A Look Behind the Scenes
If you’ve been following along on this blog, you know that I regularly use artificial intelligence tools as part of my genealogy workflow. Today’s post pulls back the curtain a bit to show how I used ChatGPT to analyze a set of Augusta County, Virginia records and develop a genealogical proof argument.
Starting with the Records
Like many of my research projects, this one began with a document—or rather, a set of documents. Two deeds from 1798 identified the heirs of Samuel Craig. At first glance, the names seemed straightforward, but as is often the case in early records, the relationships were not explicitly stated.
Adding to the mix were two marriage entries from Chronicles of the Scotch-Irish Settlement in Virginia. These entries provided critical clues, but they also introduced a potential point of confusion involving multiple Crawford men.
Step One: Transcription and Extraction
The first step in my process is always to understand the document. I transcribed the deeds and then asked ChatGPT to extract the family information. This is something I’ve come to rely on—having the AI pull out names, relationships, and patterns helps me see the structure of the record more clearly.
However, I never stop there. Extraction is just the beginning.
Step Two: Correlation
Once I had the extracted information, I introduced the marriage records. This is where the real work began. By asking ChatGPT to consider both the deeds and the marriages together, I was able to test assumptions and identify conflicts.
For example, the deeds included both John Crawford and Elizabeth Crawford. It would have been easy to assume they were husband and wife. But the marriage record clearly showed that Elizabeth Craig married Samuel Crawford. That single piece of evidence changed the interpretation of the entire document set.
Step Three: Asking Better Questions
One of the biggest lessons I’ve learned in using AI is that the quality of the output depends on the quality of the questions. Rather than asking ChatGPT to “tell me the answer,” I asked it to:
- Reconstruct the family
- Identify inconsistencies
- Suggest alternative interpretations
This back-and-forth process feels very much like working with a research partner. Sometimes the AI suggests something I hadn’t considered. Other times, I recognize that it has made an assumption that needs to be corrected.
Step Four: Building the Proof Argument
Once the relationships became clearer, I asked ChatGPT to draft a proof argument. Because I had already guided the analysis, the resulting narrative reflected the evidence I had assembled.
Even so, I reviewed and edited the argument carefully. AI can help organize and articulate a conclusion, but it is still my responsibility to ensure that the reasoning is sound and the evidence is accurately represented.
What AI Did—and Did Not Do
This project is a good example of what AI does well and where human judgment is still essential.
AI helped me:
- Extract information from complex legal language
- Organize multiple records into a coherent structure
- Draft a clear, readable proof argument
But I still had to:
- Provide the correct records
- Recognize when assumptions were being made
- Evaluate conflicting evidence
- Make the final genealogical conclusions
Why This Matters
Genealogy is not just about collecting names—it’s about building accurate, well-supported conclusions. Tools like ChatGPT can make the process more efficient, but they don’t replace the need for careful analysis.
For me, AI has become part of an ongoing workflow. It helps me move more quickly from document to interpretation, and from interpretation to narrative. But every conclusion still rests on the records themselves.
Final Thoughts
If you’re hesitant about using AI in your research, I understand. I was cautious at first as well. But used thoughtfully, it can be a valuable tool—especially when working with complex records like these Augusta County deeds.
The key is to treat AI as an assistant, not an authority.
And as always in genealogy: follow the evidence.