In this post, I share the story of my work with ChatGPT Plus in December, 2025, when I went from a light user of the free version to a constant user of the paid version.
Me and free ChatGPT since 2023
For me, using free ChatGPT has been like going to a fun cocktail party on a deadline. Just when the conversation got going, I had to go because I ran out of free prompts on the best version of Chat. I tried a few times to continue on the lesser models, but it was terribly unsatisfying – as I’m sure it was designed to be – because once I got used to the best answers, I was spoiled. I told myself that was OK, my questions weren’t urgent, and I could keep a long running thread open to explore various topics. And don’t get me wrong: I have learned a great deal and benefited from it. Looking back, I was light on learning how Chat processed uploaded documents, because that was a fast way of expending my free prompt limits.
That’s how I’d been using ChatGPT for a year and a half. Others said I should try the other models to see if I liked them better (and I understand that using all the platforms also theoretically extended my free usage). That’s like asking me to try different brains and see if I prefer thinking this way or that way. The capacity of my own brain to understand one model to my own satisfaction is limited – how could I possibly take on multiple brains? No thank you. I leave it to the experts to test and report back. And that’s how it might have stayed – me on free ChatGPT – but for a walk in the park.
Go for a walk, come home with a killer insight
About a month ago, my friend and I went out for an evening walk with her dogs, and we talked, as we do, about everything from personal to professional life. I’m saying that we didn’t mean to talk about ChatGPT, but it evolved naturally as part of us sharing what’s on our minds. She had a killer insight. What if you could train Chat to work according to your own values, parameters, ethics, and language? She upgraded to the best paid version and created a working agreement tailored to her needs. She gave me examples of how she’s using it.
I can’t tell you what a revelation this was. I was inspired. I signed up for ChatGPT Plus immediately, and the first assignment was creating that working agreement. It took me days of reflection – thinking about my goals, aspirations, needs, ethics, and honesty. My second series of inquiries had to do with privacy. If I was going to be asking sensitive questions and uploading records, how safe were my thoughts and those records? Was I inadvertently going to be training the data model to make myself obsolete? Switching metaphors from cocktail parties to T-Rex dinosaurs, I want to use the big, powerful beast, but I don’t want it to eat me.
Aside – I was an early adopter of internet search engines. I could see how it would change my way of thinking, and how memorized knowledge was going to be less important, but keyword searches and pattern recognition was going to be more important. During those early years as I went from Ask Jeeves to Yahoo! to google, I could feel my own brain rewiring. Knowing information is still important, but locating and analyzing became just as important. I still miss the twenty-something me who was a trivia master but I love the current me who is a recognized genealogist.
Evolving questions in using ChatGPT
My basic questions for ChatGPT are these. What’s the best way to leverage AI to level up what I do? And if AI has the potential to eat my lunch, how can I use it to become a person whose lunch is out of reach?
With privacy fears allayed, I could embark on something ambitious. I was sitting in the lounge area at my pilates studio, waiting for a ride home, when the project idea came to me. What if I could train Chat to create an analytical model framework to help me better understand the records I use?
My idea: a custom analytical model
Think about it. As family historians, we work like this: search for records about our ancestors, find the records, then learn about the record. (Yes, I know sometimes we need to first learn about the records, but be honest: that usually only happens when the easy search has nil results, right?) If we are working with foundational sets like census records, we might come to understand the record-making process through seeing multiple censuses for a family as they are born, marry, have children, and the children grow up and have children of their own. This is largely how I’ve worked for thirty-odd years. I learn about record creators as I find records.
The problem is, my area of genealogy is baffling. I can’t tell you how many records I’ve found that were created by some unknown-to-me law or Order-in-Council that had an effect and created a paper trail but whose origin was murky. Chinese Canadian genealogy seems to be largely gap-based. It’s like there’s one stream for the greater population and different stream for mine. I’ve learned to ask: Does this apply for all Canadians, or only some? If only some, when, for whom, and why? For example, the idea that all Canadians before 1947 were British subjects (and have implied citizenship). The answer is: mostly right, but the government reserved the right to define it circumstantially when it wished. People will send me records they don’t recognize and ask absolutely normal questions: What is this? Who created it? Why was it created? What time frame was this in place? What does this mean?
To give a short list of records that are exceptional in Canadian history: Chinese Case Files; Chinese immigration (C.I.) 5, 9, 28, 30, 36, 44, 45 certificates and forms; General Register of Chinese Immigration; KMT registrations; Ledgers of exempt admissions; NF Registers of Chinese Immigration; Port Ledger of Chinese Immigration (New Westminster); and refunds for capitation tax. And that’s when I fall into yet another in a lifetime of deep rabbit holes. I don’t know how many hours I’ve expended researching, for example, civil rights, citizenship, immigration, naturalization, and voting. I did not set out to study colonialism, exclusion, and racism, but it’s been necessary to understand the latter group in order to find out answers about the former. And I study all of that in order to make sense of the exceptional records that were created about the Chinese Canadian community.
Which brings me back to ChatGPT. It’s my new favourite toy. What if I could build an analytical model that, when trained with original sources, could help me analyze records? People. Do you know how many versions of the Chinese Immigration Act were enacted? How many Orders-in-Council (OICs)? How many circulars, regulations, and memos? How many legal cases? (Actually, that one is easy, and the reason why there were so few immigration-related legal cases is that the Privy Council made it policy that the decisions of the Chief Controller of Chinese Immigration were final and not able to be challenged in court. No court means no legal cases.) I’ve been slowly feeding ChatGPT one law after the other, developing analytical prompts, building a framework, rinse and repeat. Along the way I have learned so much about these legislative instruments that my brain hurts.
I’ve been building this model, statute by statute, OIC by OIC, piece by piece. It’s mid-level robust already, but I’m a long way from my goal. It needs rigorous testing to see if its conclusions are right. I still don’t have a good sense of its fallibility and keep an eye out for problems in its data processing. For example, I’ve identified when it’s assuming a Chinese name is a common English name (Dake, not Dick or Jack), and created a rule to prevent such mishaps from happening again.
I’m learning Chat’s limitation on creating good citations. I now understand what it can and can’t see, and how to develop a prompt that will result in a more accurate Chicago Manual of Style citation. The results are not Evidence Explained level, but they are good enough for creating wayfinders. Put another way, I wouldn’t use ChatGPT’s citations unedited, but perhaps I will be able to develop a better prompt, and in the meantime, the citations it’s generating save time and guide me back to the original record.
Next steps
Going forward, I’m inspired to write a number of ChatGPT-related blogs. Ideas that come to mind are i) what is a working agreement and why do you need one; ii) what does privacy mean to ChatGPT; iii) the problems with citations and why developing a prompt is a good idea; and iv) what to know in order to build your own analytical framework.
Afterword
I’m building a custom analytical framework in ChatGPT Plus (paid version, 5.2 as of December 2025) to help me understand Chinese Canadian records. Along the way, I’ve been blasted with one revelation after another. ChatGPT isn’t like google. ChatGPT and other large language models are growing into the adult T-Rex dinosaurs of computing: massively powerful but also with downsides. I want to use the big, powerful beast and I don’t want to get eaten, and I think the slower we are to get on board, the harder it will be. Working with ChatGPT 5.2 now is like training with a baby T-rex. I’ve learned in a short time how to set it up, what it does with private data, what keywords are important, how to create my first ChatGPT-generated prompts, and a few limits of the model. In the meantime, I am agog at what I’ve got so far, and this is only the beginning.
Oh, and every word in this blog was written by me. I’ll tell you when it’s AI-generated.
Thank yous
To the organizers and presenters at the September East Coast Genetic Genealogy Conference, to the Facebook group Genealogy and Artificial Intelligence, to Stephen Little and Mark Thompson, and to my friend who asked me to join her on a walk and gave me insights of inestimable value – thank you. Learning something big and scary is easier with a community like this. And to my friend Kendra Gaede, who said I should write up my findings, even though I feel like my understanding of ChatGPT is remedial. Kendra, this is for you.