If you’ve attended one of my webinars, read my book, or followed my social media posts, you’ve probably heard me talk about the basic structure of a good AI prompt as:

Role – Task – Format – Limits

It’s a simple framework that works remarkably well. For example, compare:

“Tell me about Italian genealogy.”, to:

“You are a professional genealogist specializing in Italian genealogy. Create a beginner-friendly research plan for finding birth records in Southern Italy. Format the response as a numbered list and keep it under 500 words.”

The second is a much better prompt. You’ve given the AI a role to play, a task to complete, a format to follow, and limits to work within.

For many genealogy tasks, that’s enough. But what happens when you need more than a quick answer? What happens when you’re working on a difficult research problem, trying to understand historical context, or tackling a brick wall ancestor? That’s when it’s time to move beyond the basics.

Thinking Like a Genealogist

One of the biggest mistakes people make with AI is treating it like a search engine. Search engines are designed to answer short questions. Genealogical research rarely works that way. When we research, we already know that context matters. A birth record means something different if it was created in Brooklyn than if it was created in Brussels. An 1890 census record in the US is going to be impossible to find. Historical events can dramatically affect where records were created and whether they survived.

As genealogists, we naturally think about these factors, but AI doesn’t always know which details are important unless we tell it. The secret to getting better answers is to include some of your own knowledge in the prompt.

Add Background Information

Let’s say you’re researching an Italian orphan born around 1870. A simple prompt might be:

*”Create a research plan for tracing an Italian orphan born in Southern Italy around 1870.”*

That’s not bad. But now consider this:

“Create a research plan for tracing an Italian orphan born in Southern Italy around 1870. Consider that Italian unification occurred between 1861 and 1871, record keeping varied by region, and orphan records may have been created by foundling hospitals, churches, or civil authorities.”

 The second one is better because you gave the AI additional context. You helped it understand the research environment before it began generating an answer. Think of it as the difference between asking a librarian for a book and explaining what project you’re working on. The more context you provide, the more useful the assistance becomes.

Tell the AI About Any Challenges

Experienced genealogists know that every research problem comes with obstacles. The surname may have changed over time. Perhaps the records are in another language, or a courthouse burned.   When you already know about a challenge, tell the AI. For example:

“Please create a research plan. One challenge is that the surname may have been assigned to a foundling rather than inherited from biological parents.”

That single sentence changes the direction of the entire response. Instead of producing a generic research plan, the AI is more likely to discuss foundling records, orphanages, naming conventions, and other specialized resources. In other words, you’re helping the AI think about the problem the same way you would.

Ask “What Am I Missing?”

This is the opposite direction of the last suggestion, and one of my favorite prompt techniques. Genealogists are experts at finding missing information, but sometimes we don’t realize what we don’t know. Above, we’re telling the AI what it might not have considered; now we’re asking it for assistance.  After describing your research problem, add a question such as:

“What additional challenges, historical events, record types, or research strategies should I consider?”

This encourages the AI to broaden the discussion beyond the information you’ve provided. Sometimes it will suggest a record set you’ve never heard of. Sometimes it will identify a historical event that affected record creation. Sometimes it will remind you of a source you forgot to check. Not every suggestion will be useful, but occasionally you’ll discover something important.

Ask for Decision Points

Research projects often reach a crossroads. Do you focus on New York records first or German records? Should you spend time on church records or civil records? Is there enough evidence to pursue a particular theory? AI can help identify these moments. Try adding:

“Identify any decision points in the research process. Explain what additional information would help determine the best path forward.”

This transforms AI from a simple answer generator into a planning partner. Instead of telling you what to do, it helps you think through your options.

Have a Conversation

Many people treat AI as a one-and-done tool. They ask one question, receive one answer, and stop. However, AI often works best as a conversation. Imagine you’re talking to a research colleague. You wouldn’t ask a single question and walk away. You would discuss the evidence, review possibilities, evaluate theories, and ask follow-up questions. The same approach works with AI. Start with a basic prompt. Review the response. Ask follow-up questions. Challenge assumptions. Request clarification. Ask for examples. The second and third prompts are often where the real value appears.

Let AI Help You Write Better Prompts

Here’s a final trick. You don’t have to create sophisticated prompts entirely on your own. Ask the AI to help. Try:

“Please improve this prompt and suggest additional context, challenges, and decision points that would help produce a better answer.”

You can also use dedicated prompt-generation tools that are designed specifically for this purpose. One of the fastest ways to learn prompt writing is to examine the prompts these tools create. Over time, you’ll begin to notice patterns and naturally incorporate them into your own work.

The Role-Task-Format-Limits structure is still an excellent foundation for prompts, and one I use regularly. But when you’re tackling a difficult genealogy problem, don’t stop there. Add background information. Explain the challenges. Ask what you’re missing. Request decision points. Treat the interaction as a conversation rather than a single question. The best genealogy research has always depended on asking better questions and working with AI is no different. The more thoughtfully you frame the problem, the more useful the answer becomes. And sometimes the difference between a mediocre result and a breakthrough isn’t a better AI program at all. It’s simply a better prompt.