How to Write Better AI Prompts for Title Insurance | Ep 93
Episode Summary
Mo Choumil and COO Hope Arviani teach title professionals how to communicate effectively with AI language models like ChatGPT and Claude. They explain why vague prompts produce poor results and demonstrate how specificity, context, role assignment, and output formatting dramatically improve AI responses. The episode includes real-world examples for title searches, client emails, and title insurance explanations. They emphasize treating AI outputs as first drafts, not gospel, and share strategies for comparing results across multiple models to achieve optimal outcomes for daily title operations.
About Hope Arviani
Hope Arviani serves as Chief Operating Officer at Alltech National Title, where she oversees operational strategy and innovation initiatives. A licensed title professional specializing in Virginia markets, she has rapidly become an AI adoption advocate within the title insurance industry. Hope focuses on translating emerging technologies into practical workflows for title processors, examiners, and agency operations teams. She co-hosts the AI Corner series on the Title Agents Podcast, demystifying artificial intelligence applications for title professionals nationwide.
Key Takeaways
- Vague AI prompts produce vague results—specificity in your request directly determines output quality, just like giving clear instructions to an employee.
- Effective prompts include five elements: role assignment (act as a real estate attorney), context (the situation), goal (desired outcome), format (bullet points, email, Excel), and constraints (word count, tone).
- When reviewing title searches with AI, upload the document and specify exactly what to analyze—legal descriptions, easements, chain of title issues—rather than asking it to generically ‘read’ the search.
- Use AI-generated content as a first draft, not final output, because you remain the expert who must verify accuracy and prevent hallucinations.
- Comparing responses across multiple AI models (ChatGPT, Claude, Gemini, Grok) by playing them off each other often produces superior results than using a single platform.
- Claude currently outperforms ChatGPT for title professional workflows and proactively asks clarifying questions when prompts lack specificity.
- Memory migration between AI platforms is now simplified through Markdown files, eliminating the primary barrier to switching from ChatGPT to superior alternatives like Claude.
Episode Chapters
| Time | Topic |
|---|---|
| 00:00 | Intro and AI Corner series overview |
| 01:45 | Why most people get bad AI results |
| 03:20 | Prompting for title search analysis |
| 05:15 | Context, role assignment, and formatting |
| 07:30 | Real-world prompt examples for title professionals |
| 09:10 | Using multiple AI models and memory migration |
| 10:45 | Wrap-up and next episode preview |
Full Transcript
Show Full Transcript (2,051 words)
In a world where change is the only constant, Mo Shumil stands at the forefront, guiding title professionals to not just grow their businesses, but to master the art of innovation. With every episode, you're handed the keys to unlock unparalleled growth and stay ahead of the curve. Get ready for a transformative journey. Hello, everyone, and welcome back to the Title Agents podcast. I am your host, Mo Shumil, CEO of Alltech National Title.
And today I have my co-host is our fabulous COO, Hope Arviani. Welcome, Hope. Hi, happy to be here again. So in our AI series or AI corner, we want to provide education and simplify artificial intelligence to kind of take the fear out of it. And then last episode, we had AI 101, Artificial Intelligence 101, just kind of explaining what is artificial intelligence, what different models from ChatGPT to Claude to Gemini to Grok.
Today, we want to teach title professionals how to communicate with a large language model, the ChatGPT, Claude, or LLM, that's the abbreviation we're going to use, effectively, we want you to leave knowing what a prompt is, why specific specificity, that's a tough word, matters, how to immediately get better results from the tools you already have. So Hope, in your brief experience, and again, in your kind of getting really entrenched in AI and the smile on your face, again, you and I are very passionate about it and the possibilities and this. So why most people get bad results like with AI? Because they're not specific enough. And I learned that the hard way as well, like when I first, it's like an employee, right?
When you hire an employee, you're not just going to tell the employee, hey, do this. No, you need to be very specific on what you're trying to get out of it, the results you want out of it and who you are, like what the purpose of the request. So you have to be specific, you can't just be generic or, you know, short. It's like asking the question in a prompt, help me with title insurance. That's very vague.
It's like walking to a closing and I'm like, hey, do the paperwork, do the title on a consumer or else do the paperwork. That's very, kind of very vague. So really the output is only as good as the specificity and the details you ask. Give us an example, when we've been using prompts for title examination, kind of as a, like get some results. How do you prompt it when you want to ask about a title search?
Yeah, and I've learned this too, like, and recently, honestly, when you upload a title search, for example, and you're like, hey, read the title search. Well, that's not going to help me. That's like if you went to an underwriting counsel and said, hey, read the title search, what's up with it? You got to pinpoint what you want to look at. So if I knew, and I'm licensed, I'm going through the title search, and I knew that there was an issue with the legal description or the chain of title or whatever the specific is, then I'm going to upload the title search and say, please review this title search on this property in the state of Virginia and focus in on the easements or requirement or whatever it is that you want to focus on.
And then give me the high level red flag or whatever, give me specifics of why the legal description does not work, why the meets and bounds doesn't make sense, whatever it is. And then it'll spit out a beautiful result. And you can compare and contrast to like, hey, review title search and you'll get a very ambiguous answer or, you know, give me the title search and give every specific. Sometimes when I write, it's like a paragraph because I want every little piece covered so that I don't miss anything. Yeah, it's like another example, like a weak prompt, write me a follow up email versus write me a follow up email to a Chicago real estate attorney named David who attended our closing last Thursday.
The deal had a complex lien issue, which will resolve at the table. Keep it under 100 words, professional, but warm and with an invitation to connect on a future deal or coffee or lunch. That's way different than write me a follow up email. Way different, yeah. You got to give it context, yeah.
So context is very important. You have to give it context and you have to give it a, what's the goal, the outcome you want and also with another way to prompt, to get a really good result, to give it a role, like act like a world-class attorney in real estate or a world-class litigation attorney or a world-class marketing strategist. The context we're trying to do. Then the next thing is, how do you want the format to come out? Is that a bullet point?
Is it very detailed? Is it an Excel format, a Word format? The more details that give it, the better the outcome you get. And I just learned that too recently from you, Mo, is like, tell it what it should be pretty much, like tell it, you know, train it. It's like the nature, it's when they upload Jiu Jitsu to you, you're like, I want it to be, this is the outcome I want, here's the result I want.
You get to tell it, I am a COO, you are a COO in a title company or whatever. And that's very helpful. Like another, another prompt for a real life scenario for a title professional is, I'm a title processor in Wisconsin, a buyer's confused about why they need title insurance when they already have homeowners insurance. Write a two paragraph, plain English explanation. I can paste into an email, keep it under 150 words, friendly tone, no legal jargon.
The result to get out that email is so, it's just sometimes a beautiful, this is perfect, I couldn't write it, I couldn't write it any better. Same, same, I'm like, man, it's going to put me out of a job. Another best, best practice, like the top AI professionals or experts, they use the first draft, the first response as a draft. Like you don't take, as I said, sometimes AI hallucinates or doesn't understand it clearly, that's one, I don't want to jump into tools because we, you and I both love, love, love Claude, Anthropics Claude is just another, another level. Like Claude will just ask you questions, like if it doesn't understand something, but love, but you want to use, at the end of the day, you are the expert.
I mean, all a GPT or Claude, they just kind of has massive knowledge, database of knowledge, they go through the scan very quickly, but at the end of the day, you as a title professional, you are the expert, you know better. So don't take that first draft as gospel, just to look at it, see if you like it, put flows, ask even, uh, different, uh, questions or kind of, uh, we want to. Fine-tune it. And sometimes I even literally get a play between different models. I have a subscription to all of them.
Like the, they're all like 20, 200, 200 bucks a month. What else? Literally I have Grok, TachyPT, Claude and Gemini, they're all paid subscription, not a free version. Sometimes they all got a different thing. So sometimes you, you don't like a result or you want to get a better, like, Hey, TachyPT, this is what Claude said.
What do you think? I had to pay them off each other. It's, it's a beautiful thing to watch. So you had the outcome you'll get. That's funny.
And TachyPT gives you two options. Have you seen that now? Like, do you like this result or this result better? And it's just different formatting. Um, but TachyPT just, uh, TachyPT is just like social media.
It wants to keep you in the end, kind of ask, Oh, you want a better version of this? Like, why don't you give it to me the first time? Right. Like, why do I have to ask for it? Like literally trying to keep you in the platform, which is done by design because Claude is way, way ahead of TachyPT.
So they're trying to get you glued to the platform. Yeah. Yep. I've seen that too. I think they just came out with Entropic too, uh, TachyPT.
Entrop, Claude is Entropics. Oh, well they said like down, um, when I was in TachyPT there, they was like, um, download and drop. Oh, you migrate the memory. So basically all you, the, what a biggest fear for people to migrate between, uh, TachyPT to Claude or Gemini. Is that all that memory?
It's a, all AI is literally memory files. It's called the, the new term you need to learn is MD or Markdown file. Markdown file. That's the new, the new format, the text formats, basically. So, um, the hardest thing in the past are using the new platform.
Like you all will have all this knowledge base, all this communication over the last three years or whatever. And I've been using TachyPT since literally week one or week two. That's late 2022, October, 2022, when it first came out. Yeah. So that was kind of a big fear for people to migrate from one platform to another.
Claude, make it very easy. Here's directions, how to kind of, uh, download the memory from TachyPT and upload it here. It's like. It's crazy. Those are the index, right?
Index.html. It's like the memory. Yeah. I've noticed that too. It's so cool.
Uh, anything you want to add as far as prompting, we want to literally keep these, uh, we're going to do a bunch of series at this is very, very important is you have to kind of jump on this bandwagon, your life will change literally for productivity, the knowledge, the speed, you'll be able to do stuff. And it's just very exciting. We want to kind of, uh, literally segment by segment. So we don't overwhelm you. I think you want to add hope, uh, to as far as prompting.
Yeah, I think we covered a lot of it. I just be specific. I've learned over the last two years as well as how to be more specific. And then sometimes I'll just say, rewrite it when I'm writing an email for like an angry climate, rewrite it. And I'm like, and I've learned, like it'll rewrite it in a totally different format, but I want to say, rewrite it in a professional soft tone or rewrite it.
So it's just being specific and prompting it the right way for what you want out of it. It's just like, if you're training an assistant, an executive assistant, how do you want them to be trained? Do you want them to be cloned just like you? Yes. Yes, I do.
So then that's exactly what Claudette is. And, uh, no, yeah, now you know how to actually talk to AI and get results you can use, but here's, uh, what we, we have not covered yet, which tools are actually worth your time. We look at a little hint, hint, hint, Claude. There are hundreds of AI tools out there right now, claiming to transform your workflow. Next episode, we'll cut through the noise and show you exactly which ones Tyler professionals are using right now and what they are using them for.
You're going to want a notepad for next one, or just record the episode, have it transcribed it for you. How smart. Well, thank you everyone. We'll see you next time. Bye.
In a world where change is the only constant, Mo Shumil stands at the forefront, guiding title professionals to not just grow their businesses, but to master the art of innovation. With every episode, you're handed the keys to unlock unparalleled growth and stay ahead of the curve. Get ready for a transformative journey.
