This is the third version of this article. I got feedback from great minds in the Facebook Group “Genealogy and Artificial Intelligence”, for which I am very grateful.

The Cliff Notes:

The guide explains that AI is a powerful tool for genealogy, especially for transcribing handwritten records and generating research ideas. However, we should use it responsibly, taking into account different levels of risk.

Key Principles:

• Risk-Tiered Sharing: The article discusses the distinction between public, non-sensitive records of deceased individuals (low risk) and information that identifies living people (high risk). Public historical records are usually safe to analyze with AI.
• The “Temporary Mode” Tool: Genealogists should use ChatGPT’s Temporary Mode when handling sensitive historical data, such as records concerning slavery or institutionalization, or when conducting research they intend to copyright or publish.
• Privacy Pitfalls: The post warns not to use the “Share” feature, pointing to a July 2025 incident when shared conversations appeared in search engine results. Treat “Share” as if you are publishing to the internet.
• Transcription Benefits: AI is described as a breakthrough for transcribing hard-to-read 19th-century handwriting, like Spanish colonial records or German surnames. It can save time and money compared to doing it by hand.

The “Golden Rules” for Genealogists:

• Protect the Living: Never post current addresses, phone numbers, or details of minor children.
• Redact or Use Temporary Mode: Use privacy settings for unpublished or sensitive historical content.
• Verify Everything: Consistent with genealogical standards, researchers should fact-check AI outputs and monitor changes to AI companies’ privacy policies.

In summary, AI can greatly help us understand history, but it should be used carefully and ethically, just like any other genealogical source.

Why I’m Writing This

I love using AI tools for genealogy research. Since late 2024, I’ve discovered that ChatGPT and other chatbots help me brainstorm research strategies, understand confusing historical documents, and even build websites for my genealogy projects. AI is genuinely useful.
But I’ve also learned that using AI tools wisely means understanding how your information is processed.

This guide is not an argument against using AI on historical records. In fact, using AI to read, transcribe, summarize, and analyze public, non-sensitive records about deceased people is often a reasonable and powerful use case—and it’s already widely practiced across the genealogy ecosystem.

The real goal here is narrower and practical: protect living people’s privacy, protect your unpublished work, and avoid accidental public sharing.

Important: I’ve included sources for every major claim in this article.
As genealogists, you know how to evaluate evidence. You should do the same here.
Please fact-check what I say by reading the official sources from OpenAI and the journalists who covered the 2025 sharing incident. The sources are at the end of this guide, and I encourage you to verify everything yourself.

---

The Real Situation: What Happens to Your Data?

When you type something into ChatGPT, here’s what matters:

OpenAI (the company behind ChatGPT) can see what you write.
OpenAI processes the input you submit to generate responses. Depending on your settings and the product mode you use, some content may also be used to improve models.
The key is: assume anything you paste could be retained or reviewed unless you intentionally choose privacy-preserving settings.

You have control over how much of your data OpenAI keeps. If you use Temporary Mode (available on ChatGPT Plus and Free plans), your conversation isn’t saved for training purposes. However, OpenAI still processes it to provide answers.

Think of it like this: If you ask a question at your local library, the librarian can see what you’re asking. They might remember it to improve their reference services. But if you use a “request a service anonymously” option, they process your question, help you, and don’t retain notes about it afterward. OpenAI’s Temporary Mode works similarly—they help you, but they don’t keep records.

---

The Simple Rule for Genealogists

Use a risk-tier mindset:

1. Never paste data about living people (addresses, phone numbers, emails, full DOBs, minor children, private family stories) unless you have a clear consent-based reason and you’re using privacy-preserving settings.
2. Public records about deceased people are usually fine to analyze with AI—especially when they’re already broadly accessible (e.g., older census pages, historical newspapers, probate abstracts).
   Don’t treat “public” as automatically “dangerous.”
3. Use Temporary Mode when you’re pasting anything that is:
   • about living people, or
   • your unpublished research/transcriptions, or
   • sensitive even if historical (adoption, assault, medical details, slavery, etc.), or
   • restricted by website/archive terms or copyright.

Why this matters: Genealogy already battles “privacy” being used as a blanket excuse to restrict access to historical records. We should protect the living without accidentally arguing for broad restrictions on public historical documents.

---

What You Can Safely Ask ChatGPT (Without Worrying About Privacy)

Safe questions—use any ChatGPT plan:

• “How do I find naturalization records for someone who arrived in New York in 1910?”
• “What’s the best way to search for German immigration records?”
• “How do I evaluate conflicting birth dates in census records?”
• “What was happening in Puerto Rico in 1880 that might explain emigration?”
• “Help me organize my research notes—I have census records, church records, and naturalization documents. What’s a good system?”
• “How do I write a family history essay?”
• “What should I look for in a baptism record from 1850s Spain?”
• “Can you transcribe this handwritten 1880s census record?”
• “I’m having trouble reading this word in an 1800s document. It looks like ‘Schindhel’ or ‘Shindhel’—what German surnames are similar?”
• “I can read most of this baptism record from 1875, but a few words are illegible. Based on Spanish colonial records, what words commonly appear in this context?”

Why transcription is often low-risk: Many commonly used genealogical records describe people who are almost certainly deceased, and the content is frequently already public or broadly accessible. In those cases, the privacy risk is generally low.

But “historical” doesn’t always mean “non-sensitive.” Some older documents contain intensely personal information (illegitimacy, institutionalization, abuse, adoption, enslavement). Treat sensitivity as its own category—separate from whether the person is living.

---

What You Should NOT Paste Into ChatGPT (Without Privacy-Preserving Settings)

Be careful with anything that includes:

• Living people’s identifiers: address, phone, email, workplace, minor children, full date of birth, account numbers, student IDs
• Recent vital records that could reasonably involve living individuals (depending on jurisdiction and context)
• Private family material, such as letters, texts, emails, and family conflict narratives, that are not yet public.
• Unpublished work product: your original transcriptions, research notes, proof arguments, draft chapters—especially if you plan to publish

Why: The main risk isn’t “AI analyzing records.” The main risk is unintended disclosure of living-person data, plus losing control of unpublished research and writing.

Temporary Mode: Your Privacy Tool

What it does: Temporary Mode tells OpenAI “process this conversation to help me, but don’t use it for training and don’t save it after I leave.”

How to use it:

• Open ChatGPT Plus or Free
• Look at the top of the window—you’ll see a toggle for “Temporary mode”
• Turn it ON
• Have your conversation
• When you close the tab, the conversation disappears

What Temporary Mode does NOT do:

• It doesn’t encrypt your data (OpenAI still sees it while processing)
• It doesn’t make your data invisible to the OpenAI company
• It doesn’t protect you if you share the conversation with someone else later

It DOES:

• Prevent your conversation from being used to train the AI
• Prevent OpenAI from keeping a record of your chat
• Make your data less useful to the company

When to use Temporary Mode: You’re pasting a specific document and asking for help analyzing it. You want fast, focused help without a permanent record.

---

A Practical Workflow for Genealogists

If you’re asking methodology questions:

• Use regular ChatGPT (no Temporary Mode needed)
• “How would I find records for someone who emigrated from Ireland in 1920?”
• Save the conversation if you want to refer back to it

If you’re asking for help with a specific document:

1. Turn ON Temporary Mode
2. Paste the document or describe it: “I have a census record from 1880 with handwritten entries. I can’t read this word—it looks like it starts with ‘S’. What might it be?”
3. Ask your question
4. Get your answer
5. Close the tab—the conversation is gone, and your document wasn’t used for training

If you’re collaborating with another genealogist:

• Don’t use ChatGPT’s “Share” feature (see below)
  
  Instead, copy the key points into an email or shared document or describe the conversation verbally or in a message

---

A Special Case: Transcribing Historical Documents

Transcription is different from analysis. It’s one of the best uses of AI for genealogy, and the privacy concerns are much lower when you’re working with documents from the 1800s or earlier.

Why Historical Documents Are Safer

If you’re transcribing a document from 1800, 1850, or even 1920, the people in those records are almost certainly deceased. The information is historical, not current. A census record from 1880 is a historical document—it’s not identifying living people.

Examples of safe documents to transcribe:

• Census records from 1800-1950
• Church baptism, marriage, and burial records from any era
• Naturalization papers from 1800-1930
• Military records from WWI or earlier
• Property records from the 1800s
• Ship manifests and immigration records
• Wills and estate documents
• Newspaper clippings from decades ago

Many of these can be pasted with relatively low privacy risk when they clearly concern deceased individuals and don’t contain sensitive content.
When in doubt—especially with edge cases—use Temporary Mode or redact identifiers.

Also consider access terms and copyright. “Publicly viewable” does not always mean “free to reuse or upload anywhere.” Some sites and archives have terms that restrict redistribution or bulk reuse. Privacy and permission are different issues—both matter.

Why You Might Still Use Temporary Mode

Even though historical documents are safe in terms of privacy, you might still want to use Temporary Mode for transcription because:

1. You’re building a unique resource.
   If you’re the first person to transcribe a particular document, that transcription becomes your intellectual work. Using Temporary Mode means OpenAI won’t keep a copy for training.
2. You plan to publish.
   If you’re transcribing for a genealogy book or article, use Temporary Mode so ChatGPT doesn’t see the document before you publish it.
3. The document contains sensitive historical information.
   Some 1800s documents mention slavery, abuse, or other sensitive topics.
   You might prefer not to have OpenAI process them, even though it’s legal.

But here’s the important part: If you’re transcribing a historical document from the 1800s just to help yourself understand your family history, you don’t need Temporary Mode. The privacy risk is minimal.

Avoiding Privacy Overreach

It’s reasonable to protect living people’s privacy in genealogy. It’s also easy to go too far and accidentally reinforce the idea that historical records should be restricted simply because they contain personal facts.

A practical line to hold:

• Protect the living.
• Handle sensitive historical content thoughtfully.
• Don’t treat public, non-sensitive records about deceased people as off-limits for AI.

That balance keeps genealogists on the right side of ethics without helping institutions justify unnecessary record closures.

How to Use ChatGPT for Transcription

Method 1: Screenshot or Image

1. Take a photo of the handwritten document
2. Upload the image to ChatGPT (click the image icon)
3. Ask: “Can you transcribe this handwritten document into typed text?”
4. ChatGPT will read the handwriting and type it out

This works surprisingly well for 1800s documents, even with difficult handwriting.

Method 2: Describe What You See

If the image is unclear or ChatGPT has trouble:

1. Type what you can read: “I have a census record from 1880. The first line says ‘Maria [illegible] wife of John Smith, age 35, occupation: farmer.’ The second line is almost completely illegible except for what looks like ‘Puerto…’ Can you guess what common words might fit?”
2. ChatGPT can help you figure out what illegible words might be based on context and common names of the era.

Method 3: Paste Text and Ask for Help

1. Type out what you can read from the document
2. Ask: “I’ve transcribed this document, but there are [number] words I couldn’t read. Here are the gaps: [illegible], [illegible]. Based on the context and the era this document is from (1880s Puerto Rico), what might these words be?”
3. ChatGPT can suggest possibilities based on historical context.

Example: Transcribing a Historical Document

You ask ChatGPT:
“I’m transcribing a baptism record from 1875 in Puerto Rico. The handwriting is difficult. Here’s what I can read: ‘En el año de mil ochocientos [illegible], en [illegible] día del mes de [illegible]…’ Can you help me guess what the illegible words might be?”

ChatGPT responds with:
“Given the date format and Spanish colonial records, those gaps likely are: ‘mil ochocientos setenta y cinco’ (1875), a day number like ‘dieciséis’ or ‘veinte’, and a month name like ‘enero’ or ‘febrero’. The full phrase would typically be ‘[day] día del mes de [month]’.”

This is incredibly helpful, and you’re not exposing living relatives—you’re getting help with a historical document.

Privacy Summary for Transcription

Safe to paste into ChatGPT without Temporary Mode:

• Historical documents (1800s and early 1900s) with names, dates, locations of deceased people
• Church records
• Census records
• Naturalization papers – before 1920s
• Military records
• Ship manifests
• Property records (before 1920s)

Use Temporary Mode if:

• You plan to publish the transcription
• You’re building a unique research database
• You’re uncomfortable with OpenAI having copies of the documents

Never paste (even with Temporary Mode):

• Documents with living relatives’ current information –
• Records with addresses or contact information of living people –
• Anything that identifies someone currently alive

Why This Matters for Your Research

This is one area where AI is genuinely transformative for genealogy. Before AI transcription tools, you had to: – Struggle with handwriting yourself – Send images to other researchers and wait for help – Pay for professional transcription services – Accept that some documents would remain partially unreadable

Now, you can upload a photo of an 1800s document and get a transcription in seconds. This is a revolutionary tool for genealogists who work with Spanish colonial records, German immigration documents, or any other historical handwriting.

The key: Historical documents are safe. You’re working with the past, not exposing the present. Use that to your advantage.

ChatGPT has a “Share” button that lets you generate a link to share a conversation.

Don’t use SHARE for genealogical research.

Here’s why:
In July 2025, thousands of shared ChatGPT conversations appeared in Google search results. People who clicked “Share” to send a conversation to a colleague or friend ended up making it searchable by the entire internet. Some contained sensitive family information, medical details, and personal conversations.

OpenAI fixed this problem and removed the feature, but the lesson is clear: “Share” means “publish to the internet,” not “send to one person.”

If you need to share your analysis with someone:

• Describe it in an email
• Copy the key points into a document
• Take a screenshot
• Anything except clicking the “Share” button

---

Why This Matters: Three Real Scenarios

Scenario 1: Your Research Gets Published (And So Does Your Data)

You’re researching the Martinez family. You paste their naturalization records into ChatGPT for help with translation. Later, you publish a genealogy article about the Martinez family. Months later, a researcher on the same family tree stumbles across your ChatGPT conversation in OpenAI’s training data—they can see not just your published work, but also your research notes, your questions, and the documents you used. You lose your competitive advantage.

How to avoid it: Use Temporary Mode for any document you plan to publish.

Scenario 2: You Share a Conversation, and It Becomes Public

You’re collaborating with your cousin on family history. You create a ChatGPT conversation analyzing migration patterns and click “Share” to send the link to your cousin.

You think they’re the only one who can see it. But if the “Make this chat discoverable” option was enabled, that conversation is now searchable on Google, with all the family names, dates, locations, and relationship information you discussed.

How to avoid it: Never use the Share feature. Send a description, a screenshot, or the key points in an email instead.

Scenario 3: A Living Relative’s Privacy

You’re researching your family tree and discover a living aunt’s maiden name, her current address, and her children’s names in historical records. You paste these documents into ChatGPT to ask for help finding more information about her. Your aunt never consented to having her current information analyzed by AI. You’ve made a choice about her privacy without her permission.

How to avoid it: Before pasting any document with living relatives’ information, ask yourself:
“Would they want OpenAI to see this?” If the answer is no, don’t paste it.

---

Quick Reference: What to Do

Before you paste anything into ChatGPT, ask yourself:

1. Does this contain living relatives’ names or current information?
   • If yes: Don’t paste it, or use Temporary Mode
2. Am I planning to publish research based on this document?
   • If yes: Use Temporary Mode
3. Would I be comfortable with a stranger reading this?
   • If no: Use Temporary Mode or don’t paste it
4. Am I about to use the “Share” button?
   • If yes: STOP. Use email or a screenshot instead

What to do instead of pasting documents:

• For translation help:
  Describe the problem. “I have a handwritten word in a 1840s record that looks like ‘Schindhel’ or ‘Shindhel’. It’s a German surname. What spellings are similar?”
• For transcription help:
  Type a few words manually and ask about the rest. “The first line says ‘Maria [illegible] wife of…’ What might the middle name be?”
• For methodology help:
  Ask without the documents. “I found two different birth dates for my ancestor. How should I evaluate which one is more reliable?”
• For collaboration:
  Copy key findings into an email. Don’t send a ChatGPT link.

---

The Bottom Line

ChatGPT is genuinely helpful for genealogy. It’s excellent for: – Brainstorming research strategies – Understanding historical context – Working through conflicting evidence – Learning how to approach a research problem

Just be intentional about what you share. The same care you use to evaluate historical sources should guide how you use AI tools.

Main rules:

1. Protect living people’s information (and redact when needed)
2. Use Temporary Mode for living-person data, sensitive content, or unpublished work
3. Public, non-sensitive records about deceased people are generally reasonable to analyze with AI
4. Treat “Share” as “Publish” and avoid it for genealogy

That’s really all you need to know. Use the tools, get the benefits, and keep your family’s research safe.

---

For Questions About OpenAI’s Current Policies

OpenAI changes its policies frequently.
To verify the current state of Temporary Mode and other privacy features, check: – https://openai.com/policies/ – https://help.openai.com/en/articles/7730893-temporary-chat

---

What About Other AI Tools?

This guide focuses on ChatGPT because it’s the most popular tool among genealogists.
Google Gemini and Claude work similarly—they both can see what you type and offer privacy options.
The same rules apply:
– Be thoughtful about what you share.
– Use privacy modes when available.
– Never use share features for sensitive genealogical data.

If you’re curious about other tools, the principles are the same. The specific features change, but protecting your research and your family’s privacy is always the priority.

---

Sources for This Guide—Please Fact-Check

I believe it’s important that YOU verify the claims in this article yourself.
Please note that while I was writing this, the links to various AI vendor policies changed, and some URLs may be broken.
Here are the sources I used, so you can check them and confirm that what I’m saying is accurate:

On ChatGPT’s Temporary Mode:

• OpenAI Help Center: Temporary Chat documentation
• OpenAI Privacy Policy

On the July 2025 ChatGPT Sharing Incident: When ChatGPT’s share feature made conversations searchable by Google and other search engines, multiple sources documented the issue:

• TechCrunch investigation (July 31, 2025): “Your public ChatGPT queries are getting indexed by Google and other search engines”
• Search Engine Land analysis (August 1, 2025): “Google indexing shared ChatGPT conversations”
• GRC Report summary (January 2025, covering the July 2025 incident): “OpenAI Shuts Down ChatGPT Sharing Feature After Privacy Concerns”
• Snyk Security analysis: “The Security Risks of GPT Chats Leaking to Search Engines”
• Digital Position explainer: “Shared ChatGPT Conversations Are Now on Google Search” ow-on-google-

On how OpenAI responded:

• Fortune reporting (August 5, 2025): “OpenAI will no longer let ChatGPT conversations be indexed”
• OpenAI’s official statement via Chief Information Security Officer Dane Stuckey, reported in multiple outlets including the sources above… (Too many more links to add…)

General AI and Privacy Information:

• Anthropic (Claude) Privacy Policy(Consumers)
• Google (Gemini) Privacy

Why I’m Including These References:

You should never take my word for it—or anyone’s word for it. These are technical and privacy matters that affect your family’s information. If you’re going to follow the advice in this guide, you should:

1. Check the sources yourself
2. Verify that the features I describe are still accurate (policies change)
3. Read OpenAI’s official documentation directly
4. If you have questions, contact OpenAI support with a screenshot of what you see on your screen

The links above will let you verify every major claim in this guide. If something has changed, or if you find that the current version of ChatGPT works differently than I’ve described, please send me an update at the contact information on looking4myroots.com.

A Note on Fact-Checking and AI Tools:

As genealogists, you’re trained to evaluate sources.
You know to ask:
– Who wrote this?
– Is this primary or secondary evidence?
– Does it have bias?
– Could it be outdated?

The same skepticism applies to AI guidance.
– Don’t assume any article (including this one) is complete or the truth.
– AI company policies change frequently, sometimes without clear announcements.
– The features available today may not be available next month.

Always verify current information by:
– Going to the official company website (openai.com, not a blog post)
– Checking your own account to see what options are actually available
– Testing the feature yourself before relying on it for sensitive work
– Asking OpenAI support directly if you’re unsure

That’s how careful genealogy research works, and it’s how careful AI tool usage should work too.