Building Your Family Tree with AI (2026)

My great-grandfather was a blacksmith in a village of 600 people in the Corrèze region of France. I knew that, and nothing else. Not his own father, not his wife's maiden name, not where the family came from before the 19th century.

In three weeks, I traced back five generations. 47 named ancestors, three portraits recovered from digitized archives, one animated via Incarn.

AI helped me. Not by searching on my behalf: AI can't do that yet. But by deciphering what I found, analyzing what I didn't understand, and speeding up every step by a factor of two to three.

Here's how.

What AI can (and cannot) do for your genealogy

A clarification before you invest your time.

AI cannot: access archive databases (Ancestry, FamilySearch, and Geneanet have their own interfaces), find your ancestors by name, or guarantee the accuracy of what it says about proper names or dates.

AI does well: transcribing 19th-century handwritten records into readable text, interpreting historical context ("what did 'cultivateur' mean on a French marriage certificate in 1870?"), formulating adapted queries for archive search engines, translating old French or ecclesiastical Latin, and connecting information when you're stuck.

AI is an analysis assistant, not a genealogy search engine. With that distinction clear, it's genuinely worth the effort.

Step 1: Interview the elders before it's too late

This is the most urgent step, and the least technological.

Your grandparents, great-grandparents still living, aunts and uncles over 70: they are human archives with no digital version and no backup system. Every death without a recorded interview means an entire branch of the tree disappears.

Before any search engine, pick up the phone. Ask for parents' first names, home villages, occupations, and the stories told at Christmas dinner. Take notes or record with their permission.

My 84-year-old great-aunt gave me the full name of my great-grandfather's father, his village, and the fact that the family had moved during the Occupation (which explained why I found nothing in the Corrèze records after 1942). One hour of conversation I could have had twenty years earlier.

ChatGPT can then structure these notes into a table: individuals, approximate dates, places, known sources. Not magic, but useful for not losing anything at the start.

Step 2: FamilySearch and Geneanet for the first records

FamilySearch is free, with no mandatory registration for basic consultation, and indexes tens of billions of civil records. It's the logical starting point for 19th-century France, as well as for US, UK, and European genealogy more broadly.

The method with AI:

1. Search for the record on FamilySearch or on a regional archive portal
2. Download the record image
3. Send the image to ChatGPT or Claude with the prompt: "Transcribe this handwritten record into modern language. Mark uncertain passages with brackets."
4. Verify the flagged passages before adding them to your tree

The time savings are measurable. A 19th-century marriage certificate used to take me 15 to 20 minutes to decipher alone. With AI, five minutes to verify the proposed transcription.

Geneanet complements FamilySearch for France in particular, especially for trees shared by other genealogists. Useful for finding branches a distant cousin has already traced. One essential caution: shared trees contain errors. Always verify the source record, never a copy from another tree.

Step 3: Deciphering old records with AI

This is where the difference is most striking.

Records from before 1800 are in old French or Latin, with abbreviations, variable spellings, and legal formulas that have long since disappeared. A baptism record from 1743, in a provincial parish register, can remain illegible for hours if you're not familiar with the handwriting style.

The method: send the image to Claude (more rigorous than ChatGPT for Latin and complex documents). Ask for a transcription and an explanation of the structure: "Who is the father? The mother? The witnesses? What exactly is the event being recorded?"

Claude identifies roles and names even in ambiguous formulas. It also explains what terms meant in the context of an 18th-century record, which can sometimes change the interpretation of an entire branch.

Atypical regional handwriting or heavily damaged records remain difficult. For complex cases, a guide on deciphering old handwriting with AI covers the advanced techniques.

Step 4: Regional and national archives to fill the gaps

FamilySearch and Geneanet have gaps. Some parish registers haven't been digitized; others are only accessible through regional archive portals. In France, each department runs its own portal. For UK researchers, The National Archives and county record offices play a similar role. For US researchers, state archives and the National Archives hold equivalent collections.

AI is useful here for formulating the right queries.

A real example: I knew an ancestor was born around 1820 in a village in the Haute-Loire region. I asked ChatGPT: "How do I search for the birth record of someone born around 1820 in Haute-Loire, France? What registers exist, and where are they available online?" It pointed me to pre-1792 parish registers on the regional archive portal, the corresponding FamilySearch collections, and the fact that the commune had changed its name in 1850, which explained why my previous searches had found nothing.

Step 5: Finding and animating ancestor portraits

This was the step I had underestimated.

Among the 47 ancestors found, three had photos in digitized archives or in family boxes. Conscription photos from the early 20th century are a goldmine in France: military enlistment registers sometimes include front-facing portraits alongside physical descriptions. Similar resources exist in many countries through military records and census photographs.

For the two photos I recovered, I used Incarn. One (my great-great-grandfather in his conscript uniform, 1908, cracked sepia) produced an animation of a few seconds where he turns his head slightly.

My mother, who had only known this man through a family anecdote (he had refused to flee during the war and stayed to work his forge), watched the video for twenty minutes. She says he looks like my uncle Bernard.

More than 12,000 photos have already been animated on Incarn: grandparents, soldiers, wedding portraits. The animation is powered by Seedance 1.5 Pro (a video AI model by BytePlus), rendered in under 2 minutes. The trial is free (1 credit offered at sign-up), then €1.99 per animation.

AI tool comparison for each step

Step	Recommended tool	Free?	Best for
Structuring interview notes	ChatGPT	Yes (basic version)	Organization and clarity
Finding civil records	FamilySearch	Yes	19th century, France and beyond
Transcribing a handwritten record	ChatGPT	Yes	19th-century cursive
Transcribing a record in Latin	Claude	Yes (usage limits)	Complex documents, old French
Cross-referencing existing trees	Geneanet	Free (consultation)	Already-traced branches
Animating an ancestor portrait	Incarn	1 free credit, then €1.99	Portraits, family photos

How long does this method take?

Honestly, it depends on the complexity of your tree and the archives available.

For my case (Corrèze, 19th century, well-documented):

• Interviews and initial notes: 4 hours
• Searching and transcribing records with AI: 8 to 10 hours over 3 weeks, in one-hour sessions
• Verifying and documenting sources: 3 hours

Total: roughly 15 hours of work for 47 ancestors across 5 generations. I estimated 35 to 40 hours without AI for the same result.

AI doesn't do the work for you. It reduces the time spent deciphering manuscripts and figuring out where to look, which often accounts for 40 to 50% of the total time in active research.

If you have an old photo in a drawer and want to start there, Incarn is accessible without prior registration for the first trial.

Frequently asked questions

Can AI build my family tree automatically?

No. Current AI tools (ChatGPT, Claude, Gemini) do not connect to archive databases. They analyze what you bring to them. The research remains human; the transcription and analysis, less so.

Which AI tool should I start with?

ChatGPT (free version) covers 80% of common needs. Claude is more rigorous for Latin texts and complex documents.

Is FamilySearch really free?

Yes. Basic consultation is free without a subscription. Some advanced features require an account, which is also free. It's the logical starting point for 19th-century archives, and a strong resource for many other countries as well.

I have records in Latin. Can AI translate them?

Yes, with caveats. Claude and ChatGPT correctly decipher standard ecclesiastical Latin from baptism, marriage, and burial records. Very specific abbreviations or atypical regional handwriting cause more problems. Always verify uncertain passages with a secondary source or a professional genealogist.