Just Because AI Can Do It, Should I?
An Experiment in Renaming Genealogy Files with ChatGPT
Have you ever looked at your genealogy files and wished they were better organized? Have you ever considered renaming hundreds—or even thousands—of files so that you could tell what they contained without opening them?
I have!
Recently, a post in the Facebook group Genealogy and Artificial Intelligence caught my attention. The writer described using ChatGPT and Claude to process genealogy documents. Their AI agents could extract information from documents, rename the files, place them in appropriate folders, and create structured data files containing the extracted information.
That sounded intriguing! Could I use AI to improve the organization of my genealogy files?
I decided to experiment.
My Original Filing System
When I began researching my family history, genealogy was largely a paper-based activity. I traveled to libraries, archives, and courthouses, made photocopies, and organized those papers in folders.
For organizing my research, I adopted the Dollarhide numbering system. My documents were assigned identifiers based on the surname and location, followed by a document number.
For example, my Crawford research included documents with identifiers such as:
Crawford.KS.xxx
Crawford.KY.xxx
Crawford.IN.xxx
When I began scanning my paper documents, I used those identifiers as the digital filenames.
The system worked well for connecting a document to my research notes and source citations. However, there was one disadvantage: the filename didn't tell me what the document contained.
A file named Crawford.KY.123 might be a land deed, a marriage record, a census transcription, or a letter. To find out, I would need to open it or locate the corresponding citation.
My Filing System Evolved
Over time, I changed the way I named my digital exhibits.
At one point, I used a pattern that included the surname, given name, birth year, document year, and description.
However, I discovered that this approach made chronological sorting difficult.
Eventually, I changed my organization so that an individual's folder begins with the person's birth year, followed by their name. Within that folder, document filenames begin with the year of the document, followed by a description.
This allows me to sort the files chronologically.
My genealogy folder now contains an Exhibits folder organized by surname, with additional folders for individuals and general research by state. I also use Scrivener to organize my research notes, including multiple Crawford projects organized geographically.
While I've renamed many of my older exhibits, I still have a substantial number of files bearing their original Dollarhide identifiers.
Could ChatGPT help me rename them?
The First Experiment
I began by uploading a ZIP file containing several older Crawford PDF documents to ChatGPT.
I asked ChatGPT to examine the documents and recommend descriptive filenames based on their contents.
The goal was to put the document date first while retaining the original Dollarhide number at the end of the filename.
For example, instead of a filename such as:
Crawford-IN-098.pdf
I might have a name resembling:
1860-07-16-IN-Jay-Census-Wikle-Crawford-IN-098.pdf
This would provide a descriptive filename that could be sorted chronologically while preserving the original document identifier.
ChatGPT examined the files and created an Excel spreadsheet containing the original filenames, suggested new filenames, document descriptions, and review notes.
Some of the suggestions were useful. Others required closer examination, particularly when the PDF contained handwritten information or multiple dates.
That led to the next step in the experiment.
Adding an Approval Process
I wondered whether ChatGPT could go beyond suggesting filenames.
Could I review the recommendations in a spreadsheet, approve or edit them, and then have ChatGPT automatically rename the files?
The answer was yes.
For the second experiment, I uploaded a ZIP file containing 31 Crawford PDF documents.
ChatGPT created an Excel spreadsheet with proposed filenames and an approval column. I could approve a suggested name, edit it, or skip the document.
After reviewing the spreadsheet, I uploaded it to ChatGPT.
ChatGPT then created a new ZIP file containing copies of the PDFs with the approved filenames. It also included a cross-reference between the original and new filenames.
Technically, the process worked!
I had demonstrated that ChatGPT could examine genealogy documents, suggest descriptive filenames, accept my corrections, and automate the renaming process.
And my original documents remained unchanged.
But Was It Worth the Time?
That was the question I began asking as I worked through the spreadsheet.
Even though ChatGPT had suggested filenames, I still needed to open the documents to verify those suggestions.
Was the date correct? Had ChatGPT identified the right person? Was the document a deed, a probate record, or something else? Did the proposed name accurately describe the contents?
Those questions were particularly important for my Crawford research, where I am trying to distinguish among several men named James Crawford who lived during approximately the same period.
I couldn't simply accept the AI-generated names without checking them.
And that meant I was still spending time opening and examining the documents.
Yes, ChatGPT could automate the actual renaming. But that was never the most difficult part of the task.
The time-consuming part was deciding what each file should be named.
AI had not eliminated that work. It had simply changed how I performed it.
Then I Remembered My Citations
As I considered whether to continue renaming files, I realized something important.
Those old Dollarhide identifiers weren't just arbitrary filenames.
When I used The Master Genealogist (TMG), I incorporated those document numbers into my source citations.
Later, when I migrated my genealogy database to RootsMagic, that citation information came along.
Thus, the original document numbers still serve a purpose.
If I encounter a citation referencing Crawford.KY.123, that identifier helps me locate the corresponding document.
In other words, I already have a system connecting my genealogy database to my exhibits.
The filenames may not be descriptive, but they are not meaningless.
And because I have retained the original identifiers in many of the filenames I've already renamed, I have preserved those connections.
Automation Doesn't Always Mean Efficiency
This experiment reminded me of something I learned during my years as a school technology coordinator.
Technology can automate a process without necessarily making the overall task more efficient.
Sometimes the software saves considerable time. Other times, the time required to prepare, verify, and correct the results outweighs the benefit.
That's what I encountered with this experiment.
ChatGPT successfully performed the mechanical work of renaming files. But I still had to make the research decisions.
And with thousands of genealogy exhibits, that could become a substantial investment of time.
Would the result justify that investment?
For me, at least right now, the answer is no.
Perhaps I'm Asking AI to Solve the Wrong Problem
As I discussed the results with ChatGPT, another possibility emerged.
Instead of renaming the documents, what if I used AI to create a searchable inventory of their contents?
Such an inventory might identify the document type, date, people mentioned, and location while preserving the original filename.
That could be especially useful for my Crawford research.
For example, I might be able to locate every document mentioning James Crawford and William Sellers or identify records connecting the Crawford, Douglas, Cave, and Dooley families.
An inventory could potentially make the information in my documents easier to discover without changing the filenames.
There's also the possibility of using information already recorded in my RootsMagic citations to help create that inventory.
After all, why ask AI to extract information from a document if I've already recorded that information in my genealogy database?
These possibilities are worth exploring, but they are ideas for future experiments rather than solutions I've already tested.
My Conclusion
I'm glad I conducted this experiment.
I learned that ChatGPT can examine PDF documents, recommend descriptive filenames, create an Excel approval process, and produce a ZIP file containing renamed copies of the documents.
Those are useful capabilities.
However, I've also concluded that systematically renaming my older Dollarhide-numbered documents is probably not worth the time it would require.
My existing citations preserve the connection between those identifiers and the original documents. Renaming everything would improve the appearance and browsability of my folders, but it wouldn't necessarily help me answer my genealogical questions.
I'd rather spend my time researching the relationships among my Crawford ancestors than renaming hundreds of documents I've already analyzed.
Just because AI can do something doesn't mean I need to use it for that task.
Perhaps the better question isn't, "What can AI automate?"
Perhaps it is, "How can AI help me make better use of the research I've already done?"
That's the question I plan to explore next.
What about you? Have you experimented with AI to organize your genealogy files? Did the automation actually save you time, or did you discover that reviewing the results required almost as much work as doing the task yourself?
I'd be interested in hearing about your experiences.
Comments
Post a Comment