Somewhere in a shoebox, a filing cabinet, or a folder of scanned files on your computer, there is probably a document from your family's past that you have never actually read.
Maybe it is a ship manifest with your great grandmother's name buried in a column of cramped handwriting. Maybe it is a baptism record written in Latin by a priest in 1887. Maybe it is a letter your grandfather sent home from a war in a language nobody in your family speaks anymore.
These documents are often the closest thing we have to hearing directly from the people who came before us, yet most of them stay unread because reading them takes skills most of us never picked up.
Paleography, the study of old handwriting, is not a class most family historians take.
Neither is nineteenth century church Latin or the old German script that shows up in half the records from Central Europe.
AI can help close that gap.
Used well, it can save you hours of squinting at a screen and guessing at letters. Used carelessly, it can hand you confident sounding text that is simply wrong.
This guide gives you a set of AI prompts for genealogy records that are built to get you the first outcome and steer you clear of the second, along with the steps to check the AI's work before you write anything down as fact.
What AI Can Actually Do With Old Family Documents
Modern AI models, the kind built into tools like ChatGPT, Claude, and Gemini, can read an image of a page and turn it into typed text.
They can also translate that text from one language to another in the same step. That is a real capability, and it works far better today than it did even two years ago. It is worth being honest about where it still struggles, though. These models learned mostly from typed text and clear modern handwriting, so a well printed record from a government office in the 1950s is usually easy for them. A handwritten parish register from 1820, written in a script nobody teaches anymore, in a language the model has seen less of during training, is a much harder task.
The model can also do something more troubling than simply getting a word wrong. When it is not sure what a letter or word says, it will sometimes fill the gap with something that looks plausible instead of telling you it does not know. That is not the model lying to you on purpose. It is just how these systems are built, and it means every transcription and translation needs a human check before it becomes part of your family record.
Getting the Document Ready Before You Ask AI Anything
The quality of what you get out of an AI prompt depends heavily on the quality of the image you put in.
A blurry photo taken at an angle, with a shadow across half the page, gives the model very little to work with.
Before you write a single prompt, take a few minutes to photograph or scan the document properly. Use natural light or an even indoor light source, and avoid the harsh glare of a flash bouncing off old paper. Hold the camera directly above the page rather than at an angle, and get close enough that the text fills most of the frame without cutting off any edges.
If you are working from a multi page record, photograph each page on its own rather than trying to capture two pages in one shot.
A little effort here saves you from a frustrating back and forth with the AI later.
A Quick Note on File Size and Format
Keep your images at a high resolution and avoid heavy compression, since compression is what turns fine pen strokes into a blur of pixels.
JPEG and PNG both work fine for this. If you are working from a scanned PDF that contains many pages, export the specific page you need as its own image file rather than uploading the entire document at once.
Feeding a model one page at a time, with your full attention on that page, produces noticeably better results than asking it to work through fifty pages in a single request.
Prompts for Transcribing Handwritten Records
Transcription is the process of turning handwriting into typed text in the same language it was written in.
This is usually the first step, even for documents you eventually want translated, because getting an accurate transcription gives you a much better foundation for translation later.
Here are three prompts built for three common situations.
The Basic Transcription Prompt
Use this for letters, diaries, and other handwritten documents where the handwriting is reasonably legible, even if it takes effort to read.
"I am attaching an image of a handwritten family document. Please transcribe the text exactly as written, preserving original spelling, punctuation, and line breaks. Do not correct grammar or spelling. If a word or phrase is illegible, write [illegible] rather than guessing. After the transcription, list any words you were uncertain about and explain why."
The Faded or Difficult Handwriting Prompt
Some documents have faded ink, torn edges, or handwriting so cramped it is hard to separate one word from the next. For these, you want the AI to be upfront about its uncertainty rather than smoothing over the gaps.
"This document has faded or difficult handwriting. Please transcribe it as carefully as possible, and for any word you are not fully confident about, provide your best guess in brackets followed by a question mark, like [guess?]. Do not present an uncertain reading as if it were certain. Break your transcription into short sections matching the original line breaks so I can compare it against the image line by line."
The Table and Ledger Prompt
Census records, ship manifests, and parish registers are usually laid out as tables with columns for names, ages, occupations, and dates. A general transcription prompt can lose that structure. This prompt asks the AI to keep the columns intact.
"This image shows a record laid out in columns, such as a census sheet or ship manifest. Please transcribe it as a markdown table, keeping every column in its original order and every row in its original order. If a cell is blank in the original, leave it blank in the table rather than skipping the row. If a cell is illegible, mark it with a question mark. Include the column headers from the original document at the top of the table."
Prompts for Translating Old Documents
Translation is a separate skill from transcription, and plenty of family documents need both. A record might be written in perfectly legible handwriting in a language you simply do not read, or it might combine hard to read handwriting with an old form of a language that has since changed. The prompts below are built around three situations that come up constantly in genealogy research.
Translating Old Church Latin
Catholic parish registers, and many Lutheran ones, were often kept in Latin well into the twentieth century in parts of Europe. The good part is that priests generally used the same handful of standard phrases over and over for baptisms, marriages, and deaths, which means the language is more repetitive and predictable than it looks at first glance.
"This is a Latin entry from a Catholic or Lutheran church register, likely a baptism, marriage, or death record. Please translate it into plain modern English, and explain any standard Latin church terms or abbreviations you encounter, such as titles for godparents, witnesses, or clergy. Note the type of record this appears to be based on its structure and wording."
Translating German, Polish, and Other Old World Languages
Records from Central and Eastern Europe often shift language depending on who controlled the region at the time a document was written. A birth record from a Polish village might be in German, Russian, or Polish depending on the decade, and the same family line might show up under three different languages across three different records. Tell the AI what you know, or suspect, about the historical context, since that context helps it choose the right translation approach.
"This document is from [region and approximate year, for example rural Poland around 1890]. Please identify the language it is written in, keeping in mind that political control of the region changed over time and could affect which language was used. Translate the full text into plain English, and flag any place names, given names, or surnames that may have alternate spellings in other languages, since that will help me search for related records."
When the Script Itself Is the Problem
Sometimes the issue is not the language at all.
It is the handwriting style. Kurrent and Sutterlin, two old German scripts used across German speaking Europe well into the twentieth century, look nothing like the German alphabet most people learn today. Secretary hand, an older English script, causes the same kind of confusion for English language records from before the 1700s. If your transcription attempts keep coming back wrong even though you can tell the document is otherwise in reasonable condition, the script itself may be the obstacle.
"This handwritten document may be written in an older script style, such as German Kurrent or Sutterlin, or an older English script such as Secretary hand. Please identify which script this appears to be, and transcribe the text into standard modern letters. Explain any letters or letter combinations in this script that commonly get confused with other letters, so I can better read similar documents myself in the future."
Checking the AI's Work Before You Trust It
None of this replaces your own judgment, and treating an AI transcription as a finished, verified record is where genealogy research can go wrong fast.
Compare names and dates against other records you already have for the same person, such as a different census year or a family bible entry. If a name or date looks slightly off from what you expected, go back to the original image and check that specific word yourself rather than accepting the AI's version by default.
It also helps to run the same document through two different AI tools and compare the results side by side. When both tools agree on a tricky word or phrase, you can trust it more. When they disagree, that is your signal to look closer at the original image, or to ask someone in a genealogy forum who has experience reading that particular script or language.
The goal is not to hand the work over to AI completely. The goal is to let it do the slow, mechanical first pass so you can spend your time on the parts that actually require a person, like judgment calls and cross referencing.
Turning Documents Into Family Stories
A transcribed and translated record is not the end goal.
It is a doorway. A name and a date on their own tell you that someone existed and roughly when, but the actual words in a letter, a diary entry, or even the margin notes in a family bible tell you something closer to who that person was. That baptism record in Latin might just confirm a date you already suspected, but the letter written in German that you finally get translated might reveal that your great grandfather wrote home about missing his mother's cooking, or worried about money, or fell in love with a girl his family never approved of.
Those are the details that turn a name on a family tree into an actual person.
That is really what all of this effort is for.
Every hour spent getting a document readable is an hour spent making sure a person's actual words survive instead of getting reduced to a birth year and a death year on a chart.
The people who wrote these documents did not know a stranger, or a great great grandchild, would someday be squinting at their handwriting trying to make out what they said. They just wrote it down because it mattered to them at the time.
Getting it translated and readable again is a way of listening to them, a hundred or two hundred years later than they expected, but listening all the same.
If you find yourself moved by an old letter or a name freshly translated after a century of sitting unreadable, it is worth remembering that the people alive in your family right now have their own stories nobody has written down yet.
Memoracy was built around that exact idea, giving people a simple daily prompt to record their own life in their own words, so future generations are not stuck piecing it together from old documents the way we so often have to.
Sign up and start your first story on Memoracy today.
