Friday's Photo: The Thompson Brothers of Erath County An AI Fail

After comparing my father's photos with those of his brother's in Tuesday’s post, Buster vs. Dad – An AI Perspective, I’ve found myself on a new mission—revisiting my collection of unidentified photos with help from AI.

This photo of two young men was shared by a cousin, who identified them as Floyd William Thompson (1889-1968) and one of his brothers. 

Floyd had five brothers:

• Aaron Whitney Thompson (1886-1964)
• Claud Clarence Thompson (1891-?)
• Edgar Richard Thompson (1896-1983)
• Walter Alexander Chisum (1876-1816) [half-brother]
• Philip Louis Chisum (1881-1962) [half-brother]

I uploaded the photo to OpenAI’s ChatGPT along with at least one confirmed image of each brother—except Claud. I’ve never come across a photo of Claud Thompson. I also provided their names and birth years. I've learned that AI performs better when I include some background information. 

AI Analysis

I first used ChatGPT 4o to upload photos and ask questions. I really liked this table created by AI. It made me consider details I might have overlooked. 

This model confidently identified Floyd as the standing figure, citing specific facial features and stance as evidence.

However, this AI automatically assumed the standing boy was older, despite the possibility to the contrary. It also made questionable observations, like claiming the seated boy had "less prominent ears"—which contradicts what's visible in the photograph. 

When I challenged certain conclusions, the AI adjusted its analysis to accommodate my input rather than defending its position with evidence. For example: 

• With the initial comparison, AI wrote, "The seated boy is likely another Thompson brother, but doesn't clearly resemble Edd, nor Floyd in his later years."

• When discussing birth years, AI responded, "Based on appearance, Claud Clarence Thompson (b. 1891) is a more likely candidate than Aaron."

• When reminded that no photos of Claud existed for comparison, the AI response was "The seated boy now appears to be Edgar Richard Thompson rather than Claud."

This inconsistency prompted me to seek a second opinion from a custom GPT specializing in historical photo analysis, Mark Thompson’s Historical Photo Analyzer, available through OpenAI’s ChatGPT platform. This consultation provided excellent historical context and thorough comparative analysis, but yielded the exact opposite conclusion, identifying Floyd as the seated brother.

Conclusions

Given the conflicting results, I decided it was best to conclude this AI comparison. Moving forward, I'll continue using AI as a supplemental tool. In my last post, Buster vs. Dad – An AI Perspective, AI did an excellent job pointing out similar features that could be easily observed. Not so much with this project. What I learned is that AI can help me see things I might overlook, but it is not a replacement for my intuition, family knowledge, or careful comparisons. 

For now, I will share this post with my family and on the Erath County Genealogical Society Facebook page, in the hope of gaining more insight from others who might recognize the faces or have additional knowledge about this family. 

If you'd like to learn more about the families I research, click here to follow my Facebook page, where you can view each post and other genealogical discoveries. 

Diana

© 2025

ChatGPT, "Comparison and Identification of Floyd Thompson in Historical Photo; Analysis of Seated vs. Standing Boy," OpenAI (https://chat.openai.com : accessed 25 April 2025).

Mark Thompson’s Historical Photo Analyzer (GPT : ChatGPT, OpenAI), "Historical photo analysis customized for detailed image study and genealogy insights," analysis provided to user DB Quinn, 25 April 2025; privately held by DB Quinn, VA.

Thompson Family Photographs, digital images, privately held by DB Quinn, Virginia Beach, VA, 2025; scanned from original photographs shared with DB Quinn in 2001 by Juanita Gleghorn, Tom Thompson, and Nadine Waugh during a visit to Texas.

Buster vs. Dad – An AI Perspective

Buster and Dad in their early teens. 

I decided to use AI to help identify some of my unknown family photos. The first few results weren’t great—a boy was labeled as a girl, another as a man, a scratch was mistaken for a clothesline, and an attempt to create a collage ended with images that didn’t look like my family at all. But when I started using OpenAI’s ChatGPT to compare photos side by side, I finally began to get useful results.

Buster or Dad?

I have several children in photos that have always been in question—were they my father or his brother? My cousin, who originally had the photo collection, labeled all of them as “Buster.” Buster was my dad’s older brother, Hairston Albritton Bryan, born in 1906. My father, Whit Criswell Bryan, was born in 1920. While I’d already guessed that some of the photos were of Dad, there was always a bit of doubt. AI helped confirm what I’d suspected.

I started by uploading one of the questionable photos and simply asked, “Can you date this?”

AI responded with an analysis of the child’s clothing, shoes, and the overall photographic style—pointing out the sepia tone and the formal studio setup. Based on these features, the photo was dated to somewhere between 1910 and 1920.

I followed up with: “This has been identified as my Uncle Buster, who was born in 1906, but could it actually be my father, born in 1920?”

The response was clear:

“While it could be your father, the styling strongly favors your Uncle Buster (born 1906). Everything about the child’s appearance, the clothing, and the studio portrait setup points to the 1910–1912 range.”

I decided to go a step further. I uploaded another photo and asked, “Could this be the same child?”

AI responded by analyzing the clothing, backdrop, and photo mount, estimating a time frame of circa 1910-1912, which lines up closely with Buster’s childhood years.

It noted that both children had oval-shaped faces, strong jawlines, and similar eye placement and spacing. The resemblance was plausible, especially when considering age progression, fashion trends, and the photographic style. If the child in both photos was Buster, the timeline worked.

Next, I shared teen photos of Buster to see if there was a visible connection. AI noted strong continuity in features—especially eye shape, nose, and jawline—linking the child in the earlier photos to the teen in the later ones.

I added this photo and asked, “What do you think of this photo of my father—Buster’s younger brother, born in 1920? My cousin said it was Buster, but I know it’s Dad.” 

AI responded with, "The boy is dressed in a classic sailor suit with a wide collar and checkered tie, a style popular from 1915 to the early 1930s. His dark stockings and leather shoes complete the look. His straight, bowl-style haircut with a trimmed fringe was typical for boys in the early 1920s, especially before school age. Based on these details, the image likely dates to around 1923–1925, which fits perfectly for your father, born in 1920."

So I shared three known childhood photos of my father, and AI came back with, "These known photos of your father clearly show a consistent progression from babyhood to adolescence, with strong physical matches to the sailor-suit portrait. Between the features and the timeline, it’s very safe to say the child in the sailor suit is your father—not Buster."

I shared three more photos, thought to be of Dad or Buster, and received the expected results.  

Dad - Whit Criswell Bryan

Buster's Baby Photo
Dad on a Donkey

Comparing Tools and Doing My Own Research

All of the photo analysis and comparisons in this article were generated using OpenAI’s ChatGPT. I also tried Anthropic’s Claude. Claude was more cautious. It provided similar comparisons of facial features and timelines, but was less willing to commit—recommending a second look and suggesting I talk to family members. 

And while ChatGPT provided useful insights, I didn’t accept everything without question. I did my own research to confirm the clothing and hairstyle information, especially since sources weren’t cited. After conducting my own research, I generally agreed with ChatGPT’s timelines for hairstyles and clothing; however, the sources I found indicated that these trends didn’t follow strict cutoffs—there was more overlap and variation than the AI responses suggested.

Final Thoughts

This process was a good reminder of the value of revisiting old photos with new tools and a fresh perspective. AI won’t solve every mystery, but in this case, it helped me ask better questions and, more importantly, confirm what I’d long suspected.

If you want to know more about the families I research, click here to like my Facebook page, where you will see each post and other genealogical finds. 

Diana

© 2025

Sources

Edward Zapletal, Publisher and Editor, More Dating Old Photographs 1840-1929 (Toronto: Moorshead Publishing, 2011).

Historic Boys' Clothing, "United States Sailor Suits: The 1920s," web page, Historic Boys' Clothing (https://histclo.com/style/suit/sailor/sailorus20.html : accessed 22 April 2025).

Jo B. Paoletti, “Clothing and Gender in America: Children’s Fashions, 1890-1920.” Signs 13, no. 1 (1987): 136–43 (http://www.jstor.org/stable/3174031: accessed 14 April 2025). 

Maryanne Dolan, Vintage Clothing 1880-1980 (Alabama: Books Americana Inc., 1995).  

Special Collections at the Richardson-Sloane Special Collections Center, "Saluting the Sailor Suit," blog entry, 2 July 2022, Primary Selections from Special Collections, blog (https://blogs.davenportlibrary.com/sc/2022/07/02/saluting-the-sailor-suit/ : accessed 22 April 2025).

TPR, "The Bowl Haircut," blog entry, 2 January 2020, The Past Recedes, blog (https://tpr76797.wordpress.com/2020/01/02/the-bowl-haircut/ : accessed 22 April 2025).

From Loyalists to Free People of Color: Stories You Don’t Want to Miss!

You Don’t Have to Live in Virginia Beach to Attend Genealogy Day at the Beach!

Every spring, I look forward to highlighting this event hosted by one of my favorite groups, the Virginia Beach Genealogical Society. Genealogy Day at the Beach is right around the corner—Saturday, April 19, 2025—and this year’s event is shaping up to be one of our best yet. Hosted by the Virginia Beach Genealogical Society and the Virginia Beach Public Library, it’s a day filled with inspiration, learning, and community connections.

But here’s the best part: you don’t have to be in Virginia Beach to join us—and there’s no cost! This event is hybrid, meaning all speaker presentations are available both in person and via Zoom. No matter where you are, you can tune in and enjoy a remarkable lineup of genealogical speakers.

Registering for Zoom access is free, and you’ll find the Zoom links posted on the Virginia Beach Genealogical Society website on the day of the event.

So spread the word! Invite your genealogy friends and make a day of it. You don’t have to live in Virginia Beach to enjoy this incredible lineup of speakers. But if you're local, we would love to see you in person! There's nothing quite like sharing the day with a friendly, enthusiastic group of people who are just as passionate about genealogy as you are. Whether you're into DNA, colonial history, African American research, or family mysteries, you'll find someone here who shares your interests.

Come learn, connect, and get inspired! 

If you want to know more about the families I research, click here to like my Facebook page, where you will see each post and other genealogical finds. 

Diana

© 2025 

Massachusetts Roots, Vermont Branches: The Children of Benjamin Reed and Huldah Pratt

Benjamin Reed and Huldah Pratt were both born in Plymouth County, Massachusetts. At the time of their marriage on 3 December 1778 in Abington, Benjamin was residing in Cummington, Hampshire County, while Huldah was living in Bridgewater in Plymouth County.

According to this list, found at The Reed Genealogy, Descendants of William Reade of Weymouth, Massachusetts From 1635 to 1902 by John Ludovicus Reed, Benjamin and Huldah had at least ten children between 1779 and 1802. Huldah was 19 when their first child was born and 42 at the birth of their last. My fourth great-grandmother, Nancy Reed, was their ninth child.

Hampshire County land records indicate that Benjamin and Huldah lived in Massachusetts until at least 1786, though brief absences during that period can’t be ruled out. This suggests that their children—Benjamin Junior, John, Huldah, and Cyrus—were born there. By May 1788, Benjamin Senior was named as a Bennington County, Vermont resident in a land purchase record, suggesting that son David, born in November 1788, was likely born in Vermont. As Benjamin and Huldah remained in Vermont, their children—Betsey, Calvin, Luther, Nancy, and Almira—are also presumed to have been born there.

This list of the Reed children is believed to have been given to John Ludovicus Reed for his book in 1899 by Mary E. Reed Roane, Luther Reed's daughter and Michael Roane's wife. According to John L. Reed, Mrs. Roane had a copy of a family record found in the Bible once owned by Benjamin and Huldah. I'm eager to connect with descendants of this family to learn more!

It was clear that someone was trying to decipher dates from a handwritten list when
naming these children. The dates circled in the list below were questioned in
the Reed book, possibly due to difficulty in reading them. I would love to
see the original document!

If you want to know more about the families I research, click here to like my Facebook page, where you will see each post and other genealogical finds. 

Diana

© 2025 

Doolittle, Amos. "State of Vermont Drawn and Engraved from the Latest Authorities," 1795, digital image, David Rumsey Map Collection (https://www.davidrumsey.com/luna/servlet/detail/RUMSEY~8~1~34674~1180289 : accessed 20 March 2025).

Find a Grave, database and images (https://www.findagrave.com/memorial/27499358/huldah-read: accessed March 27, 2025), memorial page for Huldah Pratt Read (1760–17 May 1842), Find a Grave Memorial ID 27499358, citing East Putney Cemetery, East Putney, Windham County, Vermont, USA; Maintained by Cynthia Kaley (contributor 51056978).

Hampshire, Deeds 1789-1795, Book 3:550, Reed to Coock, images, FamilySearch (https://www.familysearch.org/ark:/61903/3:1:3QS7-99Z8-29DJ?view=fullText : Mar 28, 2025), image 286 of 597; Massachusetts. County Court (Hampshire County).

John Ludovicus Reed, The Reed Genealogy : Descendants of William Reade of Weymouth, Massachusetts from 1635-1902 Vol. I  (no publisher named, 1901).

Scott, Joseph , "Massachusetts," 1795, digital image, Library of Congress (https://www.loc.gov/item/2018590108/ : accessed 20 March 2025). 

“U.S., Revolutionary War Pension and Bounty-Land Warrant Application Files, 1800-1900,” database and images, Ancestry.com (https://www.ancestry.com/discoveryui-content/view/49769:1995 : accessed 25 June 2022), images 266 to 320 for Benjamin Reed; citing "Revolutionary War Pension and Bounty-Land Warrant Application Files" (NARA microfilm publication M804, 2,670 rolls). Records of the Department of Veterans Affairs, Record Group 15. National Archives, Washington, D.C.

"Woodford, Bennington, Vermont, United States records," images, FamilySearch

(https://www.familysearch.org/ark:/61903/3:1:3Q9M-CSVG-ZQ9S-4?view=fullText : accessed 2 Apr 2025), image 26 of 208.