Making ChatGPT a Valuable Genealogy Assistant

By Laura Clark Murray

Since the release of ChatGPT in late 2022, this artificial intelligence “bot” has been wowing users by engaging in detailed conversations, answering questions, and even writing stories and essays. But at its core, ChatGPT lacks the ability to distinguish between fact and fiction. Since facts are critical to what we do as genealogists, it’s important to understand how to keep the tool in the realm of reality. If you do that, ChatGPT can become a valued genealogy assistant!

What is ChatGPT?

ChatGPT is an online artificial intelligence chatbot which generates responses to a user’s prompts. You engage with ChatGPT by typing in a question or prompt. For example, when I ask “What is ChatGPT?”, it responds:

Hello! I am ChatGPT, a large language model developed by OpenAI, based on the GPT (Generative Pre-trained Transformer) architecture. My purpose is to provide conversational assistance and answer a wide range of questions on various topics such as science, technology, history, literature, and more. I have been trained on a massive corpus of text data, including books, articles, and websites, and I use this knowledge to generate human-like responses to user queries. I can understand natural language input and provide informative, helpful, and engaging responses in a conversational manner.

Chat bots are nothing new and you may be familiar with them from interacting with customer service agents on websites and via text, though not all chatbots make use of artificial intelligence (AI) as ChatGPT does. It relies on technology known as deep learning and makes use of natural language processing (NLP) and machine learning techniques, allowing it to comprehend human language and generate human-like responses to text-based conversations.

Words, not Facts

ChatGPT will respond to most questions and quickly generate text that may sound authoritative, but may not be based in fact. Ask it to tell you about a specific ancestor of yours and ChatGPT may comply, but the details are likely to be made up. These “hallucinations” can be deceiving and appear to be factual. Asking ChatGPT for its sources doesn’t fix the problem, as the sources may also be fabricated! This is in part due to the fact that, as of this writing in April 2023, ChatGPT is not connected to the internet. Instead it relies on its training material which, though vast, captures a limited understanding of the world. More importantly, ChatGPT is engineered to string words together to create a natural-sounding response in the context of a prompt, without any understanding of what is or is not true.

So how can we use a tool that we can’t believe? By constraining it to work only with material we know to be true and explicitly telling it what we want it to do with that material. Once we understand this, the possibilities for using ChatGPT as your personal genealogy assistant are truly exciting. Let’s look at a few examples.

Write a narrative profile of an ancestor

Provide ChatGPT with the facts of a person’s life and ask it to write a profile from that material.

Prompt:
You are an expert genealogist. Write a narrative genealogical profile of Addison Mills, using only the information provided here. Do not speculate.

I then copied the facts displayed on the Ancestry page for my ancestor Addison Mills, and pasted them into the above prompt.

Result:
Addison Mills was born on July 11, 1826, in Bradford, Vermont, to Daniel Mills and Ireney Chandler. In 1850, at the age of 24, Addison married Margaret Pastores Brown, and they settled in Boscawen, Merrimack, New Hampshire, where Addison worked as a Miller. The couple had several children, including William Valentine D. Mills, Luther Manson Mills, Benjamin Franklin Mills, Addison Mills, Bernard Brown Mills, Charles Waters Mills, and Edward Beasley Mills.

Around 1860, the family moved to Malone, Franklin, New York, where Addison continued to work as a Miller. In 1863, they moved to New York City, where Addison worked as a Miller and lived at 347 E. 54th Street. In 1866, the family moved to Connecticut, where they lived in Stratford and where Addison continued to work as a Miller.

In 1882, Addison worked as a Peddler and moved to Bridgeport, Connecticut. He lived in Bridgeport until his death in 1911 at the age of 84. Addison was survived by his wife Margaret and his children William, Luther, Benjamin, Addison, Bernard, Charles, and Edward.

Be sure to review the results to ensure that everything is accurate and that ChatGPT did not concoct some incorrect details.

Put a profile in the first-person

Take a narrative profile, either one you’ve already written or one that ChatGPT generated for you, and have ChatGPT rewrite that narrative in the first person as that ancestor.

Prompt:
You are an expert genealogist. Write a narrative genealogical profile of Addison Mills, in the first person, using only the information provided here. Do not speculate.

Before hitting enter, I pasted the profile that ChatGPT generated for Addison Mills in the previous example. Here’s the result:

I was born on July 11, 1826, in Bradford, VT. In 1849, when I was 22 years old, I became a Miller at Gerritsen Mill in Brooklyn. It was on January 4, 1849, when I moved to Brooklyn.

Around 1850, when I was 24 years old, I married Margaret Pastores Brown. She was born in 1830 and died in 1913. We had several children together: William Valentine D. Mills (born February 14, 1852), Luther Manson Mills (born June 28, 1855), Benjamin Franklin Mills (born April 12, 1858), Addison Mills (born July 17, 1863), Bernard Brown Mills (born July 21, 1866), Charles Waters Mills (born October 2, 1868), and Edward Beasley Mills (born November 3, 1871).

In 1850, when I was still 24, I was a miller in Boscawen, Merrimack, New Hampshire. By 1860, I had moved to Malone, Franklin, New York, where I continued to work as a Miller.

Unfortunately, my daughter Elizabeth Duer Mills died in infancy in 1862. I also experienced the deaths of my father Daniel Mills in 1854, and my mother Ireney Chandler in 1857. My brother John Mills passed away in 1863.

Throughout the late 1800s, I continued to work as a Miller, living in Stratford, Fairfield, Connecticut. However, by 1882, I had become a peddler in Bridgeport, Connecticut. I remained in Bridgeport until my death.

I died sometime before March 28, 1911, when an obituary was published about me in The Bridgeport Times and Evening Farmer.

You might even direct ChatGPT to write the first-person narrative with a particular tone, such as “a man from the midwest in the mid-1800s.”

Revise a piece of writing to be more engaging

Here I’ve asked ChatGPT to take a somewhat boring 250-word section of background material that I wrote for a client research report and make it shorter and more engaging. While I don’t love the “Ahoy, mateys!” phrase it came up with, I can edit that bit out of the result. Or I can provide additional guidance and have ChatGPT give it another try.

Prompt:
Assume the role of an expert genealogy writer with a conversational writing style. Revise the following passage to be more engaging. The result should have a maximum of 50 words. Use only the information in the following text.

After the prompt, but before hitting enter, I pasted the following 250-word text I wanted rewritten:

The close of the 18th century and the first couple of decades of the 1800s were tumultuous years for American maritime workers. Tensions escalated between America and Britain as the British Royal Navy enacted the practice of “impressment” — forcing maritime workers found in ports around the world and on American merchant ships into military service, refusing to recognize naturalized American citizens and claiming them to be British citizens. It became critical to determine the citizenship of mariners aboard American ships. As American-born sailors often lacked formal paperwork that validated their citizenship, starting in 1796, the US government allowed them to apply for a “Seaman’s Protection Certificate,” which they could present at a custom house or an overseas consulate if they were out of the country.

To further protect the American sailor and to punish foreign nations for their impressment practices, President Thomas Jefferson signed the Embargo Act in 1807, putting a halt to trade with foreign nations. The embargo devastated maritime workers throughout the country, as they lost their means of making a living with ships sitting idle in port.

Without work in their home ports and, therefore, no income to support themselves and their families, many sailors attempted to illegally skirt the embargo by heading to Canada and other foreign ports. In New York City, 46% of the individuals who had identified as “mariners” in the 1808 city directory were no longer listed in the 1809 directory, which may have been a result of the exodus to find work.

Here’s what ChatGPT generated in response:

Ahoy, mateys! Let’s set sail back in time to the tumultuous years for American maritime workers at the close of the 18th century and the first couple of decades of the 1800s. As tensions between America and Britain grew, the British Royal Navy enforced the practice of “impressment,” forcing maritime workers into military service, causing chaos on American merchant ships. To determine the citizenship of sailors aboard American ships, the US government introduced a “Seaman’s Protection Certificate.” However, President Thomas Jefferson took it a step further and signed the Embargo Act in 1807, which devastated maritime workers throughout the country. With no work, many sailors attempted to illegally skirt the embargo by heading to Canada and other foreign ports. In New York City, almost half of the “mariners” in the 1808 city directory disappeared in the 1809 directory, likely due to the exodus to find work.

Pull out all the genealogical information from an obituary or newspaper article

Consider all the genealogical details that can be found in an obituary, wedding announcement, and other documents. Put ChatGPT to work collecting those details into a table for analysis. If you don’t already have a text document with the content of the newspaper clipping for copying and pasting into your ChatGPT prompt, you’ll need to transcribe the document first or make use of a tool to do that for you.

Prompt:
Below is the text of an obituary. State the relationship between the people in the article. Do not speculate. Format the information as a four-column table, with the headings “Person 1”, “Person 2”, “Relationship”, and “Quote”. In the table, list the relationships between each pair of people detailed in the article. In the fourth column, include the quote that describes the relationship of that pair.

In this example, I copied the text of an online obituary for P.T. Barnum, the American showman who lived in the 1800s, to the end of the prompt. Here’s the table that ChatGPT generated:

Where to find ChatGPT

In your web browser, go to chat.openai.com and create a free account. That’s it — you can start interacting with the tool in the main window. You’ll see a running list of your previous chats on the left-hand side of the window, which will be there again when you next log in. Occasionally the site may be too busy to allow you to log in. In that case, just try again after a bit of time.

It’s worth noting that OpenAI is regularly updating ChatGPT and folding new capabilities into the free version.

A paid upgrade to ChatGPT Plus is currently available for $20/month, which you may want to consider if you want priority access to new features. All the examples in this post were done with the free version.

A few tips and tricks

Start a new chat: Unless you want ChatGPT to revise the response it just provided, always start a new chat for a new topic. If you don’t, the tool is likely to take the bits from previous interactions and fold them into the new response. For example, if you’ve asked ChatGPT to summarize a passage about the Santa Fe Trail in the 1800s and then ask it to work with information about an ancestor’s World War II service, you may get an answer that weaves the details of the two topics together.

It’s all about the prompt: The more clear, detailed, and precise your prompts are, the more likely you’ll get a response that you like. For example, try alternatives to “rewrite” that provide more specific direction, such as “revise”, “restructure”, or “reduce”. There isn’t a definitive guide on how to write effective ChatGPT prompts, so you may want to play around with this to see what kind of results you get back.

Dealing with long text: As of April 2023, the limit on a prompt appears to be about 500 words. If you have a long document that you’d like to work with, try working with smaller chunks and piecing the results together.

More advanced uses

There’s a lot more that ChatGPT can do to assist us with our genealogy research, including generating a GEDCOM file from an obituary or a narrative story, summarizing journal articles, translating text, and even writing code. And with internet access and image processing features on the horizon, we’re just beginning to learn how to make the most of this remarkable tool.

Have some fun with it!

Take advantage of ChatGPT’s ability to be creative and ask it to generate a limerick for you. Here’s an original ChatGPT-generated limerick that I hope you enjoy:

Diana and Nicole, a duo with a genealogy flair,
Teach their readers to trace roots with great care,
Expertise and guidance that’s second to none,
Until family trees come to life and fun,
Thanks to this mother-daughter pair!