Do you need a simpler way to transcribe and abstract deeds and other land records?
As someone who deals mostly with poor and hard to find Southeastern US ancestors, I certainly do. I don’t think I’ve solved a single Southern US brick wall without the help of land records, even if the research subject(s) never owned land.
But transcribing and abstracting deeds can also be incredibly time consuming.
About a year ago, I started noodling with ChatGPT, Claude, and other large language models to help me with common genealogy tasks like transcribe, organizing large amounts of information, and performing basic research report tasks like listing out negative searches for client reports.
I noticed that, for deeds and plats, I was asking AI to do the same things over and over again:
- Transcribe the record
- Comment on the quality of the record (ink blots, tears, light writing, poor handwriting, etc.)
- Abstract the record (especially highlighting witnesses, neighbors and geographic features)
- Write a short narrative summarizing the record for my research log
It finally occurred to me that I could save time and energy by creating a custom GPT (generative pre-trained transformer) to do all these things for me without me having to prompt AI to follow these same steps over and over every single time.
And, after much trial and error, I think I’ve created just the AI tool I need to make deed (and other land record) transcription quicker and simpler.
So now I’m introducing “Jenn’s Deed Transcriber and Abstractor“!
I figure folks want the good stuff up top, so I’m including instructors on how to get started first. But if you’re willing, please read on to help me test this GPT and ensure it’s as useful as it can be for current and future genealogists. Thank you, friends!
How to use Jenn’s Deed Transcriber and Abstractor on ChatGPT

- Visit Jenn’s Deed Transcriber and Abstractor here
- If for some reason that link doesn’t work, go to the ChatGPT library and search for it by name
- Upload an image file (.png or .jpg) of the deed, plat, survey or other land record you’d like to transcribe and abstract. To reduce “noise,” I recommend only uploading the actual deed you want ChatGPT to process, and not entire pages of a deed book.
- Note: After testing, I’ve found this tool doesn’t read pdfs well. If your record is in pdf format, I recommend simply taking a screenshot of it.
- Follow the prompts. You’ll be asked to give the names of the grantor and grantee, and the location of the deed. You can skip this step, but I’ve had ChatGPT hallucinate some truly off-the-wall county names, so it’s easier if you help it from the get go. Remember, you’re not trying to trick ChatGPT – you’re trying to use it as your eager assistant!
- Read over the output carefully. Pay special attention to names of people, counties and localities, and roads, rivers, creeks, etc. ChatGPT is great at boilerplate, but can be hit-or-miss on transcribing proper names.
- From here, you can correct anything that looks incorrect.
- Record the output. Once corrected, this GPT gives you an output that you should be able to copy and paste into Microsoft Word, Google Docs, your research log, a note taking app, or another place of your choice.
- Tip: If copy and paste isn’t working correctly, ask it to create you an html file, and then open that file in your browser (Google Chrome, Firefox, Bing, etc.) You should then be able to copy and paste it, or use your browser’s “Print” function and then “Print to PDF.”
- Let your friendly neighborhood GPT creator know what issues and errors you ran into! I’ve tested this fairly thoroughly, but of course I can’t think of every use case or hit every error. Please let me know how this works for you so I can attempt to improve on it!
Jenn’s Deed Transcriber and Abstractor Example
Trials and Tribulations while Creating this GPT
It took a lot of tinkering to get this one into a fit state that I’m ready to share it with the world. And it still has three major issues that I’ve attempted to work around:
- Limited PDF input – When I uploaded PDFs the GPT would literally transcribe other documents I’d previously uploaded to ChatGPT. Not even to this GPT, but ChatGPT in general. This is a serious error. If I had been dealing with lots of deeds and mortgages from the same person, I may not have even caught this error. So, I need to do more research before I allow this GPT to accept PDFs. Fortunately, I was able to have the GPT shut down any PDFs that people try to upload. Sorry for any inconvenience!
- Hallucinated names and places – Normally when I have ChatGPT transcribe something, I tell the tool a little bit about it. (I.e. “Act as a professional genealogist. Please transcribe this 1846 will of Rachel West from Union District, South Carolina.”) But, with the custom GPT, if I didn’t do that, it would make up county names out of whole cloth. Further, I’ve instructed it to place anything it doesn’t recognize in brackets, but it has resisted doing that. Rather, it confidently tells me that my ancestor lived in Marlborough County, SC when they actually lived in Spartanburg, and that the grantor in the deed was from York County when he was actually from Pendleton District. Despite the fact that I have wrestled with this GPT to admit its shortcomings, no luck so far.
- No PDF output – Hope you don’t love PDFs, because this GPT sure doesn’t. I’ve actually had ChatGPT create PDFs in the past, but maybe I was working magic? Because while this GPT would pretend it had created a PDF for me, there was no actual output. So for now, I suggest people follow the steps above to copy and paste or “Print to PDF” their land record transcriptions and abstracts.
Help me test this GPT?
I’m sure there are other issues with this GPT that I haven’t run into yet. Please contact me if you run into a problem or you have ideas for a fix for any of the above issues!
2 thoughts on “Easily Transcribe & Abstract Deeds with New Custom GPT”
I’m excited to start testing this and I will let you know how it goes!
Thank you so much for checking it out, Matt! I’ve been finding and trying to squash little issues in there, so appreciate any insight or issues you run into!