Using AI to Analyze Tax Data in a Research Log

Tax research involves a lot of data – land, property, associates, amount of tax paid, and more. You may be working with a run of tax lists for several years and dealing with several individuals in those tax lists. Once you’ve carefully extracted the data, the next challenge is analyzing it.

I recently wrote about customizing an Airtable research log for tax research to match the headings in the Dallas County, Texas, rolls from 1847 to 1864. I extracted data for all the Weatherfords in the county for those years and ended up with 126 records/rows. Airtable made it easy for me to group by individual and sort in various ways. Still, I needed assistance getting the data into my research report on Henderson Weatherford. I wondered how well AI would handle the challenge and turned to Claude to find out. The AI tool did an admirable job using my data to create a table for my report in a Word document and then summarizing that data for the report analysis. Here is how I did it.

Creating a Table from a .CSV File

The first task was to get my Airtable tax list table into AI. This can be a little tricky in Airtable, so here are the steps.

1. Click on the down arrow next to “View” of the table. Unless you’ve duplicated and renamed the view and created other views, this will be titled “Grid View,” which is the default title. I have created two additional views, so I renamed “Grid View” to “Full View.”
2. Next, click on “download CSV.”
3. Select the computer file for the download. I put my Dallas County Tax List file in my Weatherford folder under  Texas > Dallas County > Tax Lists.

 

Upload the .CSV File to AI

I started a new chat in Claude and gave it the following prompt:

Using the .csv file titled Dallas County Tax Lists, provide a table showing the information chronologically.

In the past, I’ve specified the columns for the table, but since the data I provided was already divided into columns, I was curious to see how AI would proceed. Claude did a good job and accurately transferred all the data to a table but truncated it, stopping at 1852. I then prompted it to create a table for all the years, and it ended up dividing it into three portions. With the tables created, I was ready to add them to my research report in Word.

Using Writage to Paste the Table in Word

Claude provided the table in Markdown for copying and pasting elsewhere. ChatGPT defines Markdown as follows:

Markdown is a lightweight markup language used to format text. It’s popular for its simplicity and ease of use, allowing writers to create structured documents without the need for complex formatting tools. Markdown files use plain text, with special characters to indicate formatting (like headings, lists, or links), making them easy to read and edit. Common uses include writing documentation, notes, blog posts, and README files in software projects.

Pasting something like a table with Markdown formatting into your document can be a challenge. However, I had previously purchased Writage, a plugin for Microsoft Word that lets me save Markdown (.md) files directly into Word. After you add Writage to Word, it appears in the ribbon at the top of Word.

I copied the table from Claude and then used “Paste Markdown.” The table came in perfectly with the headings and all the data. I adjusted the columns and made the font smaller to fit on the page. There was so much data that the tables took up seven pages in the appendix of my report!

Analyzing the Tax Table with AI

Now came the fun part – asking Claude questions about the table data. I was curious to see how its training data would suggest possibilities for the information I had discovered. With eighteen Weatherford individuals named in the tax records, it was a daunting task to summarize the data for each one – especially since my research had found no connection to my research subject, Henderson Weatherford. It appeared these Weatherfords shared only a surname and no familial connections.

Still, I wanted to include their data in the appendix of my report, so I asked Claude to provide a summary in bullet point format for each individual and arrange them chronologically from earliest to latest.  This was a perfect use of AI’s language and analysis capability. Using the data I had abstracted from the original tax lists, Claude reviewed the tax information for each person and drew conclusions. I added this to my report and noted that the full tax table was in the appendix. For example, here is the summary for Jefferson Weatherford. He had the most listings by far in the tax lists, and this concise detailing informs the reader without being weighed down by data.

1847 – First Appearance

Jefferson Weatherford

• First settler among Weatherfords in records
• Settled on the Trinity water course
• Initially no property was listed
• Became most extensive property owner
• Multiple land acquisitions 1848-1862
• Acted as:
  • Administrator for William Weatherford
  • Guardian for Frances Weatherford
  • Agent for Sarah Weatherford
  • Administrator for Hardin Weatherford
• Owned slaves
• Significant livestock holdings

 

Wanting to analyze the tax lists more, I asked Claude, “What possible relationships can you draw from the tax lists?” I was rewarded with bullet points for documented legal relationships, property-based connections, land transaction patterns, chronological patterns, and notable points.  The AI also provided important cautions about what cannot be directly determined from the tax lists such as family structure and generational relationships. I added these to the appendix and cited Claude as the analyst.

Analysis performed by Claude 3.5 Sonnet based on the full tax list abstracted by Diana Elder, 28 October 2024.

Telling Henderson Weatherford’s Tax Story

Because the objective of this project was on Henderson Weatherford, I wanted to add more details about his tax lists and what they meant. First, I asked Claude to summarize Henderson’s tax lists by year. Notice that 1860 is the only year Henderson paid a tax. He was missing in 1861 and 1862, and then in 1863 and 1864, Samuel Beeman paid the tax for him.

1860

• First direct appearance in tax records
• Property: 2 acres valued at $250 from William Masters grant
• Livestock:
  • 5 horses valued at $300
  • 9 cattle valued at $120
• Misc property: $90 (likely blacksmith tools)
• Total Value: $760
• Poll Tax: $0.50

1863

• Listed as “Samuel Beeman per H. Weatherford”
• 41 acres valued at $205 from John Lanier grant
• Total Value: $1,035
• Poll Tax $1.00

1864

• Listed under “Samuel Beeman per H. Weatherford”
• 50 acres valued at $150 from John Frost grant
• Misc Value: $650
• No Poll tax
• Total Value: $800

I gave Claude several prompts to refine a narrative I could add to my report based on the tax data. I added my knowledge and experience with the research. I was happy with the final products and again cited Claude’s role in writing the narrative.

Henderson Weatherford’s first appearance in the 1860 tax rolls correlates with his occupation as a blacksmith per the 1860 census enumeration. The tax lists reveals Henderson owned just two acres valued at $250, purchased from the William Masters survey. This small plot was likely carefully chosen to establish his blacksmith shop. While the acreage was minimal, his total property value of $760 reflected a well-equipped tradesman. He maintained five horses valued at $300 (possibly including work horses for his shop), nine cattle worth $120, and $90 in miscellaneous property – likely representing his blacksmith tools and equipment. He dutifully paid his poll tax of 50 cents, along with state tax of $1.45 and county tax of $1.01.

By 1863, Henderson’s tax situation had changed. His records now appeared under “Samuel Beeman per H. Weatherford,” suggesting some form of business or property arrangement with Beeman, his new son-in-law. Henderson’s oldest daughter, Mary Ann, had married Samuel Hunnicut Beaman on 29 August 1860 in Dallas and the couple lived near her parents in Scyene. Samuel was an older man, about 33 and would be a likely candidate to act as an agent for Henderson.

Henderson had expanded his land holdings to 41 acres valued at $205, now from the John Lanier survey. His total property value had increased significantly to $1,035, resulting in higher taxes: $6.18 for state tax and $1.54 for county tax. Curiously, Samuel paid the poll tax of $1.00. Perhaps Henderson had died in 1862, and these were back taxes, or he was away fighting for the Confederacy.

The final tax record from 1864 reveals a significant change. The property was still listed under “Samuel Beeman per H Weatherford,” but notably, there was no poll tax paid. This absence of a poll tax, combined with Beeman’s continued management of the property, strongly suggests that Henderson Weatherford had either died or moved. The property now consisted of 50 acres from the John Frost survey valued at $150, with miscellaneous property (likely the blacksmith tools and equipment) valued at $650, for a total value of $800. The state tax was $4.00 and county tax $1.00.

1864 was the last listing for Henderson Weatherford. The tax records revealed his connection to Samuel H. Beeman and either Henderson’s death or move by 1865. If Henderson  died in Dallas County, no probate was documented.

Working with AI as a coworker for this tax project significantly enhanced the analysis. My writing time was greatly reduced, and I gained new insights from AI’s conclusions. I didn’t agree with all of them, but in many instances, AI pointed out something I hadn’t noticed.

If you’re working with large amounts of data like tax lists, I invited you to try using AI to see if it can help you with your analysis.

Best of luck in all your genealogical research!