AI Prompting Strategies for Home Builders - Greg Bray

Greg Bray: Today, Kevin, we're doing something a little different again.

Kevin Weitzel: We are going to have a single host and a single guest. Well, technically, you're married, so I guess you're not a single guest, but you are a solo guest. There we go. How about a solo guest? So, today we have Greg Bray with Blue [00:01:00] Tangerine with us, and we're going to be talking about AI prompting and why it's important to home builders. Is that the gist of it, Greg?

Greg Bray: That's the gist of it. Love this topic. I've been learning a lot about it, and I feel like it's time that we address it in a better way on this show because we haven't really done that before.

Kevin Weitzel: So, home builders, especially in the marketing side, a lot of them are already playing with it. Even the sales side, they're playing with ChatGPT and Claude and all the other platforms that are out there. But why is it important for them to pay attention to good prompting instead of just typing whatever comes to mind?

Greg Bray: I think that's the heart of whether these tools work or not is all about the prompting. Haven't we all, when we just played it, we threw something in there, we just asked a simple question, and we got a very generic response. There's even this term that's floating around, I've seen now called AI slop, with some of the things that people have been creating and publishing and putting out there. These tools will work better if you give them more to work with, and that's what prompting really is is telling them what you want and giving them more to work with.

Kevin Weitzel: So, what do you mean by giving [00:02:00] more context? You know, when you want to add more context in the prompt. What do you mean by that?

Greg Bray: If you think of AI as like a junior assistant, and if you had a person like that in your office, that you were trying to get to complete a task for you. You wouldn't just say, Hey, go do this if they had no idea what that was. And so, in the same way with AI, you have to explain what's going on. You have to give it, we call it context, right? What are the background parameters or the scenario that you're dealing with? Who are you trying to talk to? What is the goal that you're trying to achieve? All of these things help frame what is the voice of the conversation that you want it to have. Who's the audience? All of these types of pieces of background information change the quality of what you get out of the AI tool.

And Kevin, you mentioned a couple different ones. This is not anything that we're trying to make specific to just ChatGPT or Gemini or Claude or the other ones that are always coming out and new stuff. [00:03:00] My experience for the most part has been with ChatGPT to be completely upfront. That's where I've spent most of my time experimenting and learning. But I believe that these prompting practices we're going to talk about are common to all of them.

Kevin Weitzel: It's funny you bring up the fact that there's several platforms that you tend to gravitate toward, and a lot of people do. I actually started off more with Claude than anything else, and I've made the switch to ChatGPT, and I'll tell you why. I was getting more errors, more of those hallucinations with Claude and with Gemini than I was with ChatGPT. And truth be told, my girlfriend, she actually named hers, she named it Bob. So, when she asks Bob questions, she's proud that Bob gives her the right answers even when I can't, so there you go. So, could you give us some examples of some context of why this matters to builders?

Greg Bray: So, let's say you are trying to write a piece of content, maybe for your website or a blog article, or maybe it's a community description or some other short few paragraphs that you're trying to put on the website. The idea of context [00:04:00] is who's your audience? Who's going to be reading this? Explain that. And we're not talking about, I'm writing to people, right? I'm talking to first-time home buyers who don't know very much about the home buying process, or I'm talking to move-up buyers who probably have already owned another home.

That might change how the descriptions or the words that are used or the, the context there influences what's going to come out. Maybe you get more detailed and actually create like a full buyer persona that helps you understand who this buyer is. That's the kind of context. Context can also be how long do you want it to be. What is the output supposed to look like? Do you want this to be formal? Do you want it to be casual? These are the kinds of things that we talk about with context to help really make it clear what are you trying to get with this piece of content.

Kevin Weitzel: So, would this be something like a team could use role playing or even like a snatch and grab, you know, question with the team? You know, like, who is our target audience? How tall should they be when they [00:05:00] move into one of our homes? I know it's a stupid question, but you get the gist.

Greg Bray: It's definitely something that if it's the kind of thing that you're going to be reusing, having more of a brainstorming session within the team or even with the chat tool as well, to try to understand and define what would make a difference, would absolutely be helpful. These are the kinds of things marketers have been talking about, for example, buyer personas, for ages.

This is not an AI-driven concept of who is our buyer and what do they care about, and what are the things that matter to them, and what types of background and information would then be useful to them as they go through that buying journey? That's nothing that is new to AI conversations, but making sure that we have it available to use in these AI conversations can really make a difference.

When we talk about role-playing as well, it's also about telling the AI what role we want it to be. Do we want it to be the buyer that we're talking to, or do we want it to be the marketing expert who's giving us advice? And when you [00:06:00] tell the AI what role it's supposed to play, that changes its perspective and the type of output that you'll get as well.

Kevin Weitzel: So, could you tell ChatGPT to be, in the sales terms, the lay down, the person that just comes in and says, I'll buy this house. Could you do that, and then it give you what you need to do to get to that pathway.

Greg Bray: I think you can tell it that. It might be more useful to tell it to be the one that has a million objections, that tells you all the problems and all the questions. Well, what about this and what about that? Because then all of a sudden, you can start working on answers for those things. The great thing about it, too, is if you tell it to be, you know, kind of the problem child buyer, you can then say, Well, what would you recommend as the answer to that? You can help it play both sides as well. But again, this idea of we assume that the AI already knows what we're trying to do, and it doesn't. In fact, it is this blank slate, and so the more you give it and help it understand what kind of context, it just makes for better output.

Kevin Weitzel: Do you personally have [00:07:00] a favorite structure that you utilize when you're creating prompts?

Greg Bray: You know, I've had a couple. And again, I didn't make these up, Kevin. I've learned this from others and going to presentations and things. But I've found one that works really well. It's called CRIT, C-R-I-T, and it stands for context, role, interview, and task. Context, role, interview, and task. And so, what that means is every time you do a prompt, make sure you've done something in all of these categories. Make sure you've given it the context. We've been talking about that for a few minutes. Make sure you give it the role that the AI is supposed to take.

The interview, which we haven't really talked about yet, is this idea of getting it to ask you questions. We'll dive into that a little bit more because there's some power in that one, I think, as well. And then finally, the task is, what exactly is it you want it to do? I want you to write this, or I want you to create questions, or I want you to brainstorm 10 email subject lines, you know, whatever it is, the task [00:08:00] that you want it to do.

Sometimes we just kind of only do one of those or maybe two, but if you can get all four of those into a prompt, and I've even started putting a little sticky note with those because I'm doing enough of this now, to make sure I don't forget one as I do it. So, again, that's CRIT: context, role, interview, task. And there's others out there. And I've seen some that are kind of similar along those lines, but that one has been working well for me. Is there one that you've tried, Kevin, in the past?

Kevin Weitzel: Well, the one I try on a regular basis is baby with the bath water. I ask my questions and don't get where I need it to go, so I have to throw the baby out with the bath water and start completely over. So, what is something that somebody could do to not have to do that, to not have to throw that out, and to help debug the situation to where you're not having to start over?

Greg Bray: I know I've gone down that path before, too, and I think we all have, who've done anything significant. You ask it for something, it's not quite right. You say, No, I meant this. It sort of fixes part of it, but then it breaks this other part. And then you go and say, Well, change this. You start kind of veering off [00:09:00] course and get into this mess.

I think there's two ways to look at that. One is, and this is ChatGPT-specific, but I think there's something similar in the other tools. I'm just not as familiar with it. It's the canvas mode. So, if you're working on a long document, rather than having it output the whole document and then you ask it to make changes and it gives you the whole thing over again, if you open canvas mode, you can have the document in one side of the screen and the chat on the other side, and you can talk about specific sections of the document. Like you can highlight a paragraph in this longer document and say, Let's just work on this piece right now. So that way, it's not having to redo the whole thing over and over again. So, that's one way that you can help avoid this whole like over veering way off course here, and I need to reset.

The other one that was pretty powerful that I learned not too long ago is you can actually edit your prompt after you've submitted it. If you go back up, you put in a prompt, you get a reply. It's not quite what you [00:10:00] want; you just want to tweak it. Rather than giving it the next prompt, instructing it to change something, go back to the one before, click the edit button, and add the additional clarifications or details to what you already put there, and then regenerate.

So, instead of just continuing down the path, you go back up and edit it, and then it seems to do a better job when you do that, of kind of staying where you were and just making more fine-tuning tweaks than it does if you just keep asking it more and more about the thing as you kind of veer off course. It's just something to experiment with that I've found to be kind of a powerful opportunity to help avoid kind of those strange journeys sometimes that we take on this. We're like, this was supposed to save me time, and all of a sudden it's taken me five times longer if I've just done it myself.

Kevin Weitzel: What about formatting templates? Should you have some sort of prompt template as you would have for your social media campaigns? Is that something you would recommend?

Greg Bray: I think prompt templates can be really powerful, and at its heart, what I define as a [00:11:00] prompt template is taking a prompt that worked well and saving it somewhere so you can use it again. We don't need to start over every time from the beginning. And so, if we found this prompt that is good for helping us, you know, create a piece of content. Let's say a community description, and we've figured it out, but there's some places in there that are specific to the community.

There's information that's different if we're doing community A versus community B, but the rest of it is kind of similar. Save that out just in a Word document or a Google Doc or something like that, and have your fill-in-the-blanks. So, you take that prompt, you put in the information about the new community that's specific, but you've still got all the other things about, you know, your buyer, about your voice. And some of this could be in some extra documents that you load in, too. We can drill into that a little bit more.

But the idea is that you don't have to start over if you save these out, and you can just kind of cut and paste them and drop them in the next time you're ready to do it, or you can add in some of the pieces in this other document that [00:12:00] might be a little easier to work with. And then when you're ready to generate, you paste it into your chat and move forward from there. So, it's also an opportunity to share those across your company if you put these out somewhere, like on SharePoint or Google Drive, or something like that, where you can now say, Hey, we've got this prompt library that everybody can use.

So, we get some consistency, and we also help save each other time by going and saying, oh, here's a really good prompt that worked well for X, Y, and Z task, and here's a prompt that worked well for this other task, and now we know we can go find those and really get some benefit from them. Prompts do better when they're longer. Again, the more information that you give it, the better. But that's work. That's work to create these really long prompts, and so being able to save some time there can be really valuable.

Kevin Weitzel: Alright, so the other day I was on ChatGPT, and I asked ChatGPT to show me an image of the perfect set of sideburns on the planet, and it showed me a picture of me. So, there's two things about this. One, is I think it [00:13:00] was feeding my ego. Number two, which I don't know why a chatbot would want to feed my ego, but it did. But two, it was more of an exercise on keeping on brand.

And everybody knows my brand, you know, I wear my sideburns. You know, it says a couple things. One, it says, I realize they look goofy, and I like to have fun. And two, it says, you know, I hate shaving, but I'll do just enough to stay somewhat professional. You know what I mean? So, let me ask you this: how important is it, because a lot of times ChatGPT and the other platforms that are out there tend to be a little bit more robotic in their responses and less about brand awareness. How does a builder keep everything their brand, within their strategy?

Greg Bray: It's all about defining your brand voice. And sometimes I think when people hear that, they go, brand voice, I don't know our brand voice. How do we do that? And then how do we tell ChatGPT what our brand voice is? The easiest way to do it is to give it examples of things that you've done in the past that you like. Whether that's emails, whether that's articles on [00:14:00] blogs, and things that you've already created that have been approved, gone through the long process of editing and review, and it doesn't take a hundred of these.

If you have three, four, a half dozen of these different examples, what you do is you can drop those in either as files or they could be URLs to articles online, things like that, and you tell it, Hey, we're trying to create the brand voice of all of these, create a description for us of what this brand voice is. The tool will then parse all that and say, Oh, you have a more casual and lighthearted approach, but you stay professional. And it comes up with these different terms and phrases that describe the writing and the style of that writing.

Now you take that, you save it out into, you know, a Word doc or into a PDF, and say, these are our brand voice guidelines. And maybe there's words that we want to use. For example, we never want to use the word lot, we always want to use the word home site. [00:15:00] So, you can add that in. Very specific instructions, when you're writing, never use the word lot, always use the word home site. There may be terms that you want to make sure for your brand that matter. You create that brand guidelines document or brand voice document, and now you save it, and everybody can upload those. If you haven't figured out how to upload documents into ChatGPT or the others, this opens up all kinds of quick and easy ways to add a lot of context very quickly, especially if you have some standard ones.

The first time, I was like, people tell me they upload files, and I don't know how, and then it was like you just drag and drop them into the chat. Now there's a little menu they can drop down and attach file and all. But you just drag and drop them in there and boom, there they are, and they upload pretty quick. So, there's that way to simply take this document you've created, give it as background, and say, Hey, whatever you write, make sure complies with the brand voice of what's there.

You can also create one where you just kind of give it a name within your own thing. So, for [00:16:00] example, I took some of my LinkedIn posts that I've done over the years, you know, a handful of those, pasted them in, and say, Hey, create Greg's LinkedIn post voice and save it. And I just said, save it. Now you can't have so many things saved because you have to remember, but now I can say, Hey, write this in Greg's LinkedIn post voice, and it has that saved in my account so that it will remember that. And there's some characteristics to that. I have no idea if it's any good or not, but as far as whether it's a good voice, but it'll sound more like me when I use that, as opposed to if I'm doing something just more generic and random. And I know, Kevin, you've done some things with capturing voice as well.

Kevin Weitzel: We did not only the voice side, but we did it with our email persona. Here's the story. I'll give you the Reader's Digest version. When we look at our CRM, I have a radically higher response rate, not necessarily click-through rate, not necessarily an open rate, but my response rate is considerably higher than anybody else in our company, whether you're in marketing or in the sales side [00:17:00] of things.

So, what we did was we took about a half a dozen emails from each person. We put it in ChatGPT, and we asked it to give us back a persona, and mine came back as friendly, what you would typically think about my type of speaking persona, because I type the same way I speak. Sometimes it can be offensive, sometimes it can be over the top, but for the most part, it's still jovial friendly, and like I'm your advocate, I'm your friend.

When we applied that persona to other people's outbound emails, they saw a noticeable and almost instantaneous increase in response rate. So, we've improved everybody across our marketing and sales team, we've improved their response rate, just based on using my persona when they put their email into that template to push out what the email should read like if they wanted to be more in the Kevin persona.

Greg Bray: And Kevin, how did you share that with everybody? Where do they go to use that on a daily, regular basis?

Kevin Weitzel: On each person's ChatGPT, we put in the output, which is what that persona was, and then we just called it Kevin Persona. We'd [00:18:00] like the email to be in this Kevin persona, and then it gives those definition lines. I think there's about 14 different lines in there that kind of defined what my emails or what my persona was like, and then they can put that in, and that's the output that they get each time they do it.

Greg Bray: They have a document that they cut and paste and drop in there while they're working on it.

Kevin Weitzel: Exactly right. Put this in heaven persona, and that's what it does. It's funny because we had the same input document that we gave to each person for them to put out an output on. There was a couple of tweaks from two different people, but it was very, very minuscule, as far as the verbiage, the syncopation of the phrasing, but it was practically the same on each person.

Greg Bray: I think that's a great example for sure.

Kevin Weitzel: So, let me ask you this. Before I went into my little diatribe of my example, you had said something that kind of struck a chord of me, which is that your prompt should be longer, more in-depth, more of a chain. How can you create a chain prompt to give you better output?

Greg Bray: When we talk about chaining prompts together, it's also about recognizing that if we give ChatGPT too big [00:19:00] of a task to do all at once, you're not going to get the same kind of quality that you're looking for. If you can break it down into something more step-by-step and say, Hey, first I want to work together on getting the idea right, or the target audience right. And then, I want to work on the titles, whether that's a title of an article or maybe an email subject line. Then let's work on the next part and the next part. And so, instead of saying write an article and do all these things all at once, it doesn't do as well as if you give very specific, focused things.

And if we have time to get into custom GPTs, if you have the custom GPT and are using those, you can actually move content from one custom GPT to another just by using like the little at sign and referencing the name of the other GPT. So, I saw an example at a conference where they had the article that they wanted to write, and then they were going to do a webinar [00:20:00] invitation email about it. And they had a whole other custom GPT that was designed to do webinar invitation emails. And so, they took the article and say, This is our topic for the webinar and then they moved that into this whole other custom GPT without having to cut and paste or anything, just by referencing and chaining them together. And it was really powerful.

Now you've got workflows in place where you say, okay, I do X with this GPT, then I move to Y with this custom GPT, and then I move to Z with this custom GPT. I get different outputs, but those outputs are the inputs to the next step in the process. It suddenly opens things up for consistency when people can do these things repetitively the same way and better quality. And speeds up how quickly you get things done.

Kevin Weitzel: Wait a minute. So, you're going to sit there as the co-host of this podcast and tell me that we can put in ChatGPT our basic topic coverages and that it gives us a better weaponization to invite people versus the current method, which is the Mr. T method, [00:21:00] which is I pity the fool who don't listen to our podcast. Is that what you're telling me, Greg?

Greg Bray: I'm not sure that's what I was trying to tell you, Kevin. I'm not sure. I'm not sure that was quite what I was going for, but hopefully, the other listeners got something out of it.

Kevin Weitzel: True. True. All right. So what are some good starters, some starters for somebody that maybe isn't down that road of using ChatGPT and implementing it into their daily activity? What's a great place for somebody to start with that? Let's say a new person that wants to start getting into the custom GPT world.

Greg Bray: So, custom GPTs took me a little bit to get my head wrapped around those, I admit, trying to understand where are they, how do they work. I kind of recognized that these are like pre-done prompt background information, if you will, that you can save and reuse, and you can put a lot into those as well as far as background goes. So, for example, you can create a brand voice custom GPT. And so, within the custom GPT, you have all those [00:22:00] instructions that you were just talking about.

Where instead of having to cut and paste them from time to time, or pull them out of some other document, you can have those basic background information that says, Hey, every time we do X, Y, and Z, here are all the guidelines. Here's the words you don't want to use. Here's the voice that goes along with it. Here's the buyer persona that belongs with it. Here's examples of other background context things that can be helpful. Anytime you've got a repetitive task that you're doing over and over and over again, it might be time to think about that custom GPT so you don't have to go find all those documents and load them all in every time.

The other power of the custom GPTs is you can share them within your company account. So, now you can have the official company brand voice, for example, or buyer persona. If it's a custom GPT where somebody can adjust it and tweak it over time, but now everybody who's using it doesn't have to worry and just say, okay, anything I'm writing [00:23:00] needs to go through the brand voice custom GPT and make sure that it is compliant with that or have it be rewritten in accordance with that.

When you're using these, you still have these conversations the same way you do normally, but in the background, you've got all this other context that someone else has done. You can load lots of documents into a custom GPT, actually, and they can be really long. So, you could really get very complicated if you want to. It's really good for repetitive tasks and things where you want to share for consistency throughout your organization.

Kevin Weitzel: Greg, we could have a part two just on this subject alone, just on that custom GPT, you know, sharing, implementation, we can have a part two on that. But out of respect for everybody's time, let's wrap up today, and if you could give me just one tip that you'd like to share with our audience today that has to do with what we discussed today.

Greg Bray: I'm going to give you two because I'm an overachiever.

Kevin Weitzel: Oh, bonus. Bonus. Oh my goodness. I love bonuses.

Greg Bray: Alright. First one that we didn't really spend a lot of time on. [00:24:00] Ask the chat to ask you questions before it does its task. It's that interview part of that CRIT that we talked about, and I don't think we had a chance to nail that down. Before you move forward, say something to this effect, Interview me, ask me three clarifying questions, and pause after each question for a response before continuing.

What will happen is you'll give it the instructions, and it'll ask you, Oh, did you mean this or that? Or did you want it to sound like this, or would you prefer it to sound like that? And then you give it that answer, and it moves to the next question. But by pausing between each one, it also allows your answer to influence what the next question is as you go through that. I have been really surprised at how perceptive some of these questions are when I do that.

 It's like, oh, yeah, that's a great point. I'm not sure. Or, oh, yes, I didn't make that clear. I definitely want it to be the more formal answer versus the more casual answer or whatever type of thing it might be. So, [00:25:00] get it to ask you questions. Three is pretty good at getting a lot of additional information out of you. And you don't have to be short on those answers either. It could be like a whole other several sentences that you can give in the answer if you want. That's my first tip.

My second tip is I've started playing with a tool called Wispr Flow, W-I-S-P-R, flow, and it is a voice-to-text where you can basically talk to it. Now, you can do this in the ChatGPT app on your phone, but it's a little harder to do on your desktop. But this Wispr Flow, all you do is check in the input box. You make sure your mouse is in there where you would start typing, and instead of typing, you can talk. You just hold down the key, and you just explain what you're trying to do.

ChatGPT doesn't care if you're talking in fragmented sentences, if it all flows together, whatever, but you give it so much more context when you just explain to it like you and I were talking and I was trying to get you to do something, or you were trying to explain to me how you get great sideburns or [00:26:00] whatever it is. We don't talk in complete sentences, but when somehow when we sit and type, we get stuck, going, oh no, that's bad grammar. Or, oh wait, I misspelled that word or whatever, and it takes us a lot longer.

We can talk like three to four times faster than we can type. You just dump more into it, and again, using the voice instead. And so, I would say, Hey, if you haven't tried it, they've got a free trial. No affiliate commission or anything coming this way on this one. But this Wispr Flow tool has been a real game-changer in the amount of context I can quickly drop in. So, those would be my two things. Get it to ask you questions and then give it more information faster by using voice versus typing.

Kevin Weitzel: This has been helpful, not just for the home building industry and not just for marketing industry, but I mean across the whole spectrum. So, many people can learn from this. This one definitely crosses many, many channels. Greg, this has been great, and thank you for your time today. You know, if anybody wants to get in touch with you about AI prompting or anything about what we talked about today, or even, you know, websites or SEO or PPC or anything like geofencing, how would they get in touch with you, Greg?

Greg Bray: [00:27:00] Email address is greg@bluetangerine.com. That's the color in the fruit, blue. I'm also easy to find on LinkedIn. There's only a few Greg Bray's out there; feel free to reach out. Love to be helpful. See if we can, you know, all learn together because this is still a growing area that we're all trying to figure out. So thanks for the opportunity to share some ideas.

Kevin Weitzel: I appreciate you. Well, hey, thank you for joining us on the Builder Digital Marketing podcast. I'm Kevin Weitzel with OutHouse.

Greg Bray: And I'm Greg Bray with Blue Tangerine. [00:28:00]