it all to person with one E.

Diana (9m 29s):
Well, I’ve noticed that it struggles with names too and I think that’s because often they’re not usual words that the AI has been trained on. So sometimes it will just substitute something completely different, which is always kind of funny. And of course the names are very important to us as Genealogists, so that’s something we always have to check. But that’s interesting that it was very close Peerson to person. So how did you check this?

Nicole (9m 55s):
Well, since I already had created my own transcription before I gave it to Claude, I then copied and pasted my manual transcription and pasted it into Claude and said, here is the correct transcription now list all the differences. The response was a list of 15 differences. And to my surprise, in four of the cases, Claude’s transcription was right and mine was wrong. I had missed capitalizing a word. I had omitted a hyphen. I had failed to combine two words that didn’t have a space between them, although they probably should have and spelled a word correctly when it was spelled incorrectly in the document. And this could have been to Google Docs autocorrect, but of course in historical transcription and genealogical transcription, we want to preserve all the spacing, capitalization and spelling errors exactly as they were in the document.

Nicole (10m 46s):
We don’t wanna fix those up, which is really hard for large language models to do even if you tell them to. So I was impressed when Claude did that. It kept a word spelled incorrectly so that it matched the original, whereas I had corrected it inadvertently in my Google Docs manual transcription and two of the other cases, both myself and Claude were wrong. And a closer examination of the original document showed the name of the creek where the land was situated to be Potsis Creek, not Patras or Potters, which is what me and Claude had put. And then the surname of a neighbor was actually Altam, not Alton or Alan.

Nicole (11m 32s):
So there were many possibilities for some of these names, which as you pointed out are one of the most challenging things to transcribe because they’re not words that we know already. We have to actually figure out each letter and try to figure out something that makes sense when you put it together to be a name. And so the other nine differences were slight mistakes by Claude in capitalization spelling Peerson as pressin or person inaccurately transcribing super scripted letters with the quotation mark instead of the super scripted letter, missing an S at the end of the word, missing a period after the C and Joseph C.

Nicole (12m 13s):
Pearson. And so some of these differences were just very minor, but it was good to see, to really compare my transcription that I did at first with the one that Claude did. And this is such a good exercise to do to really understand the the model’s abilities compared with my own abilities. And as I was going through everything with a fine tooth comb, I realized that we both missed something We didn’t super super script the th after the 14, which isn’t really super important.

Diana (12m 53s):
I love that as a little exercise just to practice this and get familiar. I know as I talk to different people about using AI, a lot of Genealogists are still a little hesitant to just jump in and try it. Maybe it sounds daunting or difficult, but it is so easy and it’s so fun and I have to say it’s kind of addicting because it’s just fascinating to see what happens when you give it a prompt and then get the response. But this whole idea of transcribing tedious deeds becomes so much more fun when you’re working with a large language model. You know, there I said it, it becomes more fun. It really does.

Diana (13m 33s):
And so I love this. Well it seems like that Claude was really strong with this capability and it produced transcriptions with higher fidelity to the original without changing the misspellings, it expanded abbreviations and altering unusual capitalizations. So it’s really come a long way. I think when these models first came out and people tried this, it was kind of a mess and maybe, you know, if you tried it back in the early days you just gave up. So it’s exciting to see now how much better it is and that it really accurately transcribed this deed with only one major mistake,

Nicole (14m 15s):
Right? It was really good. And if we compare this to other transcription tools like Transkribus and FamilySearch full text search, it’s much more accurate. It’s so much better and it’s pretty amazing to see the differences. You know, if you’ve looked at the family text, full text search, there will be all kinds of random characters in there and it is really good to help you find your ancestor in deeds and wills and all kinds of records. But the transcription itself needs a lot of fixing up. But this transcription is very usable. So it’s exciting.

Diana (14m 52s):
It is. And one of the things I like to do is to take the document itself and I will highlight the fact that you didn’t put in the whole page, you cropped it. So it was just the specific deed because if you don’t do that it can really confuse the model, which I’ve done that before, just, you know, putting in the whole two page spread. So you always wanna crop it. But then what I like to do is take that, put that in, put in the FamilySearch full text search, you know their transcription, but then ask Claude to then compare, look at the original and then give me a really good transcription. And it does really well with that.

Diana (15m 33s):
I mean you don’t even need to have the FamilySearch full text search, you know, transcription in there as well. But sometimes I like to put that in just to have it have something else to compare to. So just another idea.

Nicole (15m 46s):
I love that. And you know with the new ChatGPT ability to reason and think and do like multiple steps, I think that would be a great prompt to give that version of ChatGPT because then it can look at the original, it can make its own transcription that it can compare that with the one you provided from FamilySearch full text search, then it can look for differences and it can kind of follow all these steps and it can think what to do and it takes longer. I know whenever ChatGPT has done that model where it thinks and tries to reason that it takes a while, but it will give me both options of responses so that I can see which one I need for the certain task I’m doing.

Nicole (16m 33s):
And sometimes I just need the basic model and other times the reasoning model does better. And it seems like anything where you have kind of multiple steps involved, like comparing two different transcriptions and correcting one, that that would be a good use for the reasoning model of ChatGPT

Diana (16m 50s):
I like that. And often I like to have the large language model give me some analysis or to make some conclusions that I can then look at and see if I agree with it. And often in something like probate or a a difficult deed, perhaps it’s understanding of the legal terms is actually better than mine and it gives me some, some better ideas of what was going on because it’s been trained on data that you know that it knows, it knows more than I do. So it’s funny to think about that idea that it could actually help you learn but it can, it can be a great tutor, especially with some of these difficult documents.

Nicole (17m 34s):
Yes, love that. Well I’m gonna put a link in the show notes for anyone who wants to read more about the OpenAI o1-preview, which is a new series of reasoning models that can think more before responding so that you can learn about it. But you might see it in ChatGPT when it asks you which response do you like better, this one that thought for a long time or this one. So it’s exciting to see the development of these models and hopefully they will help us to improve our automatic transcriptions that we’re attempting to do.

Diana (18m 6s):
Exactly.

Nicole (18m 8s):
Now let’s have a word from our Sponsor. Imagine curling up in a cozy chair scrolling through Newspapers.com and discovering a family member smiling back at you in a photo you’ve never seen before. Give the genealogy lover in your life a gift subscription to Newspapers.com and make this a reality. They’ll break through genealogy brick walls with access to hundreds of years of wedding announcements, obituaries, family stories and so much more. There’s always something new to discover on Newspapers.com with a growing archive of over 1 billion pages from the US, UK, Canada, Australia and beyond. Long after other gifts are forgotten, a Newspapers.com gift subscription will continue to delight, make the past the best present ever. For listeners of today’s show, Newspapers.com is offering 20% off gift subscriptions for a limited time just use the code FamilyLocket at checkout.

Diana (18m 52s):
Well we have some tips for everyone listening about transcribing handwritten texts with ChatGPT and Claude. And the first one is to choose images that are in prose or narrative format rather than something like tabulated data. So thinking of a census that has the columns or taxes, those are much harder for the large language model to read. And I discovered this when I was doing my Weatherford project and I ended up doing 126 different rows of tax data and at first I thought, oh this will be so fun, I’ll just put this in Claude and see if it can figure this out. And it, it did not, it was very difficult, it was not accurate at all.

Diana (19m 36s):
So that is not a good use for transcription. However, after I did my own transcription then and I had the data, then it handled the data beautifully but it, it could not do a good accurate transcription of those tax records sadly

Nicole (19m 52s):
Right now I wanna play with that and figure out the best way to use a large language model to transcribe tax records because I’ve been doing a lot of them too. And just, I wonder if it could maybe instead of typing it into your Airtable log first and then putting that in to ChatGPT or whatever, if you could just put in the column headers and have it transcribed those and then set up a table. And you know how Claude has the artifacts on the side where it’s constantly updating adding to the table that you’re making or whatever you could just, every time you get to a new person to extract from that tax record it already has the column headers you could just start typing in like here’s column one, here’s column two, like add this to my table, I don dunno, but then it’s just the same as typing it into your table, so it kind of like is dubious with it.

Nicole (20m 42s):
No, but then you know what I have to always do is go back and scroll to the top and check to make sure I’m putting things in the right, you know, field. So it’s, it’s tricky.

Diana (20m 52s):
It is and I think you just have to experiment and we’re getting a little off topic, but what I ended up doing was in Airtable I created a new base and had all my columns or fields just match the tax list. And doing that it just went super fast. It was so nice and I even had the fields for the value be the dollar amount So I could see if I was typing in the right thing because the little dollar would would pop up dollar sign once I put in the numbers. So that really helped.

Nicole (21m 19s):
That’s much better than Excel then. I mean you can do that in Excel but it’s so easier in Airtable, at least for me because I’m not as good with Excel as I am with Airtable.

Diana (21m 27s):
It was really slick, I was super excited. And then the other thing that helped me know if I was in the right column or field was that I had some single select fields for the water course. And so if I started typing in a number, you know, I was like, oh no, it’s supposed to be pulling up the Trinity or you know Bear Creek and so I could just tell right away, oh I’m in the wrong place. Anyway, it got really fast and it was kind of exciting to see how that worked really, really well and it was even better when I put it all into Claude and had it do all this amazing analysis. So yeah, it’s fun to experiment though with all these different tools we have.

Nicole (22m 4s):
Yeah, especially when there’s a record type that we do a lot of extracting or transcription of when we figure out how we can use the large language model with that type of record, then we can even save that as a saved prompt in ChatGPT or Claude projects and then we’re always ready to analyze or transcribe that document. And that’s something I really wanna work on doing because I think it’s a big time saver that I’m not really taking advantage of yet.

Diana (22m 30s):
I agree.

Nicole (22m 31s):
Okay, so we had one tip so far and then we got off on a tangent. So repeating the tip was that choose images to be transcribed that are in prose and not in tabulated data like census and taxes. So the next tip is to use image files like JPEGs or PNGs instead of PDFs. PDF images are much more difficult for the large language models to read and transcribe. And I’m not exactly sure why, but I think it’s because PDFs can be much more complex. They can include an image and then they can have a text block underneath it. They can have all kinds of things in it. So I think they’re just harder to figure out. So using a basic image file like JPEG or A PNG, it’s just more simple to parse.

Diana (23m 14s):
Exactly, exactly. And as I’ve mentioned before, another big tip is to crop your deed images from FamilySearch. So you only see one page instead of two pages of an open book. You know, you just wanna get the specific transcription. So the tool that I use for this is, you know, a screen capture, I use the Snipping tool and Nicole, I know you use Snagit I think, but you wanna get something that lets you just get specifically what’s showing on your screen and then you can just paste that right in to the large language models. So it’s good to learn how to use your tools that you might have available.

Nicole (23m 55s):
Right. I actually use both. I’ll use snipping tool or Snagit depending on how quickly I want the snip ready because like if I use snipping tool I can paste it right into PowerPoint and if I use Snagit then I have to click a couple more times before I can copy and paste it. So I use both. But one great thing about Snagit is that once I click a couple times I can control S to save it and it it will be saved into a folder on my computer. So I have both of them at my disposal.

Diana (24m 26s):
That’s great.

Nicole (24m 27s):
Alright, so the next tip is to ask for the large language model to preserve the original line lengths for easy comparison with the original. And when I figured this out it was so helpful because trying to compare the transcription from the large language model with the original was kind of a nightmare. Until I did this, it was so hard to figure out where I was and putting them side by side, which is one of the strengths of Transkribus, that it will do that automatically and puts the image side by side with the automatic transcription for you. So comparing that as a breeze, well once I figured out that you can tell ChatGPT and Claude to preserve those original line lengths, then it was great.

Nicole (25m 10s):
Everything was so much easier.

Diana (25m 12s):
It’s funny because when I first started doing transcriptions I didn’t know you were supposed to preserve the original lines and I would just type and let the word processor go to the next line and then I can never find my place in the original, in my transcription. You know, it’s funny how when we first start doing something, we may not think most logically about the best way to do it. Yeah, that’s a huge one. Well another one that I really like is to ask for illegible words to be placed in the square editorial bracket. So you can quickly review any words that the model had trouble recognizing. And this is such a great tip because otherwise your eye might not be drawn specifically to those problem areas.

Diana (25m 54s):
So that’s such a good tip.

Nicole (25m 56s):
Right. And I’ve noticed that it doesn’t put very many things in square editorial brackets, as many as I would think with some of the errors, but there actually weren’t that many errors. And in this example of this deed, there were only two words that were put in the square editorial brackets. It was the name Joseph, and then the word tenement. And it does really help you to call like notice the possible challenging words really quickly and be able to check those. So I love that. I’d like to think of a more detailed prompt that I could save for this that encourages it to do more square editorial brackets. Maybe just saying like put square editorial brackets around every name or every word that’s not in the dictionary.

Nicole (26m 41s):
Then it’ll make it so much easier to go check anything that that is like non-standard like name of a creek or the name of a person or a date. Like put square editorial brackets around every date and every name and every word that’s not in the dictionary. Then you can just so quickly go and and check those.

Diana (26m 57s):
Oh that’s a great idea.

Nicole (26m 59s):
So the next tip is to ask for the transcription to have high fidelity to the original without changing abbreviations and placing super scripted letters in square brackets with the carrot symbol. So for example you could say 14 square bracket carrot symbol, th square bracket. And then that would notify me that I need to use that markdown basically in my final transcription to superscript those letters.

Diana (27m 29s):
I have found that giving the model an example helps a lot. So I’m even wondering if you, you know, do save a prompt that you put some examples in there so you know, just like you kind of showed up with the symbol, but I’m trying to think if there would be other things that we could put in there. It might be fun to start a list of all the common problems and then you could have your prompt and say if you see this, do this and maybe it would down the road help it to be more and more accurate. Right. Another thing that you can do is if you already know the names of the people, provide them in the prompt and then you’ll get a more accurate transcription the first time through. Because as we’ve mentioned, the models really do struggle to read those proper nouns and dates.

Nicole (28m 14s):
Yes. That has worked really well for me in the past. You know, often when you’re transcribing a deed you already know the names of the people in it. Like for example, I was researching Archibald Dyer, So I knew that was his name. So I can just say transcribe this deed that has Archibald Dyer and by quickly glancing at the deed I can see the name of the other party is Joseph Peerson or whatever his name was. So if you just say that in your prompt, it really saves the model from having to struggle to read the names.

Diana (28m 40s):
Yeah. And then you don’t have to waste time in going back and correcting them. However, you don’t really have to go do it yourself. You can just respond to the transcription and say actually the name is Archibald Dyer, go fix it. And then it’ll just fix it in all the areas it needs to.

Nicole (28m 57s):
Right. That is such a good tip too that you don’t have to go through and fix all of them. You can just type all of the changes you want it to make into your next prompt and then we’ll regenerate the transcription with all the changes, which is time saving as well.

Diana (29m 13s):
Yeah, it’s way easier to do that than to copy it with all the mistakes and then to have to painstakingly go through and fix them.

Nicole (29m 19s):
Agreed. Well I wanted to mention a new program that I’m going to have to write a blog post about, but Mark Humphreys, who’s a history professor in Canada, he has created an application that will transcribe historical documents and it’s called Transcription Pearl and it uses ChatGPT and Claude to transcribe the documents. So you have to get an API for Claude if you want to try this. He walks you through it in his recent blog post called Transcription Pearl for Non Coders and Next Steps. And his original post was only in the GitHub repository as like an application you had to run through Python and now he’s made it as an application you can execute on Windows PCs, not Macs yet.

Nicole (30m 11s):
But this is really exciting because it takes kind of these learnings and takeaways that we’ve had with how large language models can transcribe and makes it into an application that has basically his saved prompts for how to best get the things transcribed correctly and it puts things side by side for you. And he actually is not a coder, but he used ChatGPT and Claude to teach him how to write the script with Python. So it’s really cool and it’s exciting for us as Genealogists and historians to be able to have these tools at our disposal to make processing and transcribing all these historical documents so much easier and to find relevant information in them.

Nicole (30m 52s):
So it’s exciting.

Diana (30m 54s):
Oh it is so exciting. I think back to client projects where I would have to have a few hours saved to do the transcription of important documents and now you know, as a professional you could really get this, this tool down. Or even if you were doing this just for yourself, you know, nobody wants to spend hours and hours transcribing, but we need the data. We need more than just pulling names out of probate. We need to understand everything that’s in that document. So this idea of having large language models help us with the transcription is really kind of game changing for genealogy I think.

Nicole (31m 31s):
Right. It really is. I remember when Judy Russell wrote her blog post about FamilySearch’s full text search and I think it was called Game Changer, her blog post. But it is, I was just thinking about how one of the recent projects I did during the 14 day Research Like a Pro mini challenge was about the sister of our ancestor, Lucindrilla Keaton and Sally Keaton. She was mentioned in the, the probate records just twice and it had her husband’s name and it had where they lived, but my transcription wasn’t very good and so I didn’t notice the clue about where they lived until I went back and really perfected the transcription ’cause I had just been focusing on Lucinda and Lucindrilla those names and I had found my ancestor and used that to prove her father.

Nicole (32m 17s):
And so I went, I, when I went back to find more about her sister and what happened to her and where she moved to and her children, then I really had to scour that transcription and there were really only like two or three references that were helpful. But yeah, we can’t just gloss over and look at only one or two of the pages in a 56 page probate record. ’cause there are clues that can help solve our genealogy questions in all these little scraps of paper. And one of them was like a receipt that mentioned like heirs that are living outside of the state. So it was just a little piece of paper that I didn’t know was gonna be so helpful.

Diana (32m 54s):
Oh, that is a great example. Yeah, it’s gonna make us better. We will definitely be better Genealogists because of AI. Absolutely. Well thanks everyone for listening. It’s been so fun to talk about artificial intelligence and we will continue to learn and to share what we learn and we really hope that all of you will give it a try. Even if it you’ve been kinda holding back, you can try these free models and put in a letter, a deed, a will, something you’ve already transcribed, put it into one of the models and see how it works. Then compare it and we’d love to know how that worked for you. So thanks everyone and we’ll talk to you next time.

Nicole (33m 36s):
Okay, 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 at DA on amazon.com and other books, sellers. You can also register for our online courses or study groups of the same names. Learn more at FamilyLocket dot com slash 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 dot com slash 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.