More in Family history.
Translate a great-grandfather's Yiddish letters for a family archive with Gemini and Claude
Copy the prompt below and paste it into the AI tool named on this page.
Prompt Recreated prompt
The source didn't share its exact prompt, so we wrote this from what it describes.
This is a scanned handwritten letter in Yiddish, written by my great-grandfather sometime between 1900 and 1926. Translate it into English, keep the date and names exactly as written and mark any words you can't read or aren't sure about instead of guessing.
Try this in:
Sends the Recreated prompt shown above. ChatGPT and Grok open with it filled in, and ChatGPT may send it straight away. Gemini opens empty, so we copy the prompt for you to paste. Every button also copies it.
Opens in a new tab. We don't see what you send.
These links are only a shortcut. We're not affiliated with OpenAI, Google or xAI, and your results may not match the source.
Optional stronger version
Our rewrite with extra checks. The poster didn't test it.
Before you start, ask me 1–3 short questions about the details below and wait for my answers. Once I've answered, confirm the plan in one line and run the whole task start to finish. Goal: Translate a set of old handwritten family letters into English accurately enough to build a family history. Questions to ask me first: - What language and roughly what years are the letters from? - How many letters are there, and are they scanned? - Who wrote them, and who are the main people mentioned? Inputs (fill in or confirm): - Scans of the letters, one at a time: [SCANS] - Language and dates: [LANGUAGE / DATES] - Known family names and places: [NAMES / PLACES] Rules: - Translate line by line, keeping names, dates and places exactly as written. - Mark unclear or unreadable words with [?] instead of filling gaps. - Keep a separate list of words or passages you're unsure about, for a native speaker to check. - Don't add background or story details that aren't in the letter. Checkpoint: Translate one letter first and wait for me to compare it with another translation before doing the rest. If you get stuck: If the handwriting is too hard to read, tell me which lines and ask for a clearer scan. Verify before you finish: Plan to compare each translation with a second tool and a native speaker before treating it as fact. Output: For each letter: the translation, a list of uncertain words and a one-line summary with date and people. Done when: Every letter has a checked translation I can trust enough to share with family.
AI tools
Outcome
On a trip to a storage unit, Paul Levine, 61, of Vancouver found over 100 letters written by great-grandfather Sam Diamondstein between 1900 and 1926, including letters to Dora during their courtship written in Yiddish. Levine fed one letter into Google Gemini and within moments had a translation: in 1901, Sam had traveled to Montreal in steerage to seek a fortune. With friend Fernando Medrano, Levine built an online family archive. Because the AI 'produced what looked like clear narrative text,' they assumed it had made some things up, so they compared three independent translations from Gemini, Claude and the tool Transkribus. They also had Yiddish speakers on Facebook double-check them and worked with a local historian. Claude also helped draw out larger themes and shape the narrative.
Evidence: Article or interviewA newsletter, journalist or interview account of what happened.
Source
- Posted
- by Denise Ryan (Vancouver Sun)
- Last checked
- Method:
- How we verify
https://vancouversun.com/news/letters-vancouver-storage-unit-ai-history
Context: Levine and Medrano treated the AI translations as drafts and checked them with Transkribus, Yiddish speakers and a local historian.
More in Family history
Transcribe and date 150 old family letters for an archive with ChatGPT
A co-chair of the California Genealogical Society's Manuscripts Committee scanned a collection of about 150 handwritten letters exchanged among Midwest family members and friends between about 1885 and 1925. The co-chair uploaded JPG images to ChatGPT with clear rules: keep the original line breaks, preserve the spelling exactly, flag unsure words and never invent missing text. The transcriptions weren't perfect but sped up the work, and every one was reviewed against the original letter. ChatGPT also explained references like Flinch cards and helped date undated letters, noting that Iowa was debating women's suffrage in 1919. Asked to find every letter mentioning an infant death, it showed some referred to an earlier loss around 1898, not the 1903 death the co-chair had assumed; genealogical records confirmed it.
ResearchFamily historyChatGPTWorkedRecreated promptEvidence: Described by posterView CaseSeparate land descriptions in an 1836 dower record with ChatGPT
Working from an 1836 Jefferson County, Indiana, probate record, the Heartland Genealogy blogger asked ChatGPT to pull out the land description. It separated Thomas D. Crawford's 113-acre tract from the 43¾ acres set aside as the dower of Crawford's widow, Jane. While drafting a blog post from the record, ChatGPT also found that this Jane appeared to be Jane Todd Crawford, remembered in medical history for surviving an operation performed by Dr. Ephraim McDowell. The writer treats it as a lead to keep researching: 'AI found the connection. It didn't finish the research.'
ResearchFamily historyChatGPTWorkedRecreated promptEvidence: Described by posterView CaseTrace a family tree from half-remembered details with ChatGPT (partial)
Ryan Aubrey didn't know where the family came from, or even the grandparents' middle names. Aubrey turned ChatGPT's effort level up as high as it would go and shared everything known, mistakes included. On the father's side it quickly found a grandfather's obituary (a high school state championship and a basketball scholarship to the University of Tampa) and went through FamilySearch and Find a Grave to suggest a colonial Virginia line of 'Awbrey's' with a possible, unconfirmed Welsh origin. On the mother's side the line got lost after the grandparents in West Virginia, so Aubrey says the next step is to 'go analog'.
ResearchFamily historyChatGPTPartialRecreated promptEvidence: Described by posterView CaseBreak a six-year family history brick wall with help from ChatGPT
Two cousins spent six years trying to prove a family story that the Jacksons descended from the Reade family of colonial Virginia. ChatGPT couldn't hand them the answer, but it helped them spot patterns in their documents, surfaced lesser-known resources and suggested they might be 'chasing the wrong Francis'. A ChatGPT-assisted search led them to Library of Virginia records on a Captain Francis Read whose ship was captured by privateers, and a Reade family Bible at the library listed the children of Gwyn and Dorothy Reade. The next morning, cousin Kelly had pieced it together: 'I've been up since 4 a.m. I figured it out.'
ResearchFamily historyChatGPTWorkedRecreated promptEvidence: Described by posterView CaseExtend a family tree two generations with Claude Code and browser agents
Working from the family's archive scans, Claude Code with memory and the Claude browser plugin cross-checked church books, ship manifests and naturalization records, built a 300-page family wiki, and found two more generations in a day; the family is now checking it for errors.
Poster says:300-page family wiki
ResearchFamily historyClaude CodeClaude in ChromePerplexity CometWorkedRecreated promptEvidence: Shown on videoView Case
Related cases
Transcribe and date 150 old family letters for an archive with ChatGPT
A co-chair of the California Genealogical Society's Manuscripts Committee scanned a collection of about 150 handwritten letters exchanged among Midwest family members and friends between about 1885 and 1925. The co-chair uploaded JPG images to ChatGPT with clear rules: keep the original line breaks, preserve the spelling exactly, flag unsure words and never invent missing text. The transcriptions weren't perfect but sped up the work, and every one was reviewed against the original letter. ChatGPT also explained references like Flinch cards and helped date undated letters, noting that Iowa was debating women's suffrage in 1919. Asked to find every letter mentioning an infant death, it showed some referred to an earlier loss around 1898, not the 1903 death the co-chair had assumed; genealogical records confirmed it.
ResearchFamily historyChatGPTWorkedRecreated promptEvidence: Described by posterView CaseSeparate land descriptions in an 1836 dower record with ChatGPT
Working from an 1836 Jefferson County, Indiana, probate record, the Heartland Genealogy blogger asked ChatGPT to pull out the land description. It separated Thomas D. Crawford's 113-acre tract from the 43¾ acres set aside as the dower of Crawford's widow, Jane. While drafting a blog post from the record, ChatGPT also found that this Jane appeared to be Jane Todd Crawford, remembered in medical history for surviving an operation performed by Dr. Ephraim McDowell. The writer treats it as a lead to keep researching: 'AI found the connection. It didn't finish the research.'
ResearchFamily historyChatGPTWorkedRecreated promptEvidence: Described by posterView CaseTrace a family tree from half-remembered details with ChatGPT (partial)
Ryan Aubrey didn't know where the family came from, or even the grandparents' middle names. Aubrey turned ChatGPT's effort level up as high as it would go and shared everything known, mistakes included. On the father's side it quickly found a grandfather's obituary (a high school state championship and a basketball scholarship to the University of Tampa) and went through FamilySearch and Find a Grave to suggest a colonial Virginia line of 'Awbrey's' with a possible, unconfirmed Welsh origin. On the mother's side the line got lost after the grandparents in West Virginia, so Aubrey says the next step is to 'go analog'.
ResearchFamily historyChatGPTPartialRecreated promptEvidence: Described by posterView Case
