Build a family tree with ChatGPT: my workflow and prompts

I tried building a family tree with ChatGPT from what I already knew about my family. I worked through four steps: collecting memories, researching the surname, searching for records and making a diagram. This guide shares that workflow and the prompts so you can start with your own information.

ChatGPT can organize names and relationships, search the web for sources and prepare an initial tree. Any person it finds needs checking. A shared surname does not prove a relationship, and a history of the name does not establish your family's history.

Before you start: what can you do in half an hour?

If you have names and basic details ready, allow about half an hour for a first draft. Research can take much longer when you need to open records, read handwriting or distinguish people with the same name.

Use one written conversation so you can copy prompts and review answers. For the research step, explicitly ask for web search and source links. You can select Search from the tools menu; availability depends on account settings and usage limits. OpenAI's search instructions.

Gather names, names before marriage, birthplaces and approximate years. "Around 1950" is useful if it stays marked as approximate. Leave missing dates blank.

Step 1: collect memories one branch at a time

I start with the family I know. The first prompt asks ChatGPT to interview me, rather than expecting me to organize a long list myself. You can answer "I don't know" and return to the question later.

Prompt 1: the starting interview

I want to build a family tree from what I know. Interview me one branch at a time, covering both sides. Start with me, my parents and grandparents, then ask about other relatives.

For each person, ask for their full name, previous names or names before marriage, approximate birth year, birthplace and relationship. Ask a few questions per reply and wait for my answers.

Keep every spelling variation I provide. Mark approximate dates, missing details and contradictions. Do not invent people or relationships to complete a branch.

At the end, give me an initial tree and a list of open questions. For each detail, state whether it comes from my memory or a document I supplied. Do not label a remembered detail as verified by a record.

If possible, sit with a parent or grandparent. A question about a town can bring back a sibling's name or an immigration story. Note who supplied each detail so you know whom to ask if records disagree.

Include names before marriage, alternative spellings, towns, approximate birth and death years, immigration dates, and relevant occupations or service history. Keep remembered information distinguishable from documentary evidence.

Step 2: research the surname

A surname can suggest a language, region or older spelling to search. Unrelated families can share it. I ask about the name's history while leaving its connection to my family open to investigation.

Prompt 2: surname origins and historical context

Using the surname and places of origin I supplied, search reliable sources about the name's origins.

Check its meaning, language roots, geographic distribution and spelling variants. Identify historical events that may be relevant to research in the place and period I mentioned.

Give a source I can open for each explanation. Separate documented information about the name from general patterns and hypotheses. Do not assume a name change or migration happened in my family without evidence.

Suggest records that could help investigate the previous generation and where to find them. If the surname has several possible origins, show them without choosing one just because it fits a story.

Ask for the name in its original alphabet. If there are several explanations, compare their places and periods with what you know before choosing a research direction.

Step 3: search for records and possible relatives

Ask ChatGPT to search the open web. Ordinary search does not automatically unlock paid genealogy subscriptions or records requiring an account or permission. If a source is inaccessible, ask for that limitation and a way to check it yourself.

Prompt 3: records and research leads

Use the information we collected to search the open web for records and people who may connect to my family.

Search genealogy sites, obituaries, publicly accessible immigration and census records, historical newspapers and burial databases. Try the spelling variants and places I supplied.

For each result, provide the person's name, possible relationship, direct link, collection name and record identifier where available, and the details actually shown in the source. Explain which details match and which conflict with our information.

Mark each result as a research lead until we check the record and relationship. Do not invent people, documents, quotations or links. Say if you only saw a search snippet. Do not describe a document as read if you could not access it.

Finish with a checklist for manual verification. Do not add these people to the verified branch of the tree.

Starting points include FamilySearch, WikiTree and Find a Grave, along with archives relevant to the country of origin. Access varies by collection; some records require an account, permission or an archive visit.

An obituary may name a spouse, children or siblings. A burial record may provide a date or place. Check who supplied the information and what supports it before treating it as evidence of a relationship.

Compare ages, locations and relatives' names, then open the source itself. Keep conflicting details alongside the matches. Do not ask the model to explain away every inconsistency.

Step 4: make a diagram and save the data

Once the information is organized, ask for a tree by generation. I keep search candidates separate from the tree based on the information I collected, so a plausible match does not become a relationship that looks settled.

Prompt 4: the diagram and GEDCOM export

Build a family tree by generation using the information we collected. Show relationships clearly and use parents and grandparents as anchor points where we have that information.

Mark what came from memory, what was verified in a document and what remains unknown. Put unverified candidates in a separate list without connecting them as proven relatives.

Prepare a diagram I can copy or print, plus structured data. Before exporting GEDCOM, ask which genealogy application I will use and check which version it accepts.

If file creation is available in this chat, provide a downloadable GEDCOM with source notes and approximate dates marked. Otherwise, provide the file contents and saving instructions. Check for missing references and conflicting relationships. Exclude sensitive details of living people from a public version.

GEDCOM transfers family history data between genealogy applications. After importing, check names, relationships and dates. Successfully generating a file does not establish that its data or structure is correct.

FamilySearch now also has a newer upload experience for user trees that can be edited further. The older Pedigree Resource File route is being replaced. Check the upload route available in your account rather than relying on old menu instructions. Uploading a file does not verify its people or relationships. FamilySearch's current GEDCOM upload guidance.

When an answer looks wrong or research stalls

If a new relative appears without a usable source, ask for an audit:

Review every person and relationship you added. State where the information came from: my memory, a document I supplied or a web source I can open. Remove unsupported claims. Move name-only matches to the unverified leads list.

If search finds nothing, add a town, a name before marriage, a date range or another spelling. Check that web search is available in the conversation.

For a specific ancestor, try:

I am researching [name], born around [year] in [place]. Which records might exist for that place and period? Give the archive or collection name, link, search terms and spelling variants. Explain how I could investigate material that is not online.

In a long chat, ask periodically for a summary containing the current tree, sources and open questions. Save it outside the chat. Use it to start a new conversation if needed.

Privacy and living relatives

A draft does not need identity numbers, full addresses or full birth dates of living relatives. Share only what the research needs and review what will appear in any file you pass on.

In ChatGPT, open Settings, then Data controls. Where available, turn off Improve the model for everyone if you do not want conversations used for model training. Options depend on the account and workspace. OpenAI's data controls guidance.

For a possible living relative, start with a short message and one question:

Draft a short message to someone who may be related to me. I am [name], and the possible connection is through [person or branch]. Present the relationship as a hypothesis and ask one question that could help check it. Do not include sensitive personal details or the full research report.

Questions about building a family tree with AI

Can ChatGPT find relatives I did not know about?

It can search public information and suggest matches. Each match needs a record check and a relationship check. An answer with a real link can still assign the record to the wrong person.

Do I need to pay to start?

You can begin with memories and open sources. ChatGPT search is available on the free plan, subject to usage limits. An external site's particular records may require an account or payment.

Does the surname prove where my family came from?

No. It can suggest a region or spelling to research. Connecting that history to your family requires sources linking the generations.

Can I print the tree or move it to another application?

Ask for a printable diagram and GEDCOM export. Check the destination application's requirements and the imported result. Keep unresolved details in a separate list.

Can the tree prove citizenship or inheritance rights?

An AI answer or chat-generated diagram is not legal proof. Ask the relevant authority which documents it requires.

Start with a branch you know

Open a chat, paste the interview prompt and begin with parents and grandparents. After organizing one branch, choose a missing detail and look for a source. Save the link beside the person it concerns and keep the open questions with the tree.

For more prompts to adapt to your work: NexAi's ChatGPT command library.