There is a small blue book in my family. Handwritten, with ink, by a man named Anders Persson Norin. He lived from 1873 and it’s unclear when, for the last field, the date of death, he left blank. Perhaps he hadn’t made up his mind. Maybe it seemed unnecessary.
In that book, there are seven generations of my family. Names, dates, places. Short lines like “Farmer in Färange” and “Emigrated to the USA in 1926” and “Declared dead in 1852 after emigration in 1849.” Behind each line a whole life. Probably a quite hard life, considering that most lived in Hälsingland in the 1500s and subsisted by owning a number of cows.
I decided to do something with that book. Not just scan it and put it in a cloud service that no one opens. But actually find out what it was like to live as these people. And since it is 2025, there is AI that is quite good at combining historical knowledge with storytelling, I decided to take that step fully.
The result is an 80-page family saga called Roots in Järvsö. And the journey there taught me a bit about how to actually research your ancestry with AI in a way that gives more than just a dry family tree.
See how my family book was made
Innehållsförteckning
1. Photograph the book and let AI decipher the handwriting
The first step was easier than I thought. I took out my phone, opened the camera, and photographed page by page in Anders Persson Norin’s blue book. No special devices. No scanner. Just good light from a window and a phone that most of us have in our pockets.
Then I uploaded the images directly into Claude and asked for help deciphering the text.
Something unexpected has happened here. Claude is remarkably good at reading old handwriting. What appeared to be a jumble of italic loops and abbreviations and partially faded ink lines proved to be fully readable for the AI. It interpreted names, dates, and locations with a precision that would have taken me hours to figure out on my own, if I even could. Old Swedish handwriting conventions, abbreviations like “f.” for född and “d.” for död, dialectal spellings of place names that no spelling guide covers, all this was handled by Claude with a calm assurance that was almost irritatingly impressive.
The result was a clean, edited text list of all names and data from the book, ready to use as source material for what was to come.
It is worth emphasizing this for anyone who hesitates: you do not need a scanned and perfectly digitized document. You don’t even need a document in good condition. As long as the photo is sharp and the lighting is okay, Claude can handle the rest. This lowers the barrier to practically zero.
2. Gather more of what you know about your family
With the clear text as a foundation, it was about supplementing with other material. There are three sources that together provide a good basis:
Physical documents. Photograph everything. Old letters, household examination records, church books if you have access to them. The same method as with the book, phone and good light.
Digital records. The National Archive’s Digital Research Room (sok.riksarkivet.se) is free and includes household examination records, emigrant records, and tax rolls for all of Sweden. Household examination records are particularly useful because they show the entire household, who lived where, and when they moved. Arkiv Digital and Ancestry are paid supplements if you want to dig deeper.
Family knowledge. Ask your oldest relatives now. Not next year. Now. They know things that are not in any record, and they are the kind of source that cannot be recreated.
3. Tell the AI what you know
Here begins the fun. With the clear text and additional information, I entered everything into Claude: names, dates, places, professions, and the short notes from the book.
The important thing is to be specific. “I have an ancestor” doesn’t give you anything. What yields results is telling the context:
“My ancestor Jon Östensson lived around 1542 in Myra 1, Järvsö parish, Hälsingland. He owned nine cows. It was a medium-sized farm for its time. Can you describe what a day in his life might have looked like, based on what we know about the 16th-century Hälsingland peasant culture?”
That question gives you a completely different text than “write about a peasant in the 1500s.” The AI has access to enormous amounts of historical knowledge about everything from the Dacke uprising to how leather was prepared and what was drunk at the Christmas table. You need to give it a coordinate to start from.
4. Use AI as a history teacher
What surprised me most was not the storytelling. It was the history lessons.
When I asked about my ancestor Per Norin, Soldier No. 113 at Järvsö company, I didn’t just get a description of an individual man. I got the entire conscription system explained, Charles XII’s wars, Poltava, and why a peasant’s son in Hälsingland in the 1800s had a sonorous war name instead of his patronymic. Suddenly it all connected.
That is the AI’s great strength in this context: it can zoom out and zoom in. It can say “In the year 1709, when your great-grandmother’s grandfather was likely young, this happened in Sweden” and then zoom in on how it impacted a specific village in a specific valley.
Questions that work well:
“What happened in Sweden during this decade, and how did it affect ordinary peasants in Hälsingland?”
“How was wedding celebrated in this estate and this region during the 1800s? What did they eat, what did they drink, who sat where?”
“My ancestor emigrated to America in 1849 and was declared dead in 1852. What could have happened? What were the most common scenarios?”
The last question gave me three plausible stories about Jonas Hansson who disappeared in America. None of them can be proven. All are historically possible. That is precisely what genealogy sometimes is: likely narratives about unprovable facts.
5. Let AI write, but steer the story
It’s important to understand what AI can and cannot do in this context.
It can write vivid historical prose based on facts, explain social structures, eating habits, festive culture, and class hierarchies, and build a credible context around the dates and names you provide it. It cannot find facts about your specific ancestors, access records you have not shared, or guarantee that the historical details it presents are perfectly accurate down to the decimal.
You are the one at the helm. The AI is a very knowledgeable writer who cannot go to the library by itself. You provide the source material, you steer the tone, you decide what to include. In my case, I made conscious decisions for the text to be humorous but respectful, that historical facts about, for example, the Spanish flu of 1918 would be accurate, and that the fictional scenes would clearly come across as reconstructions.
The result: 80 pages about nine generations in the Järvsö valley, from the 16th century cows to the 1940s grandchildren in Duluth, Minnesota.
6. From 28 lines of data to about 80 pages
The original dataset was small. Very small.
Anders Persson Norin’s blue book contained about 28 lines of factual information about the direct family line. Names. Dates. Places. Occupations in one word. “Farmer.” “Soldier.” “Merchant.” Sometimes a number of children. When counting spaces and punctuation, it totals perhaps 400 words — about half an A4 page of information about seven generations of people.
The finished book Rötter i Järvsö contains 18,137 words divided into 383 sections and about 80 to 95 pages. That’s a multiplication factor of about 45.
But it is not the AI that invented the 17,700 words. It is history.
The AI can take a small, specific, and dry dataset and place it in an enormous context of already existing knowledge. Every factual detail in the blue book activates a layer of historical context that the AI is already familiar with.
“Soldier of Järvsö Company” triggers: the division works, Karl XI’s military reform, the wars of the Great Power Era, Poltava 1709, how soldier farms functioned, what common soldiers ate in the field, how soldier names were assigned. All this is established historical knowledge. The AI knows it. It just needs a coordinate to activate it.
“Emigrated to the USA in 1926, Duluth Minnesota” triggers: the late Swedish emigration wave, Duluth as a mining town by Lake Superior, the organization of the Scandinavian immigrant community, how Swedish surnames were Anglicized and why.
“Died September 2, 1918, 18 years old” triggers: the Spanish flu, its second and deadliest wave in the autumn of 1918, why young adults were particularly affected, how the pandemic reached rural villages in Sweden.
Every line in the blue book was a key to a room that already existed. The AI opened the doors.
It’s not about the AI making things up. It’s about your family’s specific data, however small it may be, providing an anchor for history to hang on. Without the anchor, history is abstract. With the anchor, it becomes personal. “The farmers in Hälsingland were affected by the aftermath of the Dacke uprising” is a historical meaning. “Jon Östensson in Myra 1, with his nine cows, lived three years after Nils Dacke was dismembered and put up on a pole” is something entirely different.
It’s the difference between reading a Wikipedia article and reading about your own family.
7. The empty field
There is a detail in the blue book that I can’t let go of.
Anders Persson Norin carefully filled out all the fields. Date of birth. Places. Occupations. Names of children. Names of grandchildren. He built a coding system of a precision that impresses even today. And in the end, for himself, he filled in everything except one field.
Date of death. Empty.
I asked the AI to rewrite it. Not to explain it. Just to rewrite it. And it was the most beautiful passage in the whole book. That is what AI can do with a little blue book and some attention. It can give the dead a voice that is more than three lines in a church register.
8. Here’s how to get started
There is no right way into this. But there is a way that works better than most others, and this is it.
Start with the physical. Go home to your oldest living relative and ask if they have anything. Letters, photos, documents, a book. Take pictures of everything with your phone in good light. Do it now, not the next time you visit them. That kind of source can disappear without warning and it cannot be recreated.
Build a skeleton first. Before you bring in the AI, build a simple family tree with the names and dates you have. It doesn’t need to be fancy. A table in Word or a notebook will do. You want: names, birth year, death year, location, occupation. That’s enough. The reason you do this first is that it forces you to see what gaps you have, and the gaps are actually interesting. An ancestor who disappears from the records can be just as story-telling as one who is well-defined.
Upload your images directly to Claude. If you have photographed old handwritten documents, upload them and ask Claude to decipher the text before you do anything else. You don’t need to preprocess the images or have a perfect result. Just write: “Can you read and transcribe the handwritten text in this image?” Claude handles old Swedish handwriting, abbreviations, and faded ink better than you might expect.
Go to the National Archives. sok.riksarkivet.se is free and contains house examination registers, census records, and emigration records for all of Sweden. Search for your ancestor’s name and parish. Often you will find more than you expected.
Select a person and a time. The most common mistake is being too broad. Choose a specific person, place them in their time and place, and ask about that. Use this template:
“I am writing about my ancestor [name], who lived [approximately years] in [location and region]. His/her profession was [occupation]. Can you describe what everyday life might have looked like for someone of his/her status and time? Please describe the living conditions, food, the rhythm of the seasons, and how significant historical events during that period may have directly affected him/her.”
Iterate and deepen. The first answer is not the best answer. Ask follow-up questions about what piqued your curiosity. Each answer opens new doors.
Write chapter by chapter. When you feel you have sufficient historical understanding of a person, ask the AI to write a chapter. Set a tone, determine if fictional scenes are allowed and how they should be marked, read through and adjust.
Gather everything in the end. Ask the AI to write a foreword and an epilogue. Add a character gallery with all documented names and dates as an appendix. This gives the book a reference character beyond the story, and it is often the part that relatives refer to first.
It takes time. Expect ten to twenty hours spread over several weeks for a serious project. But it’s your story, and it disappears if no one takes care of it. A small blue book with 400 words can become 18,000. It just requires someone to decide to open the doors.
Select a person and a time. The most common mistake is being too broad. Choose a specific person, place them in their time and place, and ask about that. Use this template:
“I am working on writing about my ancestor [name], who lived [approximately years] in [place and region]. His/her occupation was [occupation]. Can you describe what everyday life might have looked like for a person of his/her status and time? Please describe the dwelling, the food, the rhythm of the seasons, and how significant historical events during that period might have directly affected him/her.”
Iterate and delve deeper. The first answer is not the best answer. Ask follow-up questions about what piqued your curiosity. Every answer opens new doors.
Write chapter by chapter. When you feel that you have sufficient historical understanding of a person, ask the AI to write a chapter. Set a tone, determine if fictional scenes are allowed and how they should be marked, read through and adjust.
Gather everything at the end. Ask the AI to write a preface and an epilogue. Add a character gallery with all documented names and dates as an appendix. This gives the book a reference character beyond the story, and it’s often the part that relatives look up first.
It takes time. Expect ten to twenty hours spread over several weeks for a serious project. But it’s your story, and it will disappear if no one takes care of it. A small blue book with 400 words can become 18,000. It just takes someone to decide to open the doors.
Sources
Riksarkivets Digitala Forskarsalen: sok.riksarkivet.se
Arkiv Digital: arkivdigital.se
Claude (Anthropic): claude.ai
Hälsinglands museum: hembygd.se
Indelningsverkets historia,
Riksarkivet: riksarkivet.se
