Unlocking the Potential of AI in German Genealogy at IGGP 2025
- Gail S. Blankenau
- Jun 30
- 4 min read
Following the AI-focused presentation at the International German Genealogy Partnership (IGGP) conference, it’s clear that there is growing curiosity—and even more questions—about how artificial intelligence can assist genealogists, particularly those researching German ancestors. This post expands on the session, offers follow-up resources, and encourages thoughtful exploration of AI tools—along with a healthy dose of caution.
Audience Snapshot: New to AI, But Eager to Learn
Thanks to the Whova app, presenters had a sense of the audience a few days ahead of time. An overwhelming 97% of attendees reported having little or no experience using AI in their research. With that in mind, the session focused on beginner-friendly platforms and demonstrations of free or widely accessible tools.
Language translation emerged as the leading area of interest. While photo analysis and generative design tools were also covered, it became clear that help with deciphering handwritten or printed German documents is the most pressing need for many.
Platforms Demonstrated
The session introduced attendees to several tools with potential value for genealogy:
ChatGPT (OpenAI)
Claude (Anthropic)
DALL·E (image generation)
Transkribus (OCR and handwriting transcription)
Adobe Express (AI-powered design and slide generation)
MyHeritage (photo tools and document matching) not a free platform
FamilySearch (OCR with full-text (handwritten) search functionality)
Google’s Search AI with a summary at the top
Given the audience, most of the demonstrations used free versions of these platforms, with the understanding that frequent users may want to explore paid upgrades in the future.
Documents tested included German-language birth, marriage, and death records, a personal letter, and a printed reference book excerpt. Each platform had strengths and limitations—and that variability reminded everyone that no one-size-fits-all AI solution exists.
Key Observations from the Trials
Claude by Anthropic performed well on printed 19th-century Fraktur typefaces.
Transkribus emerged as the strongest tool for handwritten German, including a surprisingly good (but not error-free) performance on a letter from the early 1900s.
Nonetheless, even the best tools made errors—especially with names and place names, which are vital in genealogical research.
These findings reinforce the recommendation: AI can serve as a helpful assistant, but it’s not a substitute for human verification—especially for those with limited German language skills.
A Shout-Out to Audience Insights
Because the presentation went long, some of the Q and A was short so all participants could have a chance to ask questions. However, there were thoughtful contributions from attendees that helped round out the discussion. One participant reminded the group that many specialized GPTs exist under the ChatGPT umbrella—something that would’ve made a great addition to the presentation slides. Her comment sparked a valuable realization: 50 minutes isn’t nearly enough time to cover the AI landscape fully, especially as new tools appear almost daily.
The same attendee pointed out Steve Little’s GPT for photograph analysis, which inspired a deeper dive (see below) after the session. One of the photo analysis trials involved dating a carte de visite from Germany. While ChatGPT and MyHeritage provided reasonably good ranges (once prompted to focus on sleeve details), Steve’s Photograph Analysis GPT offered an in-depth, accurate assessment—right down to correctly identifying the photographer’s name on the card, something no other tool managed.
Below is just the beginning and the final conclusion of a much longer and in-depth report generated by Steve’s Photograph Analysis GPT.
Explore Steve’s GPT here: Steve’s Photograph Analysis
The last point that one of the attendees made after people were leaving for lunch:
Prompt the platform to generate ideas for better prompts!
And a huge thank-you goes out to those who raised these points. They added nuance and depth that benefited everyone in the room.
Bonus Tools and Takeaways
Also included in the demo were translation comparisons between Google Translate and DeepL, using the same German church record. DeepL generally produced smoother and more idiomatic English. German documents also contain many archaic terms that neither translation tool will translate without us training the tool. Any tool can struggle with uncommon place names or Latinized terminology in older records.
If you missed the AI session by Katherine Schober, a professional translator who explored AI in the context of German translation, her perspective offers a rich complement. The two sessions intentionally diverged in focus—one on translating documents, the other on broader AI applications in genealogy—to provide attendees with distinct entry points.
Final Thoughts: Use AI—But Use with Caution
A final screenshot shared during the session showed a transcription of a 1656 death record from Bayreuth using Transkribus. Run after the handout deadline, this trial yielded one of the best results of all—an encouraging sign that tools like this are improving fast.
Still, the bottom line remains: AI is a promising helper, not a final authority. Especially when working in languages that aren’t your own, results should always be verified by someone with the necessary expertise.
Resources to Explore
Directory of Custom GPTs: https://chatgpt.com/gpts
A Taste of Steve Little's Work in AI & Genealogy:
Thank you to everyone who attended, asked questions, and shared tips. The thoughtful engagement during and after the session is what makes conferences like IGGP so rewarding. As tools evolve, so too will the ways genealogists can explore the past.
Comments