- Published on
Lesson 1 of "How to Use AI for Family History"
Welcome to our new 5-part series on using artificial intelligence in family history. Whether you're an established researcher keeping up with new tools or a newcomer starting your first tree, one thing is clear: AI is now part of the research landscape.
Over the next five lessons, we'll show you how to use AI safely and practically, without compromising your genealogical standards (or your sanity). By the end of this lesson, you'll have written your first prompt, understood why it's built the way it is, and learned how to verify what comes back. Beginners will fill in the blanks of a ready-made prompt to pinpoint where an ancestor's records were kept, while Intermediate researchers will add time windows, boundary changes, and source requirements to that same base. At the end of each lesson, the Takeaway Prompts let you put these tools to work on your own ancestor's details.
From Novelty to Daily Tool
Not long ago, generative AI felt like a party trick that wrote awkward poems. Today, AI assistants are as common as search engines, digital calendars, and spell checkers, and many people use them every day.
For family historians, that makes them useful for organizing notes, brainstorming research paths, and untangling unfamiliar historical context. But like any tool, what you get out depends on how well you use it. That is what this series teaches.
The Universal Trap: Confident Guesswork
If you're hesitant to use AI, you're in good company. You've probably seen the headlines about AI "hallucinating," which means confidently stating things that simply aren't true.
In genealogy, a hallucination might look like:
- an 1850 census entry that doesn't exist
- a set of parents for your ancestor, offered with no source
- a records archive that was never real, or a county courthouse that hadn't been built yet
Here's why it happens. These tools are built to be helpful and fluent, so a loose, unstructured question tends to get a smooth, confident answer, even when the tool is effectively guessing. "Garbage in, garbage out" still applies, but so does polished garbage out.
The way through is a mindset: AI is not an author or an oracle. It is an assistant. With clear boundaries, the risk shrinks to something you can manage, and you stay in charge of what goes into your tree.
Your Mindset: Three Ground Rules
Every good dig begins with a survey of the site. To bring AI into a serious research workflow, start with three ground rules:
- Treat AI like an eager, hyper-fast research intern. It works for you, under your direct supervision. You would never file an intern's first draft without checking it.
- Ask specific questions. Vague questions invite guesswork; targeted questions produce useful clues. The formula for a targeted prompt is coming up below.
- Expect leads, not proof. A good reply gives you a roadmap, a historical context check, or a list of repositories to explore. It never gives you a verified family connection or a finished genealogical proof.
And remember: traditional standards still rule. Whether you're hunting 20th-century vital records or tracing 17th-century colonial New England lines, original records and careful verification are always the final authority.
One Rule Before You Type: Protect Living People
An AI chat window is not a private notebook. Before your first prompt, adopt one habit: don't paste in names, birth dates, addresses, or DNA results for living relatives. Use placeholders ("a living cousin," "my grandmother, born in the 1940s") or stick to ancestors who have long since passed.
It's also worth checking the data settings of whichever tool you choose. Many let you opt out of having your chats used to train future models, and the defaults vary by service and plan.
Two Tracks: The All-Levels Fitness Class
In a great fitness class, everyone uses the same equipment and scales the movement to their own level. This series works the same way. Every lesson has two tracks built on the same skeleton:
- Beginner track. If you're feeling tech-shy, you'll gain the confidence to open an AI chat window, use plain language, and take small, safe steps without fear of breaking anything or messing up your tree. You fill in the blanks of a ready-made prompt.
- Intermediate track. If you're already comfortable with tech, you'll add constraints such as time windows, jurisdiction history, and source requirements. That lets you scale up to complex multi-jurisdictional research, like tracking town boundary shifts or parsing old deeds, and shave hours off your preliminary reconnaissance.
Meet Your Practice Families. Each track follows one invented family through all five lessons, so you can watch the same prompt grow. Beginners follow the Harlows of Worcester, Massachusetts, in 1880. Intermediate researchers follow the Calders of Lancaster, Massachusetts, from 1650 to 1730. Every name and detail is fictional.
Pick your track, or try both. The prompt formula underneath is the same.
The Formula for a Good Prompt
So how do we get useful leads instead of confident guesses? We build a good prompt. Rather than dumping a wall of requirements into a chat box on day one, we'll use a simple formula with three lightweight parts:
ROLE + CONTEXT + TASK + GUARDRAIL = A USEFUL RESEARCH GUIDE
These names are my own teaching terms, chosen to be easy to remember. AI tools use different words for the same ideas, and a bonus post after Lesson 5 connects the two. Below, each part comes with the “why” (what goes wrong when it’s missing) and an example for each track.
PART 1: The Role ("Act as...")
Why it matters: AI models are pattern-matching engines trained on everything from internet forums to academic journals. With no role, you get a generic assistant's tone and level of detail. Naming a role nudges the tone, vocabulary, and framing of the reply.
A caution: a role changes how an answer sounds, not how reliable it is. A "professional genealogist" can make a wrong answer sound more authoritative. That is why the third part, the guardrail, matters most.
BEGINNER TRACK
ROLE
Example: Act as a patient, beginner-friendly genealogy guide.
Why it works: It asks for plain language, minimal jargon, and clear foundational steps.
INTERMEDIATE TRACK
ROLE
Example: Act as a professional genealogist specializing in 17th- and 18th-century New England jurisdictional history and archival methodology.
Why it works: It points the AI toward regional framing, such as colonial town development, parent-county histories, and specialized archives. You still verify what it says. Swap in your own region and period, such as 19th-century Ireland or the American West.
PART 2: The Context and Task (Scope + What You Want Back)
Why it matters: "Garbage in, garbage out" happens most often right here. AI can't read your mind, and it doesn't know where your ancestor lived unless you say so. Without a place, a time frame, and a clear goal, it fills the gaps with assumptions: wrong jurisdictions, irrelevant record types, or invented historical details. The context is the facts (who, where, when). The task is what you want back.
BEGINNER TRACK
CONTEXT + TASK
Example: I am researching the Harlow family, who lived in Worcester, Massachusetts, around 1880. I need to find where birth, marriage, and death records were registered.
Why it works: One family, one place, one decade, one record type, and one clear goal.
INTERMEDIATE TRACK
CONTEXT + TASK
Example: I am researching the Calder family in Lancaster, Massachusetts, between 1670 and 1720. I need to identify surviving land, tax, or probate records created before Worcester County was formed in 1731.
Why it works: It sets a wider time window, flags a boundary shift (the creation of a new county), and targets specific record groups. Lancaster is just our running example, a colonial town whose changing boundaries make it a classic brick wall. The same approach works anywhere borders or jurisdictions have moved.
PART 3: The Guardrail (Limits and Safety Constraints)
Why it matters: This is the most important part for staying in control. Generative AI is designed to be helpful, which means when it doesn't know an answer, it may generate a plausible-sounding response anyway. Setting explicit boundaries lowers that risk. These constraints prompt the AI to flag uncertainty, point toward verification, and focus on finding leads rather than asserting ancestral facts. While this doesn't make the tool infallible, it ensures you remain the critical evaluator of its output.
BEGINNER TRACK
GUARDRAIL
Example: Do not invent historical records or archives. List 2 to 3 specific physical or digital repositories where these records are held today. Give me the official repository names I can search for, and tell me how I can verify that each one exists.
Why it works: It forbids fabrication and builds in a human verification step before any lead is accepted. Asking for names instead of web addresses matters, because made-up links are a common failure.
INTERMEDIATE TRACK
GUARDRAIL
Example: Do not cite compiled genealogies as proven fact. State where primary records or microfilmed local records must be inspected to verify these boundary claims, and note any known record loss.
Why it works: It aligns the output with professional proof standards and asks the AI to account for record destruction (such as courthouse fires or missing town volumes) and repository limits.
A caution: Claims about record loss are a favorite spot for hallucination. Treat that list as something to verify against a reference such as The Genealogist's Handbook for New England Research (or the research guide for your own region), not as a finding.
Where Should You Type Your First Prompt?
You don't need special software or downloads. Several free chat assistants run in a web browser or as a mobile or desktop app. Choose the one that feels most comfortable:
- ChatGPT: a widely used general-purpose chat tool with a conversational tone.
- Claude: a chat assistant known for structured, carefully worded responses.
- Gemini: Google's assistant, and a natural fit if you already use Google tools every day.
Free versions differ in their limits and in whether they can search the web, and those features change often, so check each chat assistant's current settings. If your chosen tool offers web search, turn it on and ask it to name its sources.
The golden rule: don't hunt for the perfect tool on day one. They all respond to the same prompt formula. Pick one, open a fresh chat, and try your first prompt right alongside us.
Your Takeaway Prompts
You've now seen all three parts, and here they come together into one prompt for each track. Each track has two versions of its prompt. TRY IT FIRST is a completed prompt with an answer key, so you can see what a good reply looks like. YOUR TURN has bracketed blanks for your own ancestor. Copy and paste the "try it first" version into a fresh chat window, compare the reply with the answer key, and then swap in your own details for the second "your turn" chat. Save your try-it-first prompt and the AI's reply in a document, tagged "lead, unverified," so you'll have them for Lesson 2.
BEGINNER TRACK: creating a Research Guide
Prompt A: Try it First
Act as a patient, beginner-friendly genealogy guide. I am researching the Harlow family, who lived in Worcester, Massachusetts, around 1880. I need to find where birth, marriage, and death records were registered at that time. Do not invent historical records or archives. List 2 to 3 specific physical or digital repositories where these records are held today. Give me the official repository names I can search for, and tell me how I can verify that each one exists.
Answer key: Massachusetts mandated statewide civil registration starting in 1841, so records from 1880 are formally documented. A strong reply points you to two specific places: the Worcester City Clerk's Office for local records, and the Massachusetts State Archives (or Registry of Vital Records and Statistics) for state-level copies. If the AI confuses civil records with church registers, or misses the fact that Massachusetts kept both local and state copies, it is missing the mark on local accuracy. Then check that each repository it names really exists, using the 3-minute check below.
Prompt B: Your Turn
Act as a patient, beginner-friendly genealogy guide. I am researching a direct-line ancestor who lived in [Town/City, State or Country] around [Year]. I need to find where birth, marriage, and death records were registered at that time. Do not invent historical records or archives. List 2 to 3 specific physical or digital repositories where these records are held today. Give me the official repository names I can search for, and tell me how I can verify that each one exists.
INTERMEDIATE TRACK: creating a Research Guide
Prompt A: Try it First
Act as a professional genealogist specializing in 17th- and 18th-century New England jurisdictional history and archival methodology. I am researching the Calder family in Lancaster, Massachusetts, between 1670 and 1720. I need to identify surviving land, tax, and probate records and where they are held, including any records created before Worcester County was formed in 1731. Do not cite compiled genealogies as proven fact. State where primary records or microfilmed local records must be inspected to verify these claims, and note any known record loss for this area.
Answer key: Worcester County was created in 1731, so earlier records were filed under Middlesex County. A strong reply should name Middlesex and point you to its records. Confirm it in a town history or the Newberry Library's Atlas of Historical County Boundaries (or your region's equivalent), because sources don't always agree. If the AI never mentions the change, you've just caught it guessing.
Prompt B: Your Turn
Act as a professional genealogist specializing in [Region, e.g., 17th- and 18th-century New England] jurisdictional history and archival methodology. I am researching a family in [Town/City, State or Country] between [Start Year] and [End Year]. I need to identify surviving land, tax, and probate records and where they are held, including any records created before major county boundary changes in this region. Do not cite compiled genealogies as proven fact. State where primary records or microfilmed local records must be inspected to verify these claims, and note any known record loss for this area.
After You Press Enter: The 3-Minute Check
Your prompt is only half the job. Before anything goes into your notes or tree, run this quick check:
- Pick one lead and confirm it exists. Search for a repository or record set the AI named, on the repository's own website or in a catalog such as the FamilySearch Catalog. If you can't find it, treat that lead as invented.
- Look for sources. Did the reply say where its claims come from? If not, ask: "What is the basis for that statement, and how would I confirm it?"
- Label it. In your notes, tag anything the AI told you as "lead, unverified" until you've seen an original record.
(We'll go much deeper on auditing AI answers in Lesson 5.)
BEGINNER TRACK
Check: Do step 1 every time, and you're already ahead of most people.
INTERMEDIATE TRACK
Check: Do all three, then compare the reply against a reference work or a county history before you plan a repository visit.
Coming up in Lesson 2:
We'll break ground and use our first structured prompts to build a Town Profile, a Records Roadmap, and a Map-Finder for the Harlows' Worcester, Massachusetts (Beginner track) and the Calders' Lancaster, Massachusetts (Intermediate track). You can repeat each step on your own ancestor's town in the Takeaway Prompts.
Bring to Lesson 2: your finished "try it first" prompt for your track and the AI's reply, saved in a document and tagged "lead, unverified." Lesson 2 shows you how to set that document up.
What has been your biggest hesitation about trying AI tools for your research? Drop a comment below!
