How I talk to AI (so it actually helps me finish)

Most people's first AI prompt is too vague to be useful. Here's how to level up from conversational to structured prompts that actually help you finish.

Part 3 of 3: AI for Family Historians Who Want to Finish

Welcome to all our new subscribers! It’s a great time to get started with AI. Catch up with Part 1 and Part 2 of the series. You can read all my past posts here.

---

My first AI prompt was something like “help me with my genealogy research in 1820s Pennsylvania.”

The response was... fine. Like a blog post that could suddenly talk, but hardly exciting.

It took me a while to figure out that the quality of what I got back depended entirely on how I asked. Once I learned to have a real conversation with AI, giving it context, being specific, following up, I finally got what the excitement was about.

One Setting You Need to Turn On First

Before we get into prompts, here’s something critical: turn on memory in your AI tool.

ChatGPT has a memory toggle in settings. Claude has it too and so does Gemini. When memory is on, the AI remembers context from previous conversations. This means you don’t have to re-explain your entire research project every single time you open a new chat.

Without memory, you’re starting from zero every time. With memory, the AI knows you’re researching Pennsylvania German immigrants, that you’re working on your Mueller line, that you prefer certain citation formats.

Turn it on. Everything else works better.

The Prompt Progression: From Simple to Powerful

Here is something 90% of AI don’t realize: prompts exist on a spectrum. You can start simple and conversational. But when you need consistent, reliable results, you add structure.

Let me show you what I mean using the same task—understanding a census record—at three different levels.

Level 1: The Conversational Prompt (Good for Exploration)

This is where most people start, and it works fine for quick questions:

I found this 1880 census for my great-grandfather Johann Mueller. He’s listed as a ‘moulder’ in Pittsburgh. What did that job involve?

What you get: A helpful explanation of what a moulder did in iron foundries.

Why it works: Fast, natural, gets you unstuck.

Limitation: Every response will be slightly different. Hard to compare results across multiple ancestors.

---

Level 2: The Structured Prompt (Better for Consistency)

When I want more control over what comes back, I add structure:

I found this 1880 census record for my ancestor. Please analyze it using these categories:

1. Occupation: What this job involved and what it tells us about social/economic status

2. Household composition: Who’s living with them and what that reveals

3. Neighbors: Who’s nearby and why that might matter

4. Context: What was happening in this location during this time period

Here’s the record: [paste]

What you get: Organized analysis in the exact categories you need.

Why it works: You can use this same prompt structure for every census record and compare results systematically.

Limitation: Still somewhat open-ended. You’ll get good answers, but they vary in depth.

---

Level 3: The Detailed Prompt (Best for Reliable, Repeatable Results)

When I’m working on a chronicle and need consistent depth across multiple ancestors, I get very specific:

You are a genealogy research assistant helping me understand historical records. I need you to analyze this 1880 U.S. census record and provide specific insights.

Background context:

- Ancestor: Johann Mueller

- Research goal: Understanding his life in Pittsburgh during the industrial era

- What I already know: He immigrated from Germany around 1875, worked in iron industry

Please analyze using this exact structure:

1. OCCUPATION ANALYSIS- Job title and what it specifically involved (tools, process, daily tasks)- Skill level required and typical wages for this work- Physical demands and health risks- Social status of this occupation in 1880s Pittsburgh

2. HOUSEHOLD COMPOSITION- List each person, their relationship, age, and birthplace- Family structure patterns (nuclear vs. extended, boarders, servants)- Age gaps between children and what they might indicate- Any notable absences or unusual household members

3. NEIGHBORHOOD CONTEXT- Who lives on either side and across the street- Occupational patterns in the neighborhood- Ethnic/immigrant clustering- Potential family connections based on surnames or birthplaces

4. HISTORICAL CONTEXT- What was happening in Pittsburgh’s iron industry in 1880- Economic conditions affecting workers- Immigration patterns from Germany to Pittsburgh during this period- Specific events or conditions that year

5. STORY POTENTIAL- What narrative threads emerge from this record- What questions does this raise for further research- What details would bring this person to life in writing

Format your response with clear headers for each section. Be specific. If you’re inferring something, say so explicitly.

Here’s the census record: [paste]

What you get: Deep, consistent analysis you can rely on. Same structure every time. Easy to spot patterns across multiple records.

Why it works: You’ve told the AI exactly what role to play, what background it needs, what structure to follow, and how to format the answer.

When to use it: When you’re working systematically through multiple records and need to compare results. When you’re actually writing and need reliable depth.

---

The Pattern You’ll Notice

See what happened as we moved through those three levels?

Level 1 is a question. You’re asking AI to figure out what you need.

Level 2 gives AI categories. You’re guiding the analysis.

Level 3 gives AI a role, context, structure, and format requirements. You’re designing a system that works the same way every time.

Most people never get past Level 1. They think that’s all prompting is.

But when you’re ready to finish—when you’re actually writing chronicles, not just collecting information—you need Level 3 prompts. They’re longer to set up, but they save you hours of sorting through inconsistent responses.

One More Critical Thing: The 5-Turn Rule Still Applies

Even with great prompts, remember from Part 2: conversations drift after about 5 exchanges.

After 4-5 back-and-forth turns, start fresh. Copy your key findings into a new chat. This keeps AI grounded and prevents that slow drift into confident-sounding fiction.

Memory helps here too. The AI will remember your research focus across fresh chats, but each conversation itself should stay short and focused.

This is Why Writing is So Hard

When I started building Level 3 prompts for my most common tasks—understanding records, drafting narratives, analyzing migrations—I stopped feeling like I was wrestling with AI. It became a reliable collaborator.

Those records I’d collected for years started telling me stories I’d missed. And I could finally write about them with confidence because I understood what I was looking at.

Understanding your records is what lets you finally write. You can’t tell your ancestor’s story if you don’t understand what their life was actually like. AI bridges that gap—but only if you know how to prompt it effectively.

Inside Chronicle Makers, this is what we do in Lab 1: Family Tree Foundations. We teach you to build prompts that work for your specific research. VIP members are currently working together on something I’m really excited about: we’re building a complete book-writing system with AI (not just chronicles, but full family history books). They’re testing it now before it goes anywhere else. The energy in the community around this is incredible.

Try This: Build Your First Level 3 Prompt

Pick one ancestor you’ve been meaning to write about. Pick one record you have for them.

Open ChatGPT or Claude and build a Level 3 prompt using this structure:

1. Give it a role: “You are a genealogy research assistant helping me understand historical records.”

2. Provide background context: Who are you researching? What’s your goal? What do you already know?

3. Specify the structure you want: List exactly what sections or categories you need analyzed.

4. Set format requirements: Headers, bullet points, specific vs. general—tell it how to organize the response.

5. Paste your record at the end.

Try it. See what happens. Notice the difference from “help me understand this census record.”

Then remember: start fresh after 5 turns, and make sure memory is turned on.

Quick Reference: The Series in a Nutshell

Part 1 – What AI Is: Large Language Models trained on text, not genealogical databases. Start with the real tools (ChatGPT, Claude, Gemini), not limited wrappers.

Part 2 – When to Use It: Understanding and writing, not finding or verifying. Remember the 5-turn rule.

Part 3 – How to Talk to It: Turn on memory. Progress from conversational to structured to detailed prompts. Know which level you need for the task at hand.

What’s next

This series gave you the foundation: a mental model for how AI works, when it’s useful, and how to prompt it effectively. From here, the best teacher is practice.

Pick a record. Build a prompt. See what happens.

And if you want to go deeper—if you want integrate AI into everything you do with genealogy, plus the community support that helps you actually finish—that’s what Chronicle Makers is for.

You don’t have to figure this out alone.

One more thing: writing about real people

Once you start writing, there’s something most family historians don’t think about until it’s too late: the legal side of telling real stories.

I’m hosting a workshop with attorney Elissa D. Hecker designed specifically for genealogy authors who publish family histories, memoirs, and narrative nonfiction. We’ll cover defamation, privacy, copyright, fair use of letters and photographs, consent and releases, and how to handle sensitive family stories—all in a conversational format that educates without paralyzing us with fear.

Elissa is a long-standing New York Super Lawyer recognized among The Best Lawyers in America for Copyright Law, with extensive experience across entertainment and publishing.

Get a ticket here and join us.

Happy Chronicling!

—Denyse

---

P.S. If this series helped you see AI differently, I’d love to hear about it. Reply to this email and tell me what clicked—or what questions you still have. Your questions help me know what to write about next.