Exploring FamilySearch’s New Full-Text Search Tool & AI Transcription Comparison

(Note: This blog post was originally published on the blog, Genealogy with Dana Leeds: Creator of the Leeds Method.)

A Revolutionary Tool: FamilySearch “Full-Text Search”

During RootsTech, an exciting development was announced: the launch of FamilySearch Labs. Among these experimental tools is one described as “Find Results with Full-Text Search.” Although more databases will soon be added, currently this tool can search United States land and probate records from 1630 to 1975. What makes this tool a game-changer?

  • Full-text Searches: Discover records previously difficult to locate, including unindexed documents and those where ancestors are mentioned in less direct roles, such as witnesses or neighbors.
  • Dynamic Search Functions: Utilize quotation marks for exact searches, “+” for mandatory inclusion of specific words, and “*” as a wildcard for flexible searches. Filters for year, type, place, or collection further refine your search.
  • Watch and Learn: Maximize your search capabilities by watching the short, instructional video that guides users through optimizing their search strategy.

Discovering Ancestors with Enhanced Precision

To explore this “full-text search” tool, I focused on my 6th great-grandfather, David Correy (~1708-1787), of New London, Chester County, Pennsylvania. The search led me to records I had not found previously:

  • Listed as a Neighbor: A 1767 deed identified David Correy as a neighbor providing additional information about the location of his land and neighbors at a specific time.
  • David’s Will: David Correy’s will not only confirmed his approximate death date but also provided direct evidence of my 5th-great grandmother’s father as well as naming other family members.

Experiment with AI

I decided to do an AI experiment with David Correy’s will. My goal was to compare their accuracy to the original text as well as determining whether they could handle an entire handwritten page. The four contenders were:

  • Microsoft’s Copilot (powered by OpenAI’s GPT, but “weaker”)
  • Google’s Gemini Advanced
  • OpenAI’s ChatGPT 4
  • Anthropics (new) Claude 3 Opus

Evaluating AI Transcription Accuracy

The experiment revealed varied results. Copilot and Gemini struggled with both accuracy and handling longer text segments. ChatGPT performed admirably, with only a few mistakes, though it often corrected what it perceived as spelling errors. Claude emerged as the leader, offering the most accurate transcription and usually preserving the original spelling.

Copilot's Transcription of David Correy's Will
Copilot’s Transcription of David Correy’s Will
Gemini's Transcription of David Correy's Will
Gemini’s Transcription of David Correy’s Will

With ChatGPT and Claude as the leaders, I decided to test them with an entire page of the will. Both managed to transcribe the full page—a feat that AI has struggled with in the past. Claude, again, excelled at maintaining the original line breaks and was more consistent in preserving the document’s original misspellings.

Original vs. AI: A Comparison

Based on today’s spelling, the following words were some of the “misspelled” words in the original document: Newlondon (New London), perfict (perfect), helth (health), deth (death), folowing (following), satisfyed (satisfied), and princiepaly (principally). How did these two AIs fare?

ChatGPT's Transcription of David Correy's Will
ChatGPT’s Transcription of David Correy’s Will
Claude's Transcription of David Correy's Will
Claude’s Transcription of David Correy’s Will

Both ChatGPT and Claude corrected “perfict” to “perfect” and “deasently” to decently, other words were handled differently by the two AIs:

  • ChatGPT successfully kept “New london” as one word, but changed the spellings of the other words to match today’s spellings.
  • Claude managed to accurately transcribe words that were spelled incorrectly by today’s standards, but broke “Newlondon” into two words.

While neither AI perfectly preserved every original spelling, both performed impressively overall and both offer a valuable starting point for transcribing and understanding our ancestors’ written records. Claude, however, was the most precise.

Final Thoughts and a Look Ahead

FamilySearch’s full-text searching tool is an incredible benefit to genealogists. It opens up new possibilites for uncovering parts of our family histories that were previously difficult to access. I recommend trying it out to see what you can uncover about your ancestors!

As the field of AI continues to grow, new tools like Claude are showing us the future of technology’s role in our research. While ChatGPT has been a frontrunner, Claude is an amazing newer tool that is a definite asset. ChatGPT will likely release their next version soon.

Continuing the Journey with AI in Genealogy

If you are inspired by these advancements and want to explore the role of AI and genealogy further, please consider joining Dana’s course, “AI Explorations in Genealogy & Beyond: A Course of Discovery & Application.” This 8-week course is packed with:

  • Interactive learning sessions and live, recorded Zoom meetings
  • Small group activities that encourage a community learning experience
  • Weekly challenges to help you learn how to use AI in various ways

This course is perfect for those just beginning to integrate AI into genealogical research or those seeking to broaden existing skills. Click here to find out more and register today!

By integrating these innovative tools, we’re entering a new chapter in genealogy, making it easier than ever to access our past and bring the stories of our ancestors to light. Join us as we take on this exciting exploration.