Marketing Research Reinvented Published: April 16, 2025 Watch the webinar on how to use AI for market research, using conversations and 1st party data to find ways to better connect your marketing with your audience. Using AI to Humanize Your Marketing By now, you’ve probably experimented with a handful of different AI tools to create content. And like a lot of people, you’ve encountered the same issue: without context or direction, a lot of the output is bland, repetitive, and clearly written by AI. While content creation may be the more publicized use case for AI tools, it’s not the best use of LLMs. The newer reasoning models have the ability to both scale up and improve your marketing research, allowing you to understand any business or service’s market fit like never before. What you’ll learn How to use AI and tools for general research – Tools like Perplexity, ChatGPT and Gemini Deep Research How to use AI for human and semantic research – Using Reddit conversations, product reviews, or even internal communications like meeting transcripts to gain insights How to use researched information for your branding & marketing – Using AI to create Personas/Buyer Journey Joelle Irvine Growth & AI Advisor, JI Growth Marketing With over 15 years of experience, Joelle partners with eCommerce and SaaS companies to drive profitable, measurable revenue growth through customer-focused go-to-market strategies and custom AI solutions. Previously, Joelle spearheaded growth and marketing at Moz and Bookmark, a WPP global agency. Devoted to community advancement, she avidly shares insights as an international speaker, dedicated mentor, and award judge. Beyond digital marketing, she enjoys playing the drums, inspired design, and obscure pop culture references Webinar Transcript Mohamed Hamad: Hello, everybody. Welcome to the Third Wunder webinar, our Third Wednesday Webinar every third Wednesday. Today we are gonna be talking about marketing research reinvented, how to use AI to humanize your marketing. I’ve got the lovely Joel Irvine here to help us walk through this topic. She’s a growth and AI advisor and has been playing around with AI for marketing specifically for the last—I’ll let her introduce herself in that topic specifically. But yes, Joelle, a little bit of an intro. Joelle Irvine: Yeah, hi everybody. So, I’m Joel. I’m a growth in AI advisor. I’ve been tinkering with AI and learning all the things for the past, I don’t know, six or so months. And so yeah, I work primarily with SaaS and ecommerce companies, and some agencies to help them with their growth strategies, product marketing, analytics, and I’ve now embedded some AI services into the mix, so doing a little bit of everything, but it’s always in the focus of growth marketing. Mohamed Hamad: Yeah. So we wanted to talk about AI because, you know, it’s top of mind for everybody. Everybody’s got concepts that it’s gonna take their jobs, or it’s going to revolutionize something, or it’s going to, you know, destroy the planet, whatever it is, the conversation, it is top of mind. And while the big promise of AI is that it’s going to write things for you. The way it’s been marketed most recently is it’s going to write your emails, it’s gonna write your content, it’s gonna write your social. But we wanted to take a step back a little bit and kind of rethink how to use AI and some of the features around the latest updates and how to incorporate those features in research, which is something that’s very underserved but can really help you speed up how you do research with AI. The thing behind that is that AI’s can ingest large amounts of data very quickly and find patterns very quickly. So this is where the time saving comes in. When looking through an enormous amount of blog posts, for instance, or big data sheets, or lots of call transcripts or whatever it is, it can pull it all together in seconds, highlight the things, and find the things that you really might miss out on because, you know, after a while of reading like 10,000 page reports or, you know, 20 blog posts about the same topic, you just kind of glaze over things. So, one of the things that we are discussing here is how to use AI for research. I’ll hand it over to you, Joelle, on what are you, how are you using AI for research and what are some of the tips that you would like to give us around what your workflow is like? Joelle Irvine: Yeah, for sure. So I’m gonna take a step back for a second and just tell you how this all started. So, working in many companies, you know, leading the marketing team, essentially, we’ve always hit these, these areas where each team has their own data, you know, like the marketing team has data, the product team has data, customer success, sales, everybody has their own data, but a lot of the times. That there’s a lot of gaps between shared information and teams tend to work in silos. And so I started off my exploration of AI to find a solution for this. So I started there, but that’s a really big problem to solve. So I’m really focusing on the area, which I know most right now, and that’s on the marketing side. And looking at marketing research that could be used across different parts of the organization. So in terms of my workflow, to answer your question, I use AI to help me, I’m gonna say go from A to B. So, I have my initial information like I’ll know what the product is or what the problem is I’m trying to solve, or who the user is or things like this, and then I use that data to explore other topics. So, in terms of use cases, you could think of like when a lead comes in and you want to research them quickly, you know, for prospecting or for pitching, you can learn about them very quickly. In terms of launching a new product, you want to see what’s out there in the industry in terms of competitors, things like that. So if you use an example like that, and you’re trying to understand the landscape, you would essentially need the AI to get some context. Think about when you’re briefing somebody. You need to give them enough information to do the actual task, right? I’m jumping around here a little bit, but I’m trying to get—I’m gonna get to the point. Let’s say I’m briefing somebody to say I need to do research on this. What do they need to know? Well, the AI needs to know who that person is, so you give them, you send them a role, you say like, OK, you are the best person to do marketing research with a lens in this industry. Then you give them the inputs. You say, “this is my product,” you explain your product, you say, “this is who my user is.” And I’m not saying you need to give them documents and documents and stuff. Just like highlights of who this person is, what the product is, what the problem is you’re trying to solve. And then potentially, you then go into like the instructions of what you want to give them, or what you want to get, and then some examples of how you want to receive the information. So I usually use a framework like that in my prompts. And then I go to whichever tool I want to use and I go for that. So, I tend to use it, you asked me where I use it. I use it a little bit everywhere, but that’s essentially my process. Mohamed Hamad: OK. I just wanna find out from the audience what tools they use. So I’m gonna throw out a poll and if y’all can answer this one, we just want to understand which of the major AI tools or the large language models that you use. So I’m just gonna throw it out there and if y’all can just give us an idea of what is your favorite tool from all of these. OK. We’ve got 100% chatGPT with some people using perplexity. And this could be, you know, for any task, not just research. OK, we got one person using Grok. Interesting. OK, I’ve got a few people saying they can’t select more than one, unfortunately, because this is a poll, there’s only one single answer per, but choose your favorite one, or the one you use the most. I’m curious to know who the Grok lover is. Yeah.Patty says it depends on what I’m using it for, and yes, you know, different tools are better for different tasks, really. Joelle Irvine: Compare the answers, that’s actually a really great thing. I find that using tools and comparing answers for different tasks, for different use cases, things like that is always a really good practice, especially when you’re doing deep research or you’re trying to collect different use cases. It becomes really valuable. You’ll see that some give better sources than others, some are more local, some are more broad or international. So that’s a really good point right there. Mohamed Hamad: OK. I think we got most people in there. I’m gonna stop the poll here because I would like to have a follow-up question on that one. And it is, what are people using it for the most? Again, this is a poll, so it’s gonna be a single answer, but what are you using these tools for? Is it just chatting? Are you using it for content creation or research? Are you playing around with vibe coding? Are you creating agents? And if it’s other [uses], throw that in the chat. Let us know in the chat if you’re using it for other use cases. And I think this is also only one, so what’s your favorite use case? Joelle Irvine: Exactly. Yeah, I like Claude too. I’m a fan of Claude. Yeah. OK, partner, interesting. Brainstorming, naming, blog titles, dummy data and translation. Excellent. All right. There’s a lot of translation since we’re in Montreal. Yeah, yeah. Yeah, I like the partner idea or like a co-pilot, you know. Mohamed Hamad: So creating agents is what I was thinking of when I was thinking of a partner or advisor or brainstorming, yeah. That’s great. Joelle Irvine: Yeah. Mohamed Hamad: OK, we got a bunch of votes there. I’m gonna close the poll there and, yeah, this is great. So, speaking of how to do some research on that one, because a lot of people are actually using it for research. And you mentioned creating the context, giving it information about who you are, what you do, what the product is and then creating the prompts. When you create your prompts, do you use the actual AI to help you create the prompt? Because I certainly do that sometimes where I would start off with something that I want and I would give it a prompt. I would ask it to help me create a prompt to do something specific, and then it’ll flesh it out for me. Joelle Irvine: That’s really smart. So I have certain prompts that I like to use, like I have. I’ve taken several courses and I’ve, I’m gonna say, mixed and matched my favorite prompts. And then I go from there. When there’s something that I’ve never done before, I will sometimes use the AI to help me navigate that prompt. But I also find there’s a refinement that happens as you do it more and more. So you’ll be like, this isn’t giving me what I want, I need to give it more context—or this isn’t giving me what I want, I need to give it more direction in terms of what it’s good at or what it’s not good at. And also, I’d say the last thing is, in terms of the structure of a prompt, I generally go with role inputs, instructions, format [for] what I want to receive in the end, and then clear examples, and that way I’m setting my expectations. As long as I stick to that, I generally get good outputs. Mohamed Hamad: So, giving it the correct prompt, but then how do you give it context from external sources. So for instance, if you have large amounts of data that you need to compare, for instance, or do some research that is from online-based sources, how would you go about giving the LLM the information that it needs to let’s say do a comparison or to find things online for you? Joelle Irvine: Yeah, so, if the inputs, whether it be online or actual documents, if it’s too much content, I usually do it in batches. So if I want to compare two documents, I would just do that. And potentially if it’s just two documents with no external input from the internet or whatever, I would use a notebook LM because then you have a closed ecosystem and you can just do it there. Then I would pull those results and then use a different tool to then compare that with external sources. I find the more you break it down into manageable chunks, the better the output is. Mohamed Hamad: Yeah. Yes. Well, I noticed that nobody chose Notebook LM and I’m a big fan of Notebook LM and specifically for adding a lot of information about a specific topic, and then diving deep into it. What’s your experience with Notebook LM? Let’s step back and say what Notebook LM is. It’s a product from Google, and it’s not a large language model in itself, but it’s an AI-based note tool where you can add documents, links from external websites, YouTube videos. And what it does is that it digests everything and synthesizes it into a notebook, but then you can query [it], and everything that you ask is specific to the content that you give it. And my favorite thing from it is generating audio overviews because it creates a podcast. I’m more of a listener than a reader, so I’ll generate an audio overview and listen to two AI’s talk about the subject while I’m, you know, running an errand or doing something in the background. But what’s your experience with Notebook LM in this specific case? Joelle Irvine: Yeah, so I also love the podcast function. I recently took an intensive course with Google. It was like a Gen AI intensive course, and they had these huge white papers, and [they said] “or you can listen to the AI generated podcast.” And I’m like, yes, thank you, Jack. It becomes way more interesting, especially when you’re trying to learn something or if you’re trying to ingest a lot of data quickly. And then if you need to, you can go look at the sources, look at the tables or the charts or whatever. Whatever you need to look at more deeply, you can. But I also have used it for, let’s say remixing content. OK. I know this is—we’re on the research topic, but I think it’s also interesting if you don’t know, like we know that the AI tends to hallucinate on occasion, or they tend to go off script if you don’t brief them correctly. Within Notebook LM, what I like about it is that you’re focusing just on the input that you give it. It’s not going outside of that. So if you’re doing research, like you said, comparing things or compiling data from multiple sources that you have that are tangible, that are not on the internet, that’s the best place to do that. It’s also great for taking that type of information and creating, let’s say, like an outline for a deck or creating your own podcast, it’ll give you the script, so it’s great at, again, remixing. I love the remixing terminology. Mohamed Hamad: OK, I haven’t used it for remixing, but that’s a good one. Right. So, you’ve got this research; you’ve gone off and compiled things; you’ve asked the LLM to bring you some information, but what if you have information, first party information? A lot of organizations have internal documents, they have chat transcripts, or meeting minutes, or sales calls. A lot of this information is a dearth of data that usually, once it’s created, kind of disappears into Google Drive or SharePoint or whatever it is, and nobody looks back at it. I wanna talk about how we can use first party data. Whether it’s internal or it’s on a third-party source, but it is human conversations that are had on review sites or conversations had on Reddit threads where people are actually talking about a specific product or, you know, the product you’re researching. How do you go about getting details or getting insights from this kind of information? Joelle Irvine: Yeah, 100%. So it’s funny that you mix that, but I agree. It’s essentially—it’s customer—you’re talking about customer data. And so, this is part of my big challenge that I’m trying to solve for. What I find really interesting is that, you know, for years, we’ve been trying to collect data, saying like, “OK, let’s figure out what the keywords people are searching for,” things like that. But now, it’s more about what questions are people asking? What are their biggest pain points? What are their objections? What are the questions that they’re asking? And what are the feel-good moments about the brand? Like, how can we get all this information? We have all this information. If we’re creating transcripts of our calls, we obviously have all the email backups and things like that. That’s a gold mine of insights. So collecting all that data into documents is amazing. Collecting or scraping data from Reddit or from reviews like on TrustPilot or G2 or whatever, wherever your customers are. Collecting all that data into documents is like the first step. So finding ways to scrape that, whether it be through internal data collection or—for example, like with Reddit, you can actually scrape Reddit threads through their API so you can actually collect that data. Once you have all that data, what I’d like to do is essentially—I don’t remix this data. I keep it separate. Because I’d like to know what people are thinking and asking at different stages of the journey. So I keep everything sort of separated that way. Let’s say you have sales calls, you have the customer service calls, you have the reviews on the sites, they’re at different stages. And then I use—I have certain prompts that I use to understand customer sentiments at different stages of those journeys. So I have a prompt that I’d like to use, and then I modify it depending on what my main focus is. And then I apply the same prompt on all these data sets separately. OK. And then I look at the data. And I compare and I’m like, “Oh, this is interesting.” All of this data is aligned. Great. So brand wise, we’re doing a good job, things like this, but we can address their pain points in this way. They all have the same pain points. Or you look at it again and you’re like, These are different at different stages. Why is that? Why is that happening? How can we address that? So these are the types of insights that can fuel your strategy. Mohamed Hamad: OK, so talking about understanding the buyer journey across every stage, and then taking that information and analyzing it at each step of the way. I like using deep research, whether it’s on ChatGPT’s deep research or Gemini. I find that with Gemini, it can search the web really well, and I can specifically ask it to scrape Reddit or LinkedIn for topics around the specific keyword. So I find that one is an easy way to get a lot of information consolidated in one place, but it does digest that—it doesn’t take the raw information as you would if you were scraping it, but it does digest it for you and then give you a summary of it. And I think that’s a great way to get third party information. But how do you take all of that and turn it into marketing insights? How do you bring it all together and say, OK, we’re now going to digest it and understand our brand a little bit more, adjust their messaging, get closer to the audience and maybe even for sales people, how do you create objective decks out of that? Joelle Irvine: Yeah, 100%. So, essentially, once we have that data, we prioritize what we need to prioritize. If we are noticing that there’s a big challenge, and I’m getting super—I’m not using an example here, but like let’s say the challenge is on the brand side, and there’s a lot of people asking questions. You know, is this legit or is more on the credibility side? Then you know that, OK, we have a big problem here. They don’t think that our product or whatever it is, is credible, then we need to just take a step back and say how do we adjust our messaging, how do we adjust the messaging out there in terms of search, in terms of on the website where people find us, but then also how do we account for that throughout the rest of the the the journey. On the sales side, how do we reinforce that messaging when the sales team talks to them? How do we enforce that messaging when the customer service team talks to them? It’s making sure that once you identify that problem. That there’s a continuous flow throughout their experience. Does that make sense? Mohamed Hamad: OK, so you’re taking—you’re understanding the pattern across the buyer journey, and then you’re creating internal or external content to make sure that any objection, any concern or any credibility concern is accounted for, so that the messaging is consistent across the entire team or the entire buyer journey. Joelle Irvine: Yeah, exactly, but if the challenge is—let’s say something that’s specifically within your product. And you have a different challenge. And so then at that point, it’s like, is this a marketing challenge or is this a product challenge? And if it’s a product challenge, then you bring it to your team and you talk to [them] about how we can address this. But then you also have to address it within your messaging. Because people are gonna be asking the questions. People are going to be talking outside of—beyond Reddit and beyond the customer reviews, they’re gonna be talking to each other. So it’s important that you make it as easy as possible for people to understand what you’re doing and why it’s good for them and how it’s gonna solve their problems, and if there’s something that’s blocking them, how it’s being addressed. Mohamed Hamad: OK. I think that’s great, and I think you kind of touched on something in that it could also help with product development and that if there is a consistent throughput of conversations being had about how a product functions or how it’s perceived to function, then that could also be taken to the product team and say, you know, how do we account for this or how do we make the product better, or how do we streamline the feature better, I think that’s an interesting thing specifically for product led teams. Joelle Irvine: 100%. And it would be the same for a service too. Like if there’s a challenge with the service, you pass it on to the team that’s running that service. Mohamed Hamad: Absolutely. OK. We’ve got a few questions from some of us in the chat over here and Patty asks here: “I know you have to check the work, but how do you know when the work is done and doesn’t need more fine tuning, etc. Like, when does that process end?” That’s a great question. Joelle Irvine: So, I would say, the way that I use AI tools is that I need to give it the meat early on. Give it enough information like the context and all that stuff, so that it can give me a good output and then at the end, I have to be knowledgeable enough about what I’m doing and what I’m talking about to be able to fine tune it. So, you can’t expect it to do everything for you. You have to have some knowledge. And if you don’t have that knowledge, then you have to figure out the way to help you do that fine tuning, whether it’s a subject matter expert or somebody else on the team that can review it. In terms of understanding when it doesn’t need any more fine tuning, I think that’s a learning process. I would say as you continue doing this, the fine tuning at the end should reduce because every time you do it, you’re just gonna get that much better. Mohamed Hamad: Mhm. Yeah, I feel like there’s a point where you find out also that, you know, we’ve got enough information or you don’t want to push it anymore. But that could be a subjective thing where you’re like, OK, I’ve got all the details I need. Joelle Irvine: Yeah, 100%. I mean, there’s many times where I’ve put something in and I’m like, this is not what I wanted, but then I look back at what I put in at the beginning and I’m like, maybe it’s because I could have done this better to start with. You know, given more information, more context. Mohamed Hamad: Um, Geraldine asks: “Can Notebook LM also search the web? You mentioned you use it to look at offline material.” Joelle Irvine: So, this one here is interesting because while Notebook LM is used to consolidate different sources of information, there has been a recent update where you can actually give it a search term, and it’ll find offline online sources, which you can choose from to add into the notebook, right? So if you want to start off with the topic, you do a little bit of a search. It’ll give you a list of things it found online because it’s a Google product. It has the Google index, and they built in a search engine to find these topics. Once you add them to the notebook, anything after that within its chat or whatever it outputs is specific to the content that was added to the notebook. That’s a great question, and Notebook LM [is] one of my favorites. It’s great. Mohamed Hamad: Um, Another question from Patty is: “How would you validate your end product?” Uh, Patty, can you give me a little bit more context? [The] context there is from an analysis perspective when thinking about market research, how would you validate your end product? Joelle Irvine: Uh, like end product of the market research or like, sorry, I just wanna make sure I understand. I don’t know if she can provide more information. If I can take a guess—if we’re talking about the end product of the research itself, it would be kind of like my own knowledge. I would be able to look at it and say, yes, this makes sense, or I’m still missing more information because I have that experience. If you do not have that experience, let’s say you’re trying to do a benchmarking exercise. Do a first search like what we were talking about earlier, when you’re like, what do I need to prompt or what does my final product need to look like? So if I’m doing [a] customer benchmarking exercise, you can do a search first to be like, what goes into a benchmarking exercise? What are the things that I need to include in there? What are the insights I need to gain from it? What’s the best structure, or like, is there a template that I can use? And then use that as your guideline so that you know what you need to have at the end. I hope that answers that question. Mohamed Hamad: I think it does. Yeah. We’ve got a follow-up question from Geraldinean again. Um, I think it’s a follow-up from The conversation about Notebook LM. “Have you used it for market research, like trying to find your ideal clients?” Joelle Irvine: Yeah, for sure. So, I think once you know what the use cases are at a general level, I’ve used it to flesh out personas, I’ve used it to flesh out user stories. I don’t know if you’ve ever created user stories, but it’s very much—let’s say the growth marketing director uses this to solve this, so that I feel like it really helps to figure out what those things are, and then you can use that information to better target for your audience specifically. I haven’t used it for um doing cold outreach or anything like that. That’s not something that I do right now. I’m very much more of an inbound kind of gal. Mohamed Hamad: Mhm. Following up on that one, I use deep research, whether it’s ChatGPT’s deep research or Gemini’s Deep research or even Perplexity has a great deep research tool. And one of the things that you can do for, let’s say, finding your ideal client, or doing any search, analysis is one finding conversations on the web about a specific product or service and then using it to analyze any pain points that they have, and that way, what ends up happening with deep research, it searches the web for maybe 20,30, 40, sometimes up to 170 different websites and pages. And if you direct it in a specific audience or industry or even location, it’ll find that information, and will digest it into a report. And these reports can be really very robust. ChatGPT goes up to 30 pages sometimes, and Gemini goes—it’s a lot less, but 15-20 pages. But you can get some really deep insights on what people are saying, their tone about it, and just depending on how you frame the research prompts, will give you quite a deep understanding of what your ideal client is actually saying or looking for. And using that, if you’re doing sales outreach, creating an objections plan, or finding out objections for what their concerns are, or even creating messaging about their pain points and concerns in that early stage of awareness of their buyer journey. So that’s a great tool there. Joelle Irvine: I’ve also recently— I’m working on an agentic workflow so that once leads come in, I’m able to get to do a quick research automatically. So that’s something I’m building right now, which is cool. It helps with understanding who your audience is, what their pain points are, what the company is, and if it makes sense and it’s a good fit. I’ve also done exercises where I’ll put in like three different users, it’ll create a matrix of personas for me, and it’ll tell me is this a good fit? Is there a willingness to pay for it like that type of analysis, which becomes really interesting, as well. Mohamed Hamad: Absolutely. So, we’re a little bit over time, and this was a great chat, lots of interesting insights, and also great questions from the audience. If you want to connect to Joelle, you can connect with her on LinkedIn, and if you want to discuss AI with her or any of her work flows or how she does market research or any of that, you can give her a shout on LinkedIn. If you’d like to explore AI with us at Third Wunder, we do a lot of custom agents and we also use it internally for a lot of the work we do, mostly on research and ideation. In saying that, this has been a great episode. Thank you, Joelle, for your time. Thank you everybody for attending and your great questions. Any parting words there, Joelle, before we wrap it up for the day. Joelle Irvine: Thanks so much for having me on today. It was really fun. And yeah, if anyone has anything, if anyone wants to chat, feel free to reach out. I love talking about all this stuff, thanks so much. Mohamed Hamad: Awesome. Thank you, everybody, and I hope to see you in the next one. Upcoming Webinars The Refresh: Realign your brand