While FamilySearch’s experimental lab showcases how AI can transcribe cursive handwriting, I had never tried it myself. Since I have several handwritten letters I’d like to feature in blog posts, I decided to test ChatGPT’s transcription abilities.
I started by asking whether it could transcribe cursive. It responded:
Yes, I can transcribe cursive handwriting from images, as long as the writing is reasonably legible. Please upload the image you’d like transcribed, and I’ll do my best to provide an accurate transcription.
Next, I uploaded the image of the following letter and asked for a transcription.
8-14-83
Page 1Briles
(Mrs. Vernie)
Yeleta Buchanan Bradshaw
2204 Rolls Court off. Road
Ft. Worth TX 76114Briles Line on?
Dear Marcia:
Was so happy to hear from you. Out of my 18–20 names in the Gen. Helper I received about 20 letters. Yours was the only Briles. My husband’s line goes like this:
Young Dr. Henry Nease Sr. b. 22 Dec 1805 Randolph Co NC
d. 6-9-1883 Ellis Co TX m. 1 Mar 1828 NCBriles, Mary b. 20 Feb 1806 NC
d. 3 Aug 1884 Ellis Co TX – Italy Cemeteryall born
Children: N(f), m(m), also Arnold
Valentine, John, Jacob *Henry Nease Jr.
(on 1850 Census in Italy, TX, Ellis Co. 1859)Henry Nease Young Jr. b. 4-1-1845 Randolph Co NC
d. 3 July 1913 Jean, TX Young Co m. 28 June 1885Paschal, Virginia Gertrude “Gertie” Aug 1, 1867 – TX
d. 15 Mar 1953 Ft. Worth, Tarrant Co TXChildren: Henry Clyde, Carrie Maude, Mozelle,
John Marvin, Valentine Wallace, Ethel,
Ladora, Vera Gertrude *Katie Bell
Willie Pierce, Oliver NewmanYoung, Katie Bell b. 1900 Ellis Co TX
m. (1st) John Henry Bradshaw b. 12-12-1903
d. 19 Mar 1960
So, how well did ChatGPT do? Below is the transcription it produced, with any errors struck through and the corrected text following in italics. I initially attempted to use bold for the corrections, but since some of the original transcription already included bold text, that approach proved a bit confusing. After comparing the original letter to the transcription closely, I realized that ChatGPT interpreted the underlined portions of the letter as bold text.
8-14-83
Page 1Briles
(Mrs. Vernie)
Yeleta Buchanan Bradshaw
2204 Rolls Court off. Road
Ft. Worth TX 76114Briles
Line on?Letter.017Dear Marcia:
Was so happy to hear from you. Out of my 18–20 names in the Gen. Helper I received about 20 letters. Yours was the only Briles. My husband’s line goes like this:
Young
Dr.Sr. Henry NeaseSr.b. 22 Dec 1805 Randolph Co NC
d.6-9-18836-1-1883 Ellis Co TX m. 1 Mar 1828 NCBriles, Mary b. 20 Feb 1806 NC
d. 3 Aug18841883 Ellis Co TX – Italy Cemeteryall born NC
Children: N(f), m(m),alsoClyde Arnold
Valentine, John, Jacob*& Henry Nease Jr.
(on 1850 Census in Italy, TX, Ellis Co. 1859) – no parenthesis around this statement in originalHenry Nease Young Jr. b. 4-1-
18451848 Randolph Co NC
d. 3 July 1913 Jean, TX Young Co m. 28 June 1885Paschal, Virginia Gertrude “Gertie” Aug 1, 1867 –
TXARK
d. 15 Mar 1953 Ft. Worth, Tarrant Co TXNote in column next to Paschal, Virginia that was not transcribed: mother a Newman
Children: Henry Clyde, Carrie Maude,
Mozelle,Mary Ella (Mae)
John Marvin, Valentine Wallace, Ethel,
Ladora, Vera Gertrude *Katie Bell
Willie Pierce, Oliver Newman – (Newman transcribed in bold when original was not bold)Young, Katie Bell b. 1900 Ellis Co TX
m.(1st)#1John Henry Bradshaw b. 12-12-1903
d. 19 Mar 1960
While I still needed to carefully compare the original letter to the transcription for accuracy, using ChatGPT made the process significantly faster and more efficient. Thanks, ChatGPT!
Curious about what AI can do for your genealogy research? Try uploading a handwritten letter to ChatGPT or another transcription tool and see what it reveals!
Teresa, on Writing My Past (https://writingmypast.wordpress.com/2025/05/20/adventures-in-transcribing-a-will-with-ai/) just compared several AIs and their cursive transcription capabilities. All were somewhat lacking. I’ve sometimes found that copying and pasting the transcription into Word and then proofing/correcting is faster than doing it all myself.
I likely got lucky in that I was using more current cusive verses 1700 cursive and current language verses 1700 language. It’s unfortunate that her will isn’t one of the documents in the Experimental Labs on FamilySearch. While that AI does make errors in the transcription, it does know the language.
Marcia – you’re absolutely right – not only did the AI have to overcome the tight cursive of that era, but it also had to interpret the period language as well. Sadly, much of what I need hasn’t been included in full-text search at this point. There’s not much from Kent at all ☹️ Maybe one day…
I do use the Grab Text feature in Snagit to transcribe longer newspaper articles when I’m short for time. Mostly, though, for my style of learning and processing, I prefer to do the work myself as it helps things stick. AI is a great tool for specific tasks.
Comments are closed.