๐ŸŒณ Family History & Genealogy
Text size:
โ€น Back to the guide

Techniques: ways to use AI

Each one is a quick, copy-and-paste recipe. Pick whatever you need โ€” you don't have to read them in order.

Most of this works with the free AI assistants you may already have open in another tab. One idea to keep in mind:

AI is a research assistant You are the researcher The record is the source
โš ๏ธ The one rule that matters most AI predicts what is likely. It does not verify what is true. It can read a record in seconds โ€” and it can also invent a name, a date, or a whole "source" with total confidence. Treat everything it gives you as a helpful lead, then check it against the original record. Never cite the AI. Cite the record.

Pick something to try

New to this? Start with one of the first two โ€” they surprise people the most.

๐ŸŽ™๏ธ Turn your family story into a podcast

This is the one that makes people gasp. A free Google tool called Gemini Notebook (you may have heard of it as "NotebookLM" โ€” Google renamed it in July 2026; it's the same tool) can take something you've written โ€” a life history, a stack of notes, a few letters โ€” and create a friendly audio "podcast" where two voices talk through it, by name, like hosts of a little radio show about your family.

It's wonderful for sharing with relatives who would never sit down and read a 50-page history, but would happily listen in the car.

  1. Go to notebook.google.com and sign in with a free Google account.
  2. Create a new notebook and add your sources โ€” paste in your text, or upload documents.
  3. Choose the Audio Overview option and let it generate. In a few minutes you'll have a podcast about your ancestors.
Remember The hosts sound confident and may add little flourishes for color. It's built from what you gave it โ€” so the facts are only as good as your notes, and a charming aside is not a documented fact.

๐Ÿ“ท Ask your phone about an old photo (and bring it to life)

Two different, almost magical things you can do with old pictures:

Ask what you're looking at

Open Google Lens (built into the Google app and Google Photos) or any AI chat app that accepts pictures, point it at a photo, and ask. It can read a faded inscription, identify a military medal or uniform, or guess a date from the clothing and car in the background.

Please look at this photograph and estimate when it was taken based on clothing styles, hairstyles, and the photographic format. Explain your reasoning and give a date range.

Repair, color, and sharpen

You don't need a subscription to start. A few that are genuinely free:

MyHeritage is still the easiest one-stop for colorize, enhance, and animation (look for "In Color," "Photo Enhancer," "Deep Nostalgia," and "LiveMemory") โ€” a few free tries, then a subscription.

Feeling more real is not the same as being more accurate The color of a dress or the way a face moves is the computer's educated guess, not a fact from a record. Enjoy these as a way to feel closer to someone โ€” not as evidence.

โœ๏ธ Read handwriting you can't make out

Faded ink, old-fashioned cursive, a letter that's been folded for 80 years โ€” AI is remarkably good at reading it. Take a clear photo, upload it to a chat assistant like Claude or ChatGPT, and ask:

Please transcribe this handwritten document exactly as written. Note any words you are uncertain about with a question mark in brackets. Do not correct spelling or modernize the language.

Got a whole diary or journal โ€” dozens or hundreds of pages? The simplest path is still the one above: photograph the pages and feed them to a chat assistant a few at a time. There's also a specialized tool called Transkribus built for large batches โ€” but be warned, it's a power tool with a real learning curve and limited free credits. If you're not in the mood to wrestle with new software, stick with the photo-and-ask method.

Always open the original too Read the AI's transcription with the real image next to it. Where it put a [?], that's your cue to look closely โ€” those are often the names and dates that matter most.

๐ŸŒ Read a record in another language

So many of our records are in Latin, German, Italian, Swedish, or another language we don't read. A chat assistant can read the handwriting and translate it and explain what the old terms mean โ€” all at once.

Please read this handwritten record and tell me everything you can โ€” names, dates of birth, baptism or marriage, parents' names, places, and any other details you can make out. Then give me a plain-English translation, and note anything you are unsure about.
A tip from the parish books Indexes are often wrong โ€” a clerk's "Jose" can turn out to be "Jura" (George) once you read the actual entry. If a name in an index never leads anywhere, have AI read the original page itself. The answer is often hiding in plain sight.

One good habit: for foreign records, transcribe first in the original language, then translate โ€” don't translate straight from the photo. For typed/printed pages, DeepL (free) is excellent on European languages. Already on FamilySearch? Its built-in Full-Text Search uses AI to read unindexed handwriting โ€” worth a look before you go elsewhere.

๐Ÿ—’๏ธ Turn an obituary into a timeline and a story

An obituary is packed with names, dates, and places all jumbled together. Watch what happens when you hand one to AI and walk it through a few steps. Paste in the obituary, then ask these one at a time:

Please read this obituary and create a chronological timeline of the key events in this person's life. Include every date, place, and person mentioned.
Now write a short biographical paragraph based on this timeline. Write it as a story, not a list.
Based on what you know about this person, what additional records might exist that I haven't mentioned? What should I search for next?

In about five minutes you've gone from a clipping to a timeline, a readable story, and a research plan for what to look for next.

๐Ÿงฑ Get unstuck on a brick wall

When you've hit a dead end, AI makes a tireless brainstorming partner. The trick is to tell it what you already know and what you've already tried. Fill in the blanks and paste this in:

I'm trying to learn more about [PERSON], born about [YEAR] in [PLACE]. Here is what I know: [3โ€“6 facts]. Here is what I've already tried: [searches/records]. Some things that may matter: [religion, language, migration, the county or parish]. Please suggest the record types most likely to help, where they might be kept, and 5 concrete next searches. Also tell me what could explain why I'm not finding the record I expect.
Ideas, not answers Everything it suggests is a lead to go verify in real records. If it names a specific document or source, look it up yourself โ€” don't assume it exists just because the answer sounded sure.

๐Ÿ”Ž Understand old words, jobs & places

Records are full of words that have faded from everyday use. Just ask, in plain English:

What did the occupation "cordwainer" mean, and what would that person's daily work have involved in the 1880s?
This address appears in a 1910 city directory: [address]. What neighborhood is that in today, is the building still there, and what landmarks were nearby?
What was daily life like for [a coal miner / a farm wife / a shopkeeper] in [place] during [time]? What challenges would they have faced?

๐Ÿ“– Turn a few facts into a life story

Maybe all you have are a handful of dates and a couple of family sayings. AI can weave them into a warm paragraph that's actually worth handing down โ€” as long as you keep it honest about what's known versus guessed.

You don't need to buy a "memoir kit" A free chat assistant is itself the most capable story writer there is โ€” the same kind of AI the paid services charge $60 to $200 a year for. Ask it to write the story, then ask it to try again warmer, shorter, or in the first person until one version sounds like your family. (Many people find Claude writes the most natural prose.) Just feed it only facts you've checked, and never let it invent dates or dialogue.
Here are the facts I know about my grandmother: [list them]. Please write a short, warm life-story paragraph using only these facts. Do not invent events. Where you add background about the time and place, keep it clearly general, not specific claims about her.
Somewhere to keep it A story is only a gift if it's saved. The Legacy Vault has a Life History Journal made for exactly this โ€” paste your finished story straight into it, and it stays private on your own device. When you're ready to make it a book or recording, see Preserving & passing it on.

๐Ÿ“ Capture your whole life story

The hardest part of writing your own story is the blank page โ€” so don't start there. Start with a question. Here are a few from the Life History Journal in your Legacy Vault, just to get you talking:

Answer one out loud or in rough notes โ€” don't worry about writing it well. Then let AI tidy it into something worth keeping. Paste your answer into this:

Here are my rough notes answering the question "[paste the question]": [your notes, however messy]. Please turn this into a warm, first-person paragraph using only what I said. Don't invent details or add events I didn't mention, and keep my voice.
Then keep it somewhere safe Save each finished story in the Life History Journal โ€” it stays private, on your own device. There are many more prompts there than the handful above. Once you've gathered a few, Preserving & passing it on shows how to turn them into a book or a recording for your family.

๐Ÿงช Test a tool before you trust it

Ancestry, MyHeritage, and FamilySearch are all racing to add AI features, and more arrive every few months. Before you lean on one โ€” or recommend it to a cousin โ€” spend ten minutes finding out what it's actually doing.

The test costs nothing and never fails to teach you something:

  1. Pick an item you already know the answer for โ€” a photo you can date, or a clipping with a sourced publication date already attached.
  2. Run the AI feature on it and read the reasoning, not just the answer.
  3. Compare it to what you know.
  4. Then run it a second time and see whether you get the same answer.

What that looked like when I tried it

In August 2026 I ran Ancestry's Photo Insights โ€” a paid feature that analyzes a photo or document and suggests a date, a place, and a description โ€” on two things from my own gallery. Three lessons came out of it, and all three apply to any tool of this kind.

1. Some of the "findings" were my own data handed back to me.

I started with a studio work portrait of my father, Lowell Wesley Robbins. Photo Insights reported the location as "United States (likely Utah)" โ€” and its own reasoning told me why: "the metadata links the image to Lowell Wesley Robbins (born in Ogden, Utah; died in Salt Lake City, Utah)."

It hadn't found Utah in the photograph. It found Utah in my family tree, because I put it there โ€” and I had tagged that very photo about five minutes earlier, in another part of Ancestry that had prompted me to identify people in my pictures. Two lines later it admitted there were "no visible landmarks or signage" to go on. So it told me Utah because I had just told it Utah.

The date panel on that same portrait, though, was genuine analysis: the narrow tie with its bold geometric pattern, the cut of the suit, the color process, the airbrushed studio background. That's real work, and it's work I couldn't do myself without a reference book and an afternoon. Same feature, same photo, two panels โ€” one reasoning from the image, one quietly reflecting me back at myself. The only reason I could tell them apart is that it showed its reasoning.

2. It was impressive on the hard part and wrong on the easy part.

Next I tried a newspaper clipping saved to my great-great-grandfather Daniel Spencer โ€” "Last Pioneers Entered Valley In Fall Of 1847," listing the pioneer companies and their numbers. Here it made a distinction that trips up plenty of experienced researchers: "the content date (1847) is not the publication date," and it backed that up from the halftone printing, which showed the image was a later reproduction rather than an 1840s original. Genuinely sharp โ€” that's the mistake I've seen people make over and over with clippings.

Then it transcribed the total in the article as 2,005. Enlarge the clipping and it plainly reads 2,095. One digit โ€” in exactly the kind of number someone copies straight into their tree without a second look.

3. I ran it twice, and the answer changed.

Same clipping, minutes later, nothing altered โ€” and a different date range came back:

I know the real answer: 21 July 1939, Salt Lake City, from the newspapers.com source already attached to the clipping. And Ancestry knew it too โ€” its own "Compare details" panel was displaying 21 Jul 1939 in a little table directly beneath the suggestion. So the tool showed me the correct date and its own conflicting guess side by side, with a Save date button sitting right there, ready to write the guess over the fact.

โš ๏ธ The rule that falls out of all this Never save a suggestion that contradicts something you already have a source for โ€” no matter how reasonable the reasoning sounds. And if two runs disagree with each other, you've learned something important: the tool is estimating, not knowing.
๐Ÿ’ก The part it couldn't do โ€” and I could That clipping was published on 21 July in Salt Lake City. Pioneer Day is 24 July. It's a Pioneer Day feature โ€” and knowing that narrows a 108-year range down to one week of the calendar. Photo Insights correctly identified it as a "commemorative feature" and still never made the connection, because it isn't from here and I am. That kind of local knowledge is what you bring that no tool has.

Four questions to ask of any AI feature

โœ… Free tools make a good control test Upload the same photo to a free assistant like ChatGPT, Claude, or Gemini. They have no access to your tree, so they can only work from the picture itself โ€” which shows you exactly how much of a paid tool's answer came from the image and how much came from your own data.
Please look at this photograph or document and estimate when and where it was made. Use only what you can actually see in the image โ€” do not guess from anything else. List the specific visual clues you are using, tell me how confident you are in each, and say plainly what you cannot determine from the image alone.

Staying safe and accurate

โ€น Back to the guide