Carole McCulloch, family history researcher and founder of the Essential Genealogy Academy, explains how AI tools can help genealogists tell the story of their ancestors.

Genealogy and AI: a use case

We have 16 great-great-great-grandparents. Most came into the world in the early 1800s and departed, leaving behind records that are faded, handwritten, and occasionally written in a script designed specifically to defeat the modern researcher. My goal is to write a biography of each of my 16. That ambition, not so long ago, might outlast me.

Then I found a letter. And then I found AI.

The letter had been sitting in Ancestry's archive for several decades. Four pages of Victorian copperplate, written on 23 October 1891 from 25 South Molton Street, Hanover Square, London. A chance discovery. The writer was Edward James Hall. The recipient was his cousin Nellie, a woman I would come to know as Ellen Warren. What I did not know when I opened it was how personally it would land. Edward James Hall was the brother of my 2x great-grandmother, Mary Ann Hall. I was reading a letter written by my own family.

Within those four pages, Edward had left me something extraordinary: a cast of characters connected by blood, grief, and 30 years at a brewery in Dover. And threaded through every relationship he mentioned was a family puzzle I had not known to look for: the tangled, affectionate story of the Hall and Warren families, whose connection I could only unravel with AI's help.

My goal is to write a biography of each of my 16. AI has not shortened the research. It has made the writing possible.

How can AI help family history research?

My research begins with a genealogy platform, such as Ancestry, MyHeritage and FamilySearch. Here I gather records and let the platform's hint system surface potential matches. That confirmed data is what I bring to general-purpose AI tools. I do not ask AI to find my ancestors. I ask it to help me do something with the ancestors I have already found.

The workflow has six steps:

Stage	What happens
Gather	Collect records and tree data in genealogy platforms. AI hints surface matches automatically.
Transcribe	Upload document images to Claude or ChatGPT for a working first draft; review and correct every line.
Extract	Ask AI to pull every name, date, location, relationship, and occupation into structured form.
Interrogate	Hold an iterative conversation. Ask questions, request labelled hypotheses, push back.
Summarise	Feed supporting documents to AI for focused summaries keyed to specific surnames and dates.
Transform	Provide a brief of verified facts and request a narrative first draft. Rewrite in your own voice.

Transcription

Transcription came first. Four pages of faded Victorian script became editable text within moments. I reviewed every line, corrected misreadings, and noted uncertain words in square brackets. The paralysis of staring at illegible handwriting dissolved almost immediately.

Think of AI as a First Draft Specialist. You remain the Final Editor. Once transcribed, the letter yielded 11 named individuals, three countries, two diseases, one brewery, and a marriage that had ended in tragedy.

Extraction

Extraction came next. Edward writes as Victorians do; in flowing prose that scatters genealogical gold across paragraphs. Names appear mid-sentence, relationships are implied, and locations are dropped in passing.

I asked AI to extract every person named, every relationship, every date and location into a clean, structured list. For the first time, I had a cast of characters:

• Aunty, whose funeral Charlotte and Emily were arranging in Bexley.
• Cousin Alfred, 30 years at a Dover brewery, is now recovering in Australia.
• Nellie herself is now living in Brussels.
• Ina, whose affairs Edward wished to discuss, whose identity would take months to confirm.

Summarisation

Summarisation kept the research manageable. Identifying 11 people generates a substantial pile of secondary material:

• parish records
• census returns
• probate documents
• emigration
• marriage records

I routinely use AI to summarise that material before deciding which documents merit close reading. When you are researching 16 family lines simultaneously, that saving compounds enormously. One principle holds firm: every AI summary is a signpost, not a destination. The source always has the final word.

Transformation

The transformation turned data into a narrative. With a verified body of evidence assembled, I built a structured brief of confirmed facts and brought it to Claude, asking for a narrative first draft with all contextual detail clearly flagged.

What I received placed Edward at his writing desk in Hanover Square, composing words that would travel to Brussels and tell Nellie that her mother was gone. I rewrote it, checked every claim, removed anything invented, and the result is now published in my ancestral archive. I am always the author. AI is my co-writer.

Which AI tools are best for family history research?

The AI landscape in genealogy generally falls into two categories:

1. general-purpose tools you bring your research to
2. platform-specific features built into genealogy databases

ChatGPT and Claude are the strongest partners for transcription, extraction, and iterative family analysis. Perplexity adds live web search for current archive news.

On the platform side, Ancestry's Listen and Explore feature makes tree navigation more accessible; MyHeritage's AI Biographer generates narrative life stories from your tree data; and FamilySearch's expanding full-text search makes documents searchable by any word they contain.

Tool	Best for	Free?
ChatGPT	Transcription, research planning, and historical context	Yes
Claude	Long documents, iterative family analysis, narrative drafting	Yes
Perplexity	Live web search combined with AI, current archive news	Yes
Ancestry AI	Hints, Listen and Explore, tree matching	Limited
MyHeritage AI	AI Biographer, photo restoration, smart matching	Limited
FamilySearch	Full-text search across billions of digitised records	Free

How should you talk to an AI tool?

The most important skill in AI-assisted genealogy is not which tool to use; it is knowing how to talk to it. The step most researchers underestimate is the iterative conversation. AI is not a one-shot oracle. You push back, ask follow-up questions, and request that every hypothesis be clearly labelled as such.

Example

The key question I brought to AI was deceptively simple: if Nellie is Edward's cousin, what relationships could produce that cousinhood? AI laid out the logical options. I fed it the facts I held; that Sarah Ann Hall was Edward's mother; that Charlotte Elizabeth Hall appeared in connection with the Warren family; that the Hall family shared grandparents in James Cruise Hall and Elizabeth Tonbridge. AI's response crystallised what I had been circling. If Charlotte was Sarah Ann's sister, and Charlotte had married Joseph Warren and had a daughter called Ellen, then Ellen Warren was Edward's first cousin. The letter was a family death notice.

That one iterative exchange gave me three months of confirmed research to pursue. Ina turned out to be Edwina Maurice, who had married James Robert Scott in 1888 and died of peritonitis a year later. Scott had indeed remarried quickly, prompting Edward's dry note: he soon forgot his first love. Nellie was Ellen Warren, now Ellen Brandenburg, living in Brussels with her husband Nicholas Clement Brandenburg and their four children. The Warren family, whom I had not known to look for, had been hiding in plain sight in a letter I could not read.

What is the C.R.A.F.T. prompting framework?

A vague question produces a vague answer. The C.R.A.F.T. framework gives you five ingredients for a precise, powerful AI prompt:

• Context: Tell the AI who you are and what you are researching, for example “I am tracing two interconnected Victorian families, the Halls and the Warrens, across three generations.”
• Role: Ask the AI to adopt relevant expertise, for example “Act as an experienced genealogical researcher with knowledge of Victorian London records.”
• Action: State precisely what you need, for example “Extract every person named in this transcribed letter and list their apparent relationships to the writer.”
• Format: Specify how you want the answer, for example “Present findings as a structured table: Name, Relationship to Writer, Location.”
• Tone: Set the register, for example “Keep it accessible. I am experienced in genealogy but new to AI tools.”

One further principle underpins all of it: bring your data to AI, not the other way around. When you arrive with confirmed facts from primary sources, you set the terms of the conversation. The AI responds to your verified evidence and cannot invent an ancestor you have not given it room to imagine.

Summary

Somewhere in an archive, a database, or a shoebox in a relative's attic, there is a document with your ancestor's name on it. It might be a letter written in a hand you cannot read, a census return that raises more questions than it answers, or a probate record dense with names you do not yet recognise. You have walked past it, photographed it, bookmarked it, and told yourself you will come back to it when you have more time.

This is the moment to go back.

Start with what you have. Open Ancestry, gather the records you have already collected, and let the platform's hint system do its background work. You are not starting from scratch. You are starting from everything you already know, and that is more than you think.

When you find the document that stops you, the one that resists you, upload it to Claude or ChatGPT and ask for a transcription. Do not wait until you can read it yourself. The first draft is not the final word: that responsibility stays with you. But the paralysis ends the moment you upload, and what you find in the readable version will surprise you.

Once it is legible, ask AI to extract every name, every relationship, every date and location. Let the tool do the sorting while you do the thinking. Then comes the conversation that changes everything: not a single question, but a sustained back-and-forth in which you feed AI what you know and ask it to show you what the evidence suggests. Label the hypotheses, challenge the inferences, ask what the documentary record would look like if the suggestion were true. Then go and find out.

As your research grows, use AI to summarise the supporting material so you can move quickly through volume without losing accuracy. And when the evidence is solid and the story is clear, bring your verified facts to AI and ask for a narrative first draft. Then rewrite it in your own voice, because the story belongs to you, not the tool.

You are not just learning to use AI; you are learning to direct it. And the ancestor waiting at the other end of that process has been waiting long enough.

The workflow is in place. The tools are free to try. The only decision left is which ancestor you are going to explore first. Pick the one who has been waiting the longest. Upload your records. Design your best prompt and direct your AI tool strategically.

AI will assist you in revealing your ancestor’s story effectively.

About the author

Carole McCulloch is a veteran family history researcher and "edupreneur" with over 30 years of expertise in British genealogy. Since the 1990s, she has specialised in the social histories of the Allery and Cutting families, tracing their paths from Devon and Pembrokeshire to Surrey. Carole migrated to Australia in 1949 and now lives in Wodonga, Victoria.

In 2020, she founded the Essential Genealogy Academy, where she serves as a lead educator for a global community of family historians. With a portfolio of over 40 courses, she bridges the gap between traditional archival research and modern digital storytelling. Her current work focuses on the intersection of heritage and technology, specifically guiding researchers through Ancestry in the Age of AI to transform raw data into compelling ancestral narratives.

See also

How to search The Gazette

Images

Adobe Stock

Publication date

26 March 2026

Any opinion expressed in this article is that of the author and the author alone, and does not necessarily represent that of The Gazette.