Testing Persuasive Messages Using AI

NOTE: This is a raw transcript and contains grammatical errors. The curated transcript will be uploaded soon.

Fantastic. I’m going to start. Thank you guys for helping us as our impromptu tech support group. We’re going to be talking about the science of persuasive messaging, and we’re going to be talking about in the context of using AI and AB testing to improve the messaging on our website. As always, our goal with messaging is to improve the performance of our website. So that means that when people come to our website, they get what we’re about. They get what we offer and they get the reasons why they should continue to explore our website, to explore the products and the services that we offer.

Persuasive messaging also ensures that those people who are in the wrong place leave quickly if they’re not supposed to be there or if they’re not in our target market. They get that quickly and they can move on and look elsewhere for the solutions that they’re looking for. And, you know, when you’re marketing and branding, that’s usually the hardest part is accepting who is not in your group, even though like maybe some of those people would buy and really focusing your messaging on who your target market is. Because as we always say, if you’re marketing to everyone, you’re marketing to no one.

My name is Brian Massey. I’m a I’m the senior conversion scientist at Conversion Sciences. I founded Conversion Sciences way back in two thousand and seven. We have, been testing and optimizing websites ever since. We use a steady diet of analytics, very nutritious user input and customer research and experimentation. So primarily AB testing, which is really the best data that we can collect to tell us if our ideas are really as good as we think they are or if, in fact, our visitors aren’t interested in our hypotheses.

So first, a cautionary tale. I’m going to assert that AI will be the third person in every relationship.

And this year I was I was dating a lovely woman and we were texting while I was on the trip. And I didn’t realize that she had her own chat agent that she named Charlie. And she was putting my texts through Charlie and asking Charlie just how sincere I was.

And I’m a little bit embarrassed to say I didn’t fare very well.

The chat, you know, said that I was more focused on keeping you on the line than actually acknowledging the impact my actions have on her. I was using classic bread crumbing. I was not meeting her with vulnerability. I was not acknowledging her needs.

This sounds pretty terrible. And the agent even recommended that it could drop a chat in which she could break up with me and you could make it short and strong or graceful with a bit more closure.

This was my first real experience of AI impacting my personal life. Of course, we’re all over it on the business side of things, but I think that this is an indication that AI is going to be the third person as well.

So here’s a study from Pew Research. This is a focus on the AI summaries that, Google and other search engines are offering. And it is showing that when, people see these AI summaries, much fewer of them are clicking through to websites and more of them are ending their browser session in the chat window. So in the search window, rather.

Both are indications that not only are they not clicking through to your website, but they’re getting the answers that they need from the search pages.

We’ve been kind of tracking bot traffic. This is a combination of AI bots that are coming, doing, scraping our sites to do training, those that are coming, people that are coming from search. So from these AI search bots.

And then also a small portion of this, I suspect, but something we really want to pay attention to is agents that are doing work for our visitors. So instead of them coming and visiting all of our sites and deciding who gets on their shortlist in terms of what they’re going to buy or what services they need, they’re going to have the agent do it. And I’m going to take you through an example of how deep research is making this really, really powerful and really, really easy. And you’ll see how you can use these deep sea research reports when we start diving into messaging.

So this is a, you know, this is a classic hockey stick and something that we should be watching.

We’re also seeing the shrinking organic traffic. I don’t think that a large portion of this yet is from AI, but it is as as AI starts to take more and more of the traffic down, it’s going to become more and more competitive. And so we’re going to get less of that sweet organic search traffic that we’ve been working so hard to get.

And just as an interesting aside, AI influencers, completely generated influencers, for those of you that are in the influencer market are coming online. So you’re going to be able to not go out and you’ll be able to go out and find almost any sort of influencer that you want because they can dial and change that influencer at will.

The other thing, of course, that we’re looking at is when is AI going to start coming for us, the CROs, the optimizers. There are a number of services out there that will let you type in your URL for a landing page or a home page or will let you upload an image and it’ll start giving you heuristic ideas, best practices ideas, make the button pop more, change the CTA, change the headline, reorder the content. You know, some of the things that we do when we go in and start working with a website, that heuristic best practices approach. Now, Baymard, who I think does a fantastic job on customer research and is a great source of hypotheses for our clients, said that they looked at those tools and found that they didn’t really have high accuracy rates. And this is based on Baymar’s research.

Of course, Baymard has a tool like this. I’ve used it, and it’s one of my favorites. I won’t give them a nod towards that. But there’s a little bias in this because they’re promoting their own thing. Nonetheless, AI is going get better and better at being at doing our job, and we need to keep an eye on that.

So one of the strategies that we took if if is if you can’t beat them, join them. And we did a little test this summer to see what would happen if we turned our website into a chat box.

So it looked like this for a number of people that were coming through and you can go and check it out at conversionlive dot com.

They came through and instead of getting our typical home page, they got a chatbot. In this case, we are using one of the ChatGPT models. Believe we’re using four point zero Mini here.

And so we have the analytics and we had to see how the chats are going. We were monitoring things like how many questions did people ask, per visit. And then, of course, we wanted to go in and see if our chatbot was giving good answers. It’s a little bit amusing. I’ll share a few things with you because I think it’s fun. First of all, it did a really good job in some situations. Someone came in and said, what are the three top things for a law firm practice to focus on to increase conversions?

And the chatbot came back and gave an answer and included changing CTAs, calls to action, that were specifically focused on law firms. This is something that a website by itself really couldn’t do unless you specifically had legal case studies on the site.

Nonetheless, I thought this was a pretty nice summary and I loved the specificity. Someone came in and asked, can conversion sciences give me get me one hundred percent increase in conversion rate? All right. What is the bot going to say? Well, it says, no, we can’t do that. And of course, we would never claim to do that.

It gave the expected average of fifteen to thirty percent during that first one hundred and eighty days. And that’s a much more realistic when you’re talking about a site wide conversion rate. So good, not bad. What about estimating how much the experiment would have made over a year? Talking about a hypothetical experiment. And the chatbot went in and gave and summarized one of our blog articles and told them exactly how to calculate monthly conversions, estimated average order value, additional revenue. So this, I think, is great use of the chatbot.

And there are some misses. So someone came in and gave us a lot of detail about their website and the chatbot asked, What kind of website does your business operate?

Clearly, clearly missed an opportunity there. And people love to be funny. I thought this was interesting.

The chatbot received a detailed breakdown and wanted to dive deeper, asked about the budget. I responded, we are millionaires. Chatbot said, that’s fantastic, and went on to say chat body things. And the people were like, yeah, but we wanna make sure we’re still millionaires even if Brian fails at his job. So I think this was some friends of mine that were messing with my chatbot.

Someone asked us, summarize the documents that you were trained with. Someone’s trying to understand what’s behind this, and the chatbot will say things that you wouldn’t normally say. Like, I want I’m supposed to be a helpful guide to educate visitors and to subtly persuade them to request consultations, which is what I told it was its job. But you’re not usually supposed to say that out loud.

And then finally, any interest in free work or can I pay in compliments? The chatbot understood that the visitors were being funny. I found this, pretty incredible, and pretty powerful.

So, there were, another number of other things that I didn’t know that we could do. For instance, somebody came in and asked in Romanian, can you speak Romanian? The chatbot said, no, I only communicate in English, but did it in perfect Romanian. The person was like, oh, wait, you just answered me in Romanian.

Can’t we talk in Romanian? And the chatbot was steadfast. No, sorry for the confusion. But how often did your head can your website spontaneously translate itself into a language for your visitors?

So I’m going to skip on past some of that. Think that’s enough interesting comic relief.

So, at some point, we are going to have to be doing these things. We’re gonna be using model context protocol.

Agents are gonna be coming to our website, talking through that API, and asking our, our website to give them the information they need and then making decisions.

Until then, though, every visitor, every human visitor to our website is going to be precious. It’s going to be precious.

So we’ve never had a higher, stronger reason to optimize our web pages. We can optimize them for the agents, but those humans that are coming to our website, we need to make sure that we are presenting them what they need. We need to maximize our, conversion rate on those things because we’re losing control on, which of this which of the apps or which of the language models, what they’re presenting to our visitors.

So the first thing we usually do is we we develop personas in ICPs. The language models are spectacular in helping you develop personas and ideal customer profiles.

It helps you as a marketer understand how you should speak to them, what sorts of features and benefits you should talk about and things like that.

The problem is that as humans, we really can’t keep all of our ICPs in mind.

So they’re great for creating the messaging like you develop those personas and ICPs and you do your first pass on the copy.

But after that, as you’re refining and rewriting or let’s say we want to make our messaging more powerful, more persuasive. So you’re talking about rewriting your messaging different ways and testing that.

And every time you rewrite that, are you able to keep like this is one of our personas, Blair, in mind that he must know, can I trust you? What are my competitors doing? If I hire you, how much work will be required for me?

How will I know CRO is working? So all of these questions. But the thing is, there’s not just one persona. These are the four personas that we use. So when us or our copywriters go in and write, are we keeping all of these people in mind?

No. As humans, we really can’t do that.

Persuasive copy also means going beyond features and benefits.

So we really want to understand the research mode, the way people are coming. And I’m gonna show you a very simple model that you can understand and that you can use to communicate with the language models to develop these very different variations.

So first thing I wanna do is I’m gonna do a little experiment. If you would, pick up your phone or if you on your if you wanna open a browser and go to menti dot com, what I would like to do is collect a little bit of data.

And the question that you’ll be faced with is this one. So when you go to menti dot com, if you’re not using the QR code, type in eight five eight six four six two three.

Eight five eight six four six two three.

And hopefully, you guys had time. If you were gonna use the QR code, you had time for that. Maybe I’ll just pop backwards one more time and make sure I’m gonna do the same thing here.

The question we’re asking is how do you prefer to learn? Oh, here we go. We’re getting some results. So do you like to understand the underlying concepts?

Do you like to research and get all of the details? Do you like to dive in and get your hands dirty and try things? Or do you like to look for knowledgeable people who can teach you how to do things?

I, just to let you know, I’m a getting my hands dirty sort of, I think, guy.

And it looks like we’ve got a pretty strong group of well, I’m showing what other people are, yes. Exactly. Thank you, Georgia. Showing the results. So you guys might be influenced. There might I might be skewing the results towards this, getting your hands dirty thing by showing you the results because we’re social animals and subconsciously we think that we’re judging things independently, but our brains are designed to take shortcuts. And this, by the way, is why it’s so hard for us to rewrite messaging that’s different is because we just tend to write the way we write, even when we think we’re not.

Well, I think we’ve got a pretty good idea of the makeup. The vast majority of us are getting our hands dirty. A nice nod of us are researching to get all the details and then another group of us for studying the underlying concepts. Now I can use this to kind of change the way that I go about.

So this is a very hands on approach. This is a very hands on presentation. So I’m glad that so many of you are like that. So let me explain how we think about these different ways of learning.

You can imagine a grid and this, by the way, is from Jeffrey and Jeffrey and Brian Eisenberg, Waiting for Your Cat to Bark.

And what they said is that there’s four two indices that we can look at. Do people are people researching in an emotional mode or a logical mode? And are they going to make quick decisions or are they going to make deliberate decisions?

And the Eisenbergs gave them names. So those that are going to be logical, quick decision makers, we call them competitors. It’s important for you to introduce what’s in it for them. They are looking for ways to make themselves better.

They do want to understand the underlying concepts that go into this. If you, you know, make a claim that you are faster, better, easier to use, one hundred percent organic. You’ve got to prove to them why how you did that. How did you make it easier?

How did you make it easier to use? How did you make it faster? Whatever you’re whatever you’re selling.

Those that are being in a logical research mode and make deliberate decisions, we call them methodical. They need to understand the processes, your processes. How did you become an expert in what you do? How did you go?

What was the process that you went through to, make your product better than everyone else’s? They want to know the details. They are not going to get on the phone with you. They’re not going to get on a demo with you.

They’re not going to buy until they think they already know all the answers.

Then the spontaneous folks. There’s a lot of you here.

Today, they make decisions or they’re in research mode where they’re making decisions emotionally and quickly.

Think about it this way. When someone has done all the research and they’re like, Okay, I want to go with this company or I want to buy this product, they want to come in and just buy. They don’t need any more education. They don’t need any more justification. You just need to be able to tell them, here’s where you buy, here’s how you choose the size, the color, or here is how you arrange that consultation or that demo. They are just looking to take action. Some people do this through their entire learning process, as I suspect some of you guys do.

And then the humanists. So the humanists are are gonna in a mode where they’re doing emotional decision making and deliberate decision making. They want to know who you are. They want know how they’re going to feel if they buy your product or work with you. They are about connection.

And so these are the people that read your about page. These are the people that are paying more attention to testimonials and looking for people like them who have used your product or services.

So this is a great way of modeling things because when we’re doing persuasive messaging, we need, when we’re testing, we need very different bodies. We can’t really go into our copy and change a few words and expect to detect, an an effect. We really do need to change the mode.

And this allows us to, you know, even address a variety of, research modes on the same page. Here’s a quick, way that you can take these different messages and lay them out on a landing page or your home page. At the top, of course, those spontaneous, quick, emotional decision makers, just need a way to take action. This is why it’s often, successful to put your, your form in the hero area for those quick decision makers.

Your headline and your subhead has to tell the competitors they’re in the right place and give them a reason to keep reading and keep scrolling. So this is where you gotta really hit on what you do in your headline and in your subhead. It’s why it’s going why you’re better, what’s going to improve the competitive world. And then the methodicals and the humanists, since they’re more deliberate in their their research, they’re going to scroll.

So we typically put methodical and humanist content further down on the page, and they generally require more content.

If you hook that competitive, they’ll scroll and do more research, but you gotta make sure that you do that at the top of the page.

So the challenge is how do I write like someone else?

This is what’s between our ears that keeps us and our teams from really writing good copy. We are going to write like ourselves. And the two great biases are I should never do something that didn’t work for me before or I should always do something that worked for me before. This may not be true.

As you’re writing, you’re writing for what has worked for you. But your audience today may not want that. They may need, in order to feel comfortable and confident converting, they may need what you’re not putting on the page. Your biases are getting in the way.

Language models don’t have this problem. You can tell them exactly, who they’re supposed to be, and they will follow those rules. You can tell them to have biases, but I don’t know why you would do that.

So what we wanna do, what we need is a way to talk to the language models that communicates these different modes of, research that we have in mind. This this this is really gonna help us get different copy out. And it turns out that there’s a way to do that. So Myers Briggs has been around since nineteen sixty two. All the language models have been trained on, years, decades of research. They understand the language of Myers Briggs. As a personality type, Myers Briggs has fallen out of favor and, in favor of the the big five or the ocean, which, is better, across, genders, better international effects.

It’s just, what psychologists are using to understand how people process things. We’re not interested in understanding our visitors’ personalities.

We’re interested in their research mode when they’re coming to our website, and this is a great model for communicating with the language models.

So looking at our grid, we find out that the competitors have an n and a t in their Myers Briggs their Myers Briggs indicator. So that’s intuiting and thinking.

The methodicals are sensing judging. These are the, you know, they they believe there are certain rules for how things are due. These are the gatekeepers. They have processes and they guard those processes. And that’s why when you flip your switch on, the lights come on. The spontaneous folks, which most of you guys are, they’re sensing, perceiving. They’re very outward focused and they follow what they see in the world to build excitement and solve their problems.

And then the intuitive feeling humanists.

These, as I said, these folks, they have an internal frame of reference, but it’s how they feel more than how they think.

So we can use this to map our, modes of research into the language models.

So let’s ask the question, how do we write copy for different styles? Well, the first thing I would do is if you’re not familiar with, this, I would you could ask the language models to tell you. Say, imagine that your Myers Briggs type contains an n and a t, intuitive thinking. How would you like to have information written for you?

And the language model will come back and tell you. The NTs, which are competitors, they like underlying theory, systems, and principles. They like precise language, clearly and systematically, explained things. They explore complex ideas and nuances, visual models, charts, graphs.

What if we ask the language models to tell us about the SJs? They like organizational elements. They like headings, numbered lists, bullet points, real world examples. These are the guys that are reading your case studies. They need the evidence, practical uses, and it goes on. So if you’re curious and want to understand these different modes of research, then, the language models can help you out. Fortunately, you don’t really have to understand them because we’re gonna let the language model be the expert.

So this is a block of copy on one of our older versions of our website that I wrote. And I went through and I said, well, what mode do I write in? And I went in and I saw that sometimes I’m writing in competitive, sometimes I’m using humanist words, Sometimes I’m using spontaneous words, and it’s a mishmash. Our proven scientific methodology mixes brilliant solutions with data driven decisions that deliver guaranteed revenue increases. You see how I’m kind of switching voices. What did I say earlier? If you’re marketing, if you’re communicating with everyone, you’re communicating with no one.

So I want to just kind of give your ear a little understanding of how the language changes for each of these research modes before we start doing this on one of your sites. So let’s let’s rewrite this and really focus on that humanist person.

So instead of the conversion optimization agency, it says making connections that matter.

And it goes on to say, we believe what optimization should be more than numbers. It’s about understanding people and creating meaningful experiences. So are you hearing that NF, that humanist who’s looking for connection? Our methodology combines analytics with empathy.

We’re uncovering their hopes, dreams, and desires, forging authentic connections.

This is strikingly different, and I think almost anyone can hear this difference.

Now, as we’re doing this, you’re going to have initial gag reaction because you don’t write like this.

You’re going to think, well, I I would never respond to this. Me, I, well, I tend to be a little bit humanist forward myself. So this might appeal to me. We are trying to get past that gag reflex so that you can test messages that your visitors might need.

So I’m going to encourage you to, as you’re coming out with these messages, fix what is inaccurate. But for the test, go ahead and use the content, the majority of the content, the language models are building out so that you can see if you have a large group of visitors that are coming in this research mode. Here’s another example. Let’s look at those methodicals. They make decisions, deliberately and logically.

Results driven optimization. Our proven methodology delivers efficient solutions and guaranteed revenue lifts through rigorous analysis. We talk about rapid and reliable business grow growth, actionable steps to increase conversions, research, transparent reporting, consistent optimization. Can you hear it? Can you hear how different it is?

Many of our B2B websites write in this mode. So this may sound more palatable to those of you that are doing b to b. But I’ll tell you, you know, testing shows that you’re missing out on a number of other voices.

Here’s an extreme example. So I had to, do this for the SPs like most of you guys. Unleash your website’s potential. We think outside the box to find bold new strategies that will take your website’s performance to new heights.

This had a severe gag reaction to me. It’s making promises that we wouldn’t normally make.

And so we would maybe test this one last.

So again, here’s the model that we can lay this out. So if you have different modes of writing, you can lay them out like this.

To tell you the truth, we’re, testing a number of things like putting the same rewritten paragraph multiple times on the page. So it’s essentially repeating what is being said, but we’re testing if the different segments of visitors will self select themselves and read the paragraph that most appeals to them. It’s a fascinating time. And we’ve never had this capability prior to the language models.

So I would invite you to open your favorite AI chat and can play along if you would like. I have the prompts in a a document that I’ll be able to send you, after the, after the the presentation, along with these slides. But choose a landing page or a home page that you would like to follow along with.

And I’m gonna follow along with one of our examples. So this is a company that, offers, human resources services to other businesses. They had two primary personas. One is the chief HR officer and one is the chief financial officer.

And just kind of summarize their personas, CHRO, the HR, the chief HR officer, Her challenge is balancing strategic HR initiatives with genuine care for employees versus the CFO, which focuses on profitability and risk mitigation and is actually was called out skeptical of HR’s value. Are they, you know, a little too, you know, flower child versus the methodical CFO? And so I imagined the two voices that I thought would be the first to test would be that humanist voice that genuine care for employees, and then the methodical voice, which the CFO follows.

I went in and asked the language models to remind me what each of these are. The humanist values, empathy, personal growth, emotional well-being, harmonious relationships, and that methodical SJ, value stability, risk management, practicality, and efficiency.

I chose a portion of their HR landing page. So this is the landing page that’s targeted to the CHRO. It’s targeting to that humanist.

And I took a section here. These are the, what we call claims. These are the claims that need to say something relevant and then a paragraph that proves that they can do that.

So I first thing I did want to do and the first thing that you want to do is find out how you write.

So you can take you can take the entire page of copy or you can take a section of the copy like I did here, and we’re gonna ask the language model to tell you what style we write in.

So the first thing, I do is I teach the language model how to use this language I’m using. So instead of constantly typing Myers Briggs s j, Myers Briggs n t, I go ahead and tell it that we’re using Myers Briggs to group our, website visitors. If they have an s j in their method in their type, they’re called methodical. If they have an n t, they’re called competitive, etcetera. So this allows me to say, write rewrite this for methodical. Rewrite this for competitive.

You can request, these prompts and the slides, using the QR code that’s on the screen here, and I’ll show you this QR code again.

And then I’d say I want to evaluate the copy on my website to see which of these groups my writing style appeased in appeals to. Please analyze the following copy. Tell me which of these four groups it is most likely to appeal to and why. And then I paste in, the text from that section of the page.

So evaluation of the copy. Here’s what it said. The website copy you’ve provided appears to align most closely with the methodical group. So, again, this is our example where we’re on a landing page that’s targeting the HR person, which we think is a humanist, but the copy is, like the rest of the website, very methodical.

So this could be a problem. What we found out is that on our research mode graph, we’re heavily into the methodical world, and it’s a fantastic hypothesis that we may want to make this more humanist to appeal to that chief human resources officer.

So I go in, I say, Rewrite the copy provided to appeal to the humanist group. So it takes the headline Minimize risk. Work with a U. S.-based team that understands federal, state, and local labor laws, reducing the risk of legal issues and penalties.

And the rewrite says, Protect what matters. Partner with our caring team who understand the human impact of compliance, protecting your people while fostering a workplace where everyone feels secure and valid and valued.

Can you hear the difference in that?

And again, the team that wrote this will probably have a gag reaction. It’s like, we don’t write like this. But you produced the persona that said that this is where that CHRO is. This is what they care about.

The language model’s done a fantastic job here. Here’s another one. Free up resources. Spend less time on payroll benefits and other HR functions.

Focus more on the strategic priorities that drive your business forward.

And the humanist version is unlock true strategic potential. Okay? Free yourself from administrative burdens to focus on what truly matters, supporting your people’s growth, build meaningful workplace relationships, and bring your organization’s values to life.

Can you hear the difference? This is very different copy. If there is a group that wants to be spoken to to this, they will show up in an AB test. We you will be able to detect that.

So, designing the test, you would take your control group, which is the the copy as it is, and and you would set up a test in VWO, to change that copy. This is something that you could do using the WYSIWYG editor. It’s a very simple test to run and, run your test. The conversion goal on this is people scheduling a services team consultation and, let it run and deliver the results.

You can also use this framework to rewrite copy, to appeal both to the methodical and to the humanist.

So you could say, write it for both.

I I don’t have any data that says that this is not a good approach. I think maybe for two of them, it’s okay because you end up with paragraphs that say partner with a dependable, compassionate US based team. So dependable CFO, compassionate CHRO, creating a safer and more equitable workplace, minimize risks of legal issues, and create a safer, more equitable workplace. So you end up each sentence kind of has a this and that to it, which can work.

But as I said, as you start to water down your your copy and try to appeal to more people, you end up not appealing to fewer people. So I would test first, having playing these out and having a, a methodical and a humanist in the case of this example, separate from each other that they can find their way to.

You can also ask the language models to simplify things. So we know that, the methodicals like bulleted stuff that they can scan through. They like, frequent headlines. So use bullets and highlighting.

Rewrite this using bullets and highlighting to make the copier easier to read. And, so here is I thought this was interesting. The language model added this bottom line element to it. So you’ve got, you know, something that appeals and then a bottom line that, really would appeal to those methodicals.

And for those of you that are in the ecommerce world, if you want to test product page copy, this is always a problem because usually your top products don’t get enough sales to do an AB test. We have to aggregate multiple products. So going in and and rewriting the product description copy for each of those products means we might, you know, as a human, we sometimes we might nail it, sometimes we won’t. Well, the language models will enforce consistency. So every time you ask it to rewrite something, it will rewrite it in that voice very exactly. So you can say you’re gonna test like a more humanist product descriptions across, ten or twenty products that you need to really get the the the statistical sample size, up enough.

So it’s never been easier to create these messaging experts. I’m gonna take you through very quickly, the process that I think you should use to, create a what we call a messaging, expert. And we do these for our clients.

Building a messaging agent means that we wanna collect our clients’ internal marketing documents. You will collect your internal marketing documents. We’re gonna do a deep research report on your industry so that, you have a a big report that talks about what’s going on in the industry, how big it is, its growth, who the competitors are, what the primary personas are, what their care abouts are, what their objections are. It’ll all come out in this.

And so it’ll be used to train your agent, your custom GPT or your Gemini Gem, to be an expert in your industry. So we use that deep research report and those those documents of yours to train it, and then we can begin asking questions. So those internal documents are your personas and I p ICPs, your brand guidelines, any industry reports you have. Some of, the larger businesses have brand voice documents, how they like to talk to customers.

We want to make sure that your messaging agent is using that voice even though ultimately what we want to do is test the boundaries of that voice and change it as much as possible. And you can write, you can follow brand guidelines and still do a humanist voice and a methodical voice. The language models know how to do this.

And then, of course, if you have any analytics data that will help you identify visitors, we want to include that. So we’ll take those things.

I’m a big fan of, Trust Insights Casino prompt. If you go to this link, ConverseEye dot com slash Casino, you’ll get the PDF, from Trust Insights. You paste this into your paste the whole prompt into your ChatGPT window, and it will walk you through and ask you a series of questions to make sure that you provide the context of what you’re trying to do, that we are you know, we’re trying to create a messaging agent and the audience. The audience will be us marketers for this company.

And what the scope is, you know, we wanna make sure that you include competitors and that you include who the primary buyers and visitors are and what their titles are and the intent, the narrator, the outcome. It’ll take you through and make sure that the prompt you’re using contains all of the elements that we need. The best use of AI is to help you write prompts for AI. So the first step is going into your chat, asking it all the right questions and telling it to write you a prompt and then using that prompt to get what you ultimately want.

This is a fantastic way of doing that with the deep research.

Then you go in and you choose deep research on the on the, in the language. It will take between twenty minutes and forty five minutes. It’ll be a long report. It’ll be a fantastic read. When we read these reports, we very quickly get up the learning curve on our customers industries.

And then we’re gonna open our custom GPT or Gemini Gem or Anthropic has a custom language model, a custom agent you can open.

And we’ll call it, you know, the marketing expert. You’ll attach you’ll attach so you create all of those. Here’s the prompt that we tend to you that we use for this agent. So you are an expert in the, in this case, the commercial trade industry was our was the example for this one working for the company. You’ll insert your your industry there.

A company that provides a certain product or category, this prompt as the others are available through this this QR code here.

And so then we upload all of those things. So the research that was done and the playbook market reports, one of these is the personas documents. And so we’ve got an agent that we can turn to time and time again that is an expert on the industry and also knows how to talk to these different modes of research.

So you have your agent and you can start asking it to rewrite copy. And I think that rewriting is probably the best use of AI. Honestly, I’ve looked at hundreds and hundreds and hundreds of websites.

And if you’re concerned that the AI is going to be too generic, the AI is right better than eighty percent of marketing teams. I hate to say it, but it’s true. And it’s for the reasons I talk about. We come with our own biases and we tend to be careful. We tend to not take risks. What we’re doing here is using AB testing to allow ourselves to take some risks so we can really find what our visitors want.

So while AI, I wouldn’t necessarily have it write something for you actually, I would, to tell you the truth I think it’s best used to rewriting. So you go in and you determine, here are the things I want to say. The language model is simply rewriting them in different voice. So you’re beginning to understand how your visitors want to receive information. And I think that these tests will influence your brand voice. If you find out coming to your website that you have a huge number of people that are spontaneous like you are, then, you know, you you don’t talk, you know, in flighty claims and we are the best and we do this and we do this. You say, we’re going to solve your problems and here’s how you take action.

So to summarize, we need to develop really different copy treatments to understand what motivates our visitors. Right?

As humans, we just can’t do it. We just can’t do it. The best copywriters will struggle to write in these different voices.

AI does not have this problem, and I’ve just shown you a model that allows you to very easily create, different versions of your copy, whether you create the agent or not, to create different versions of your copy that you can test. And I’ll emphasize it again.

Test the output of the language model so that you can really understand, what’s going on with these different visitors. And there’s negotiation. I know there’ll be other people on the team going like, we don’t write like that. We don’t talk like that. I hate that.

Let me test it and get the data, and maybe I’ll prove you right. But if I prove you wrong, then we found something very powerful, most important, like, that creates a ceiling in conversion optimization that will move the business forward because we’ll be talking to our visitors the way they wanna be talked to.

And I am available for, questions here.

I believe we have, ten minutes or so. Let me pop over here and see what sorts of questions. Does nobody have any questions yet?

I think we have a couple of minutes to spare, folks. If you’ve got any questions, maybe even an idea that you have that you would want to build on top of what Brian discussed, feel free to do so.

Yeah.

Thanks so much, Juan.

There’s any question? Oh, there is. Let me quickly take it up to the screen right away.

And I will, I’ll put see if I can put this, prompt back up for those of you that want to reach out to me and get these prompts.

Ben, the question is now on the stage.

What is the best way to gather an understanding of the current persona of your existing customers?

Do what I’m doing. Test them. Try these different modes of research. Try language in these different modes of research and see what’s going on. So, you can do an AB test, or you can take the approach I talked about.

Rewrite a section of the copy in two or three different ways, put all of them on the website, and then you can use tools like, well, VWO so that you can see where the scroll maps are, where the attention is lingering, where people’s mouse is moving, which for a segment of your visitors is a proxy for where they’re reading, and kind of see where the heat is.

If you have enough transactions or generating enough leads to do an AB test, I think that will be obviously the result. You could test one of these at a time.

But in the pantheon of hypotheses that you have so if you go through and have these heuristic hypotheses, and you’re saying, we want more relevant images. We want to put captions and overlays on the images so people understand why we chose those. We want to get rid of stock photography. We want to increase the number of images that are on our e commerce product pages. Those are a few ideas for images.

Those are some of the early strong ones that you’ll get from heuristics. Getting the right messaging is going to blow those out of the water. If you find the right voice to communicate persuasively, that is, like I said, it creates a ceiling. If you don’t have the right words, you can change the calls to action and rearrange the content and all of that stuff and not really move things, not really move things, along those lines until you find the right words that tells the visitor in the right place, gives them a reason to keep reading, and then lays your value proposition out on the page.

You’ve got to get that right. Otherwise, you’re going struggle with everything else. So I I would I would focus and prioritize these sorts of changes. It was always hard. We’ve not been able to do this easily in the past. Now we have these tools.

Awesome. I hope that answers your question, Joe. Let us know if there’s a follow-up to it.

Coming on to the next question from Caroline. It’s on your stage right now.

In a hero section of a landing page, would it be best to test each writing style as a stand alone, or would it be better to have there’s something here for everyone style approach using a bit of all four. I think so. If we’re just talking about the hero area, I think you wanna test them one at a time. So you really will want a, you know, if your if your if your current style is is methodical, and you wanna test more competitive language, which incidentally, I think is usually a good bet, or spontaneous language. It’s at the top of the page.

We you know, if we target those quick decision makers, we’ll also get the attention of the methodicals and the humanists, but we gotta make sure that that content is below as they scroll down. I would test them individually one at a time like that in the hero area. Down in the body of the page, I would mix it up more. The danger you have is even if, you know, as I said, if you if you try to if you try to write a hero area, for even two different modes of persuasion, it can get a little bit muddied. And, you know, the the amount of lift so if you test what you have currently against, humanist or what you have and then test against competitive, the amount of the lift will give you an idea of the size of the segment of visitors that are coming from those things. And so that will inform what you do going down the page.

There’s another question here.

What’s been your experience generating AI copy using popular copywriting formulas like PASS? So personally, I love PASS. And in fact, it’s one of the first hypotheses that we put out there is that we want to aggravate the problem a little bit. So a lot of times, when our writing, we’re very aspirational. And we can do this for you. And we can do that for you. And we’re the best this.

Look how beautiful these products are.

Whereas, one of the keys is to aggregate the problem. So I don’t remember exactly what PAS stands for. Phenyl, maybe you can tell me, but I think it’s present the solution or aggregate the problem and provide a solution.

And I don’t think we do enough of that. So that is a fantastic idea. And you can go to your language model or go to your chattypd and say, rewrite this and aggregate the problem. And I found the language models do a fantastic job of this. Emphasize the negatives if they don’t take action on this page or if they don’t buy my product or don’t buy my solution.

Pain, agitate, solution is what PAS is. So I was close.

But agitate, I think, is the thing that we’re often afraid to do and something that we should do more of.

That’s a great question.

I hope that answers your question, Ben, there are a couple of talking points in the chat section too. Do you mind taking them through?

Yes. Let’s see here. Oh, there are a number here. Do you have any quick tips on how to translate some of these formulas to comms like email, SMS, push?

And if I understand the question, my my thought is, you know, you can ask the link. Once you have the this agent that understands your visitors and things like that, you can also train it on, what you’re doing in email and SMS. So you can go into analytics and say, like, here is, or go into your marketing automation system, say, here’s an email I sent. Here are the results, you know, the open rates and the click through rates and the conversion rates on it. And the language model will start to learn what works and what’s different, what doesn’t.

So it can write them from scratch for you. That’s a whole other presentation. But I would, honestly, I mean, I haven’t tried this, but I’m going to try this. Take an email that I send, send it.

I’m then gonna ask my language model to rewrite it in a different voice, And then I’m gonna send it again. And what I’ll probably do is send it to the nonopeners, so anybody who didn’t open the first email. So it should still have a fairly large segment to test and see if I get better or worse open rates and click through rates and conversions on the website. That’s an interesting idea.

But the answer is yes.

You don’t have to write copy anymore. All you have to do is take a first pass. Say, I want this copy to include these primary points, and then let the language model change the voice for you. It’s, absolutely. I think this I’m focused on the website here, but absolutely, email, SMS, all of those.

Question. Thoughts on synthetic audiences? Ah, yes.

So this is one of the things that we think is an existential threat to our business. Can you develop synthetic audiences? In other words, you can tell, a language model like, split yourself into one hundred people, look at this variation. Half of them will look at this variation, half of them will look at this variation, and predict what an AB test would deliver. And I believe the language models will become very good at predicting what a fully run AB testing tool or model would have done.

So I haven’t seen any evidence that they’re here. I’ve read a couple of papers on synthetic audiences. They’re all still kind of coming, kind of coming. I suspect they will be in the expensive enterprise range.

And what we’re going to need to do is run the language models or rather the synthetic audiences against some existing AB tests that we have to see how accurate they are. So it’s going to take a while for us to get up enough of those analyses to really begin to believe that these fake audiences can predict the outcome of an AB test. But this is, AB testing is, like I said, the best, content we can develop. It gamifies marketing. It’s fun to do AB testing and place your bets. And, when it goes away, it’s gonna be a very, very different, it’s gonna be a different agency that I’m running.

I love that question. Are there, are there is there a low fidelity way to start today while we’re working towards a thirty to forty page report?

Well, you don’t have to work toward towards a thirty to forty page report. You just have to ask it. Literally, do do the casino prompt, place it in, select deep research, and it will do the research for you.

If you but, I mean, honestly, many times, I’ll open up a a chat.

I’ll place that in that that says, here’s how I talk about an s j. Here’s how I talk about an n t. And I’ll say rewrite this for Myers Briggs n t. Rewrite this for Myers Briggs s j. Don’t need that full report. The the the messaging agent is something that provides you and your business a repeatable way, and a consistent So So It’s it doesn’t take a lot, but, yes, you can open a chat and just say, rewrite this for an NT.

The Myers Briggs type NT. Give a little more detail. Would it be best to submit your LLM existing communications from customers, reply emails, or SMS responses to understand the persona and messaging.

So I I upload I use my emails, my past emails so that the the language models, will maintain my voice. So there’s a certain way that I like to write.

I think it’s distinct from, a lot of newsletters that people get.

It’s generally shorter. It’s generally punchier. It’s, those sorts of things. So I I use my emails to kind of train the voice of the language models. But, in terms of I have not gone all all the way through, the process of taking all of my emails and all of the results of those emails, the open click and conversion stuff, and trained a language model to write things from scratch based on that. So I will Joe, I will defer to your own research on that.

Folks, it was a pleasure. And thank you so much, Brian, for this amazing workshop and a thorough walk through of the nuances of how persuasion and messaging comes together in business, in organizations, in teams, and, obviously, validate and make it more personal. And thank you so much, Brian, for this amazing session, and we’ll see you again maybe, you know, another session of Convex.

I certainly hope so. I I love doing this. I appreciate you inviting me.