More in Family history.
Translate old Norwegian and German records with Gemini, Claude and ChatGPT
PartialEvidence: Screenshots or logsPosted Last checked
Gemini did the best job translating Kitty Cooper's old records, but Claude gave a partial and ChatGPT made stuff up.
How to use this
- Tap Copy prompt.
- Open Gemini.
- Paste it into the message box.
- Replace anything in [CAPITAL LETTERS IN BRACKETS] with your own details.
- Send it, then read the answer before acting on it.
New to this? What's a prompt?
A prompt is the message you send to the AI tool. Copy the one on this page, put your own details where it asks, and read the answer before you act on it.
Prompt
Improved prompt
Our fuller version of the source's prompt, with fill-in details and rules, or one written to avoid what went wrong. Not tested by the poster.
Our fuller version of the person's prompt. They didn't use this version, and we haven't tested 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: Transcribe and translate an old handwritten record exactly, and mark anything you can't read instead of guessing. Questions to ask me first: - What language and roughly what year is the record? - What kind of record is it (baptism, marriage, census, letter)? - Do you want a word-for-word translation, a summary, or both? Inputs (fill in or confirm): - Record image: [UPLOAD PHOTO OR SCAN] - Language and year: [LANGUAGE / YEAR] - Names I already know: [NAMES] Rules: - First transcribe the original text line by line, then translate it. - Mark unreadable words as [unreadable] and uncertain ones with a question mark. - Never fill in names, dates or places that you can't read. - Keep the original order and layout; don't turn it into bullet points. Checkpoint: Show the transcription first so I can compare it with the image before you translate. If you get stuck: If the image is too poor to read, say so and tell me which parts you could read. Verify before you finish: Check that every name and date in the translation appears in the transcription. Output: A line-by-line transcription, then a word-for-word translation, then a list of uncertain words. Done when: I have a translation I can check against the record, with every guess marked.
Customize this prompt
Answer these questions and we'll put your answers into the prompt for you. You can leave any of them empty.
Your file stays on your device. Nothing is uploaded to TriedPrompts.
Your prompt
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: Transcribe and translate an old handwritten record exactly, and mark anything you can't read instead of guessing. Questions to ask me first: - What language and roughly what year is the record? - What kind of record is it (baptism, marriage, census, letter)? - Do you want a word-for-word translation, a summary, or both? Inputs (fill in or confirm): - Record image: [UPLOAD PHOTO OR SCAN] - Language and year: [LANGUAGE / YEAR] - Names I already know: [NAMES] Rules: - First transcribe the original text line by line, then translate it. - Mark unreadable words as [unreadable] and uncertain ones with a question mark. - Never fill in names, dates or places that you can't read. - Keep the original order and layout; don't turn it into bullet points. Checkpoint: Show the transcription first so I can compare it with the image before you translate. If you get stuck: If the image is too poor to read, say so and tell me which parts you could read. Verify before you finish: Check that every name and date in the translation appears in the transcription. Output: A line-by-line transcription, then a word-for-word translation, then a list of uncertain words. Done when: I have a translation I can check against the record, with every guess marked.
Scroll to see the whole prompt. Your answers show in green.
3 blanks are still empty: Record image; Language and year; Names I already know. They'll stay in brackets in your prompt, so the chat tool may ask you about them.
What you type stays in this browser tab. We don't send it or save it, and it's gone when you leave the page. If you use a "Try this in" button, your answers go to that tool as part of the prompt.
Try this in:
This prompt has blanks to fill in, like [UPLOAD PHOTO OR SCAN]. We'll copy it instead of filling it in, so you can edit it before you send.
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, Anthropic or xAI, and your results may not match the source.
The prompt they actually used
The exact prompt from the source.
translate
AI tools
Outcome
Genealogist Kitty Cooper uploaded a poor scan of a Norwegian record to Claude, ChatGPT and Gemini and asked each to translate it. Claude said it was hard to read and gave only a partial translation. ChatGPT made up something that sounded totally plausible but didn't seem anything like the original Norwegian. Gemini did well, though it organized the result into bullet items. Cooper's new method is to upload the same document to Claude and Gemini and merge the results. Cooper then tried handwritten German records, where Gemini did the best job; Claude, puzzlingly, decided one person was English.
Evidence: Screenshots or logsThe source shows screenshots, a transcript or the AI's output.
Known limitations
PartialThe source says it only partly worked, or had a significant problem a person had to deal with.
What went wrong: Gemini did well, Claude gave only a partial translation, and ChatGPT made up a plausible but wrong one.
Source
- Posted
- by Kitty Cooper (Kitty Cooper's Blog)
- Last checked
- Method:
- How we verify
https://blog.kittycooper.com/2026/05/further-experiments-with-ai-and-genealogical-documents/
Context: Cooper's original prompt was just "translate," per a reply in the comments. Compare AI translations against the record.
More in Family history
Find leads on an 1800s Edinburgh ancestor's parents with ChatGPT
ChatGPT gave Janice Wellard new places to search for an ancestor's parents, but it read burgess roll dates wrong and repeated itself.
PartialChatGPTEvidence: Described by posterView CaseMatch census tick marks to family names with ChatGPT
ChatGPT matched 1790 and 1800 census tick marks to the Reed family's children for a genealogy blogger in a few prompts.
WorkedChatGPTEvidence: Screenshots or logsView CaseCatch a wrong birthplace in a family tree with a NotebookLM Audio Overview
A NotebookLM Audio Overview made a genealogist notice that an ancestor's recorded birthplace didn't match the records.
WorkedNotebookLMEvidence: Described by posterBeginner-friendlyView CaseSearch 40+ sources on a brick wall ancestor's FAN Club with NotebookLM
NotebookLM searched 40+ sources at once with citations and helped Lisa Rex confirm a migration pattern from Ohio to Illinois.
WorkedNotebookLMEvidence: Described by posterBeginner-friendlyView CaseTranscribe an 1829 handwritten deed with Claude and check it against a manual copy
Claude 3.5 Sonnet transcribed an 1829 deed almost perfectly for genealogist Nicole Dyer, but misread the key surname Peerson.
WorkedClaudeEvidence: Screenshots or logsBeginner-friendlyView CaseCompare four AIs transcribing a 1700s will found with FamilySearch full-text search
Dana Leeds tested Copilot, Gemini, ChatGPT and Claude on a 1700s will. Claude was most accurate, but each made spelling mistakes.
PartialClaude +3Evidence: Screenshots or logsView Case
Similar cases
Break a six-year family history brick wall with help from ChatGPT
ChatGPT couldn't answer outright, but it spotted patterns and a wrong lead that helped two cousins end a six-year search.
WorkedChatGPTEvidence: Described by posterView CaseSeparate land descriptions in an 1836 dower record with ChatGPT
ChatGPT separated a 113-acre tract from a 43¾-acre dower in an 1836 record and linked the widow to a figure in medical history.
WorkedChatGPTEvidence: Described by posterView CaseTrace a family tree from half-remembered details with ChatGPT (partial)
From vague details, ChatGPT found a grandfather's obituary and a possible Virginia line, but the other side stalled in West Virginia.
PartialChatGPTEvidence: Described by posterView CaseTranscribe and date 150 old family letters for an archive with ChatGPT
A genealogy volunteer used ChatGPT to transcribe about 150 letters from 1885 to 1925, then date them from clues in the text.
WorkedChatGPTEvidence: Described by posterView CaseTranslate a great-grandfather's Yiddish letters for a family archive with Gemini and Claude
A man used Gemini and Claude to translate Yiddish family letters from 1900 to 1926, checking for made-up text with native speakers.
WorkedGemini +1Evidence: Article or interviewView Case

