How to Turn Reddit Threads into High-Converting Marketing Assets (Vibe Marketing 101)

0:01All right, we are live back here with Veru who is going to guide us through another lesson for vibe marketing 101 and this time we're going to talk a little bit about Reddit. Is that right Veru? Yes, that's right. Cool. So yeah, looking forward to to seeing what you have for us today. And yeah, give us kind of an intro in terms of you know, what we're going to be talking about and what you think people will will get out

0:34of this lesson. Absolutely. So I think Reddit has always been you know, you know, you always say that it's a gold mine where you can get a lot of ideas, a lot of business ideas for that matter or even content ideas and many the AI overview and things that we see is like extremely reliant on Reddit and stuff like that because of the UGC conversations, right? So when we do research, I think Reddit plays a very vital role for content research per se. So this is a beginner and semi-beginner friendly sort of a thing because we'll have code as well part of this workflow. So what we

1:15will be exploring in this workflow is that when we get into Reddit, this time we're not taking comments into consideration but just the post, the overall post and stuff like that because if we take comment and things like that, the entire workflow can get a little bit bit messy. So I'm just going to take the post around a specific you know, our universal specific theme is marketing automation, right? So we're going to take something around marketing automation and look into the posts and identify what has been talked across these posts, what's the summary, what are they addressing, what's the pain point that's being addressed over there and what content theme we could

1:52actually get from there, derive from these posts. And then based on those content themes, what sort of hooks and angles can we look into so that the next step is like going to the content ideas based on the hooks and angles and then build content on top of it and then the content can be across multiple channels. But right now we're just going to look into the

2:14research so it's not one-dimensional. Cool. Yeah, I think Reddit is is such an interesting platform. You know, Greg talks a lot about Reddit as well to find you know, startup ideas and and things like that. Yeah, if you if you look broadly at at Reddit and it's kind of place right now in the marketing ecosystem, I think there's a couple like patterns or nuggets that that I take away from it and you know, like you mentioned Google search results prioritize Reddit. If you if you take Reddit and you go and plug that into you know, SEMrush or Ahrefs or something like that, like their growth has been

2:55astronomical as search engines prioritize user-generated content. I don't know about you but like when I search for something on on Google, I'm like putting Reddit in my query like every single time, you know, I have a golden doodle so as as my dog so it's like uh best activities for golden doodles Reddit, you know, because I we ultimately I think we want to hear from like real people and you know, when you if I just search for for uh you know, best things to do with my golden doodle, then I'm going to get a bunch of blogs that have like you know, stuff that is irrelevant, a lot of

3:38filler and it's hard to sort of get to the bottom of it and I I think like searching for Reddit is almost like you're asking a friend for a recommendation or something like that. And then you know, these communities or these groups on on Reddit are are just so close to the topic so you inherently kind of trust what they're talking about and and things. The other interesting piece about Reddit is if you get into Reddit and you create a post or something within a community, if you're trying to like you know, if it sounds like marketing or something like that, you're going to get roasted, you know, so it forces you to

4:24have like this authentic value first kind of vibe that that you bring to the table if you're going to get any traction there. I agree and I think in this workflow right now for the sake of getting the research ideas, how this entire flow looks like we've put it together. But imagine if we could get the comment section as well within the Reddit and then we do this entire research, it's going to be absolutely like you know, filled with a lot of perspectives and you have option to like you know, fish through those perspective and say these are standing top when it comes to the pain point. It's very I think like you

5:03said, founders can actually like look into it and easily get the pain points of their audience and stuff like that from Reddit without having to spend too much in terms of the market research and time there. Yeah, it's also a great place to find like pre-validated ideas that resonate or ideas for content, whatever, that resonate with an audience and and you can be pretty sure that if it if it works or resonates on Reddit it's probably going to work elsewhere. Like you know, I found really interesting things within let's say the innate end community in Reddit and then shared it on X and things and it's gone completely

5:42viral and I know a lot of other creators leverage that approach as well. So yeah, excited to to see what you have for us today. Yeah, I'm just going to share and I'm also excited to like share the workflow. I hope you're able to see my screen. Yep. Awesome. So just to give a little bit of brief over here, we have couple of nodes. Node one is the native Reddit node and if you see the common nodes over here is the code node. Code node is going to help us structure the data that we are getting out of each of the nodes

6:20before. And this node is a LLM chain node where we are using Grok over here to do one-step analysis and we're also using open AI so you know, generally you know, James you've always been encouraging us to use multiple models of LLM so that we are able to like extract the best out of each of the steps. So we

6:45are using Grok over here for a reason. We are using open AI again here and then the third one when we wanted to look into hooks and angles like we have all the artifacts with us and we want to get a very strong hook and angle and stuff like that, we've used Claude over here. And we have a sheet node and set node ideally to store the data and to direct the data in terms of if you don't want to code, you want to basically like structure simply, you can use the set node as well. So this is the

7:15overall picture of the workflow. Okay and when you run the entire workflow, maybe give us a quick snapshot of the final output and then we can maybe work backwards a little. Yeah, so this is the final output. If you see this is the here you will see Reddit post. Sorry for scrolling through quickly. So here is the Reddit post. So we will get the

7:40Reddit post, the title, the subreddit. And we will get the summary of these Reddit posts like what is being addressed in this post and what is the theme behind this post. And what are the points that are discussed in each of this post. And from there we are trying to derive content theme. Like if you want to do okay, these are the posts, how do we basically like

8:03build content themes around it. Now based on those themes, we are coming up with some hooks and angles like you know, we can use so that for the content ideas and content creation for the later purposes. So these are something that we will be doing in the research. Perfect. I'm going to go to the workflow. So I've always a big fan of manual trigger and it's totally like up to our choice to basically go for a chat trigger and then like you know, the trigger happens when you chat but you know, this is like this gives me more control. So I'm starting it with a manual trigger where you click so that

8:43the entire workflow runs. So I'm going to go to Reddit. So this is the Reddit native node, right? So I've kept the subreddit to be marketing automation and I have a keyword saying that I want posts around marketing workflow. Now it will give me it will take me some so I've mentioned that I just need 10. You can go up to 100, 150, how much ever you want and I've gone for hot instead of relevance so that I get posts that have multiple comments and stuff like that that are basically being

9:15engaged a lot. Makes sense. So I have these let me just think is an issue with the yeah, zooming. So these are the posts that I got and if you see there are like a lot of irrelevant, too much data that's there and I'm going to like remove this and get the structured output out. Just to double click in into that real quick. So um what what benefit do we get when we just extract what we need from a

9:50data source and feed it into the LLM. Just I I just want to highlight like why that's such an important step. I think when we only take a specific thing and then run it. Like I've done something a big blunder when I tried to do this first workflow before was that I just took the content that I got from Reddit and then I ran through LLM. Two things that I faced, one was it does take a lot of API calls, right? Like because it's too heavy for any model to run. Like if I'm going to give 100, it's going to be like very difficult for it

10:25to run. And secondly, when it takes so much data, the output that you get is not going to be perfect because it usually only takes in the first few things for it to basically consider for the entire analogy that it's going to do. So we have to do something. Like if we want to take multiple posts, we can always will have to send it by batches. That's the rule. Like the hard and fast rule that we have to follow in any automation that we are going because we need to make sure we cover all bases. And at the same time it's not like loading the API

11:04or LLM chain for a long time. So yeah. Yeah. Makes sense. And I I think we see this uh as a common theme throughout multiple workflows that we're building ranging from, you know, a beginner workflow like this to the complicated setups like, you know, using Claude MCP or whatever. Um you know, the the LLM uh generates a lot of power when you arm it with with data, but arming it with too much data uh uses up a lot of context, a lot of tokens, and it increases the surface area for um hallucination and and errors. So um our job is to kind of create these tools or these filtering mechanisms um

11:51that help us get the right data um to generate the right output from from the model. Exactly. And I think that can also be automated, right? Like for example, I've used a code over here saying that here only pick up certain things. And in the same code I can still further add. Let's say I have like 100 data that I've taken, right? Now it is not necessary that whether it is Reddit or any search mechanism that we go for, we would

12:17probably get a lot of junk data. So we can also like add context saying that remove the junk data either in the code or if you're like shying from using a code or something, I would really suggest split in batches, add another AI node, and say that here remove the data that is like junk, that is not relevant to this particular theme or something. So that it gives out data that we would want to like take into consideration for research and analysis. Yeah. And also just to highlight this again, you're not

12:52a software engineer or developer, right? So when you need code like this, walk me through your process again for how you're generating that code. Yep. So I think it's a very simple thing that I do. First time what I do is usually what I used to do is that I just mention in ChatGPT that this is where I want I'm looking for code and things like that. Often times it never works out. But the best way is to basically take a screenshot of what we are looking at. If there is an error along with the error, that I I just take a screenshot and then say I'm trying to like

13:28structure this output and I want the output to be termed like this. So if you see the return has to be title, self text, and subreddit because that's the output that I want. So give me a code that I can actually add it to the code node. So it basically like gives me the code and I copy paste it. Sometimes the code can be when you paste it one time, it'll actually work and the second time it doesn't work. Most of the cases it happens. So basically then I go back to ChatGPT and then say that it has to be dynamic. It has to be like consider all

14:01sorts of AI output from the previous node. So it gives me a better code and stable code. Yeah. I think that uh you know, we've been talking a lot about like two things that you need to know as a Vibe marketer are what questions to ask and what good looks like. And I think that that, you know, goes across the spectrum whether you're you're Vibe coding or or you're Vibe marketing. So uh often times we don't know like what code to put in or whatever. I'd say 99% of the time. And I do the same thing. Um you know, my desktop is littered with screenshots of like errors and things like that. So

14:43I'm just going back and forth until I get something that works and I'm testing until I, you know, work through the bugs. I think a lot of people expect a one-shot output. It doesn't work the first time and then they get frustrated, but that's just the wrong approach. Like no one's one-shotting stuff. Maybe like

15:021% of people or less, you know? Yeah. Yeah. If you if you even see this, right? The code is super user-friendly. Right now if you I don't think I'll be able to write a code, but I can understand what this code is doing because of a lot of two and fro code conversation, right? Like it has a lot of null data over here if you see. Like this data or this data will be not necessarily relevant. So it basically removes null. So skip entries with no title or self text. It removes nulls and things like that. So we get the right set of filtered data that we are looking

15:35at. So this tells that this part is basically to skip the entries with no title or self text. And this part is basically to remove nulls. So basically it gives you an idea that what you're going to do with this code. Perfect. All right, nice. So you have your code, you're removing null, you're,

15:54you know, getting some structured data. And I think the next step you said is you're leveraging Grok. Yeah. So I'm I've just used a very very simple um you know, uh prompt over here. And usually like you say, right? Like uh just go to even if you're looking for a prompt that you're you want to put it in an A10, I just go back and uh ask ChatGPT that I want something like this and and the output has to look like this. So it gives me a very structured prompt. And the advantage in doing this is that let's say if I'm going to use Grok, I go and use Grok to

16:31get the prompt. If I'm using GPT, I go to GPT to get the prompt. If I'm using Claude, I go to get go to Claude to get the prompt. I've seen differences over here because uh Grok has a different algorithm in terms of returning an output and it needs certain things in the front and it has to be structured. The prompt has to be structured as per Grok. The same way Claude works that way and also GPT. So I think going to your specific LLM models to get the prompt is a great way to tune

17:02the model like you want. Mm. That's a really good insight. So um you know, and and we talk a lot about using different models at at different stages or for for different jobs. Um what do you find that Grok is particularly strong at doing? Um couple of things. One is that when you take Grok, right? The one of the things is the real-time data data it gives and also the logical reasoning that it has when it comes to short type of content overall like, you know, text-heavy content. Like when there is text-heavy, I obviously go for O3 Mini because it can take the load. Grok at

17:43the moment in the API level level cannot take the load. So I basically go for a logical reasoning plus uh very relevant data output that I'm looking for. I we choose Grok. But if it is text-heavy and we need a very in-depth analysis, O3 Mini. Okay, cool. Sounds good. So what are you doing here with with Grok? What's your prompt and what are you instructing the model to do? So basically I'm asking Grok to basically look into the self text, which is the Reddit post, and asking it to come up with a summary for

18:17each of the post. And what is the theme? What does the theme cover? Like summarize the Reddit post in one sentence as a summary. What is the main idea of this particular topic? And what is what are the pointers that it is entailing in the post, right? Like what are it what is it addressing in each of the post? So I think these are very very important. Like breaking the post or any sort of text that you get, whether it's a blog or any smaller content for that matter, as much as you break that into multiple set of information, it becomes easy for us to like research and do

18:52analysis in better way. Cool. Very helpful. All right, cool. Makes sense. Uh so uh what happens uh what happens next? So we get the post summary. Uh we also get So if if you see here, it's like all the data is like not very structured again. So there's a theme, there's a post summary, and there is what else? Yeah, points from the Reddit post from this particular post. How is

19:20it broken down? So you will get that. Now the next thing since the output is like not really structured, it is JSON output and it has things like this. We're going to structure it using a code. Yep. Uh by the way, I saw your battery uh notification come up. You may want to plug in. Yeah. There are a lot of battery anxiety people out there. You know, when I sometimes when I make a post on X and I'm like in the red, people are like, "You better charge your

19:46phone." I'm like, "Thanks. I know that." Yep. Yeah. It's all good. Cool. Okay, so we're back to structuring the output again. Yep. So we get an output from Grok, the summary, theme, and what point it entails each of the Reddit post. So it looks like this, Reddit summary, theme, and points. Yep. Okay, makes sense. So the next thing is we are merging these two nodes, which is the clear like the clean that we saw from the Reddit as well as the post from the the summary theme and

20:26points from here. And we are again putting this into OpenAI model because this time we have like lot of content for it to like go through. So what we are trying to do is a very simple prompt. We want to like look at content idea themes. Like I want to basically looking looking into all of this, like specifically the text and the summary. I want to basically like come up with content idea themes and what each of these idea themes should entail as pointers. So this will be helpful for me to further break that down into different content ideas and further take it down into angles and, you know, hooks

21:03and stuff like that. So this is the next step that we are doing. Okay, and on this step you're not using Grok, correct? You're using a different model. Yeah, I'm using OpenAI over here, ChatGPT. Cool. So here if you see, we get something like uh hey, these are the content themes that you can explore, introduction to uh marketing automation

21:26based on this particular post and theme. What can you What can you actually like get into it? Like you can define what marketing automation is, you can break down popular marketing automation platforms. What are the benefits of automation? So what things you can actually like consider when you take that as a cluster or the uh you know,

21:44topic to basically explore. Got it. Okay. And we are using a set node here because to some extent I wanted to like come out of code and I wanted to like juggle between code and set notes so that it is easier for someone who doesn't want to code to look into this. Um for example, I just want to get this

22:05theme like this. Mhm. So I'm going to use content theme and this is again, I go to ChatGPT to get it. And I want to basically slice this particular content and I want to base structure that content separately. So I get something like this. Okay. I'm going to merge this this one and the Reddit post and the summary of

22:32the post and things like that together. I'm going to with merging, you're just bringing these different kind of structured outputs together in into one place. Exactly. Yeah, otherwise you would only get the output from the previous node. Right. And again, we can use multiple option. We don't use append, we use combine because let's say that if you see it is basically for the Reddit post and you're getting the content ideas around that. So to get throughout, it's basically based out of

23:04the Reddit post. So you can combine it. Append is like when you have let's say from node one, there's like different set of things that you're trying to get as an output and node two, there's like different set of output. When you want to get both the output, that is where you use like append. We don't use

23:21combine over there. Okay, perfect. Now I'm going to redirect all of that into an Excel like Reddit post title summary, theme, subreddit, points from each Reddit post, and content theme. Yep. Like you see here. you're Yeah, so now you're you're starting to get all this into one place, into a Google Sheet so that you know, you can really sort of look at, share, edit, you know, start to paint the picture of how you can

23:52actually make this usable as well. Yep. Exactly. And the other reason can also be that when we have these Google Sheets like you can also use Google Sheet in different places here as well, you can store it. The advantages of using Google Another advantage of using Google Sheet is that uh you know, let's say for example, you want to test from let's say

24:16the workflow stops or there is an error. Right? You don't have to go through the entire step from the beginning. This also I learned very late that I just simply like duplicate this uh remove the trigger. I use the trigger and this to further run the set the other the next set of my workflow to see if it is working. Else, every time I do the test I'll have to like, you know, start from the beginning. It takes time and it also

24:43like takes a lot of token. Yeah, sounds kind of like a little bit of a like a version control type uh thing maybe a coder would use that to isolate the variable that's causing an error instead of going back to the beginning or something like that. Yeah, so basically I tried to do that. I learned it very late and I would really suggest someone to not go through this because it definitely cost a lot of

25:08time and API. The The best lessons are learned the hard way, I guess. Very true. So we got all of that from Google Sheets and what we're going to do is we're going to basically take the content theme and ask it to look into hooks and angles to explore those themes. Because I think for any sort of post, whether it's a blog, LinkedIn, or Twitter the hook and angle is like very important. I think the angle actually differentiates the content that you are trying to build versus what you

25:41see outside. Yeah. So I think this is a important part of the research as well. Right. And and just to um you know, help folks that you know, may not be familiar with what is a hook or what is an angle or you know, something like that. Like a hook is one of the most important pieces of your content because that's what's grabbing people's attention and sort of like bringing them in. You know, when you watch a video on YouTube or whatever or a short-form piece of content, like they all start with a really strong hook in the first, you know, two or three seconds. Uh it gives you a reason you

26:19know, why you're going to engage with that content. And the angle is kind of like your unique uh take uh on the topic. Um and what benefit uh you're giving uh the the people from engaging or reading or watching or whatever. Exactly. Yeah. So even if you see over here, right? Like what could be interesting angles and hooks that you can explore when you say introduction to marketing automation

26:49says stop downing in task. Uh how marketing automation saved businesses 20 hours a week. Uh and it basically explores angles like the psychological relief of eliminating repetitive repetitive tasks. Uh the paradox of automation, how mechanizing processes create more authentic customer connection. Like these are angles that are super uh important for us to build content. Like imagine if I have multiple angles and I can filter through those angles and say that hey, this is the angle that I want. And you can add a human in the loop so that before creating content, you select those angles and build ideas and content on top of it. Yeah. And you

27:30know, what's valuable about this too is that you know, you have one topic or whatever and it's difficult to think of all of these angles and all of these hooks and and things like that. Um this helps to solve the cold start problem. It gives you a bunch of options to to look at. It gives you a bunch of different things that you could potentially test out in the market to understand maybe a unique spin that customers or I'm sorry, competitors are not utilizing. So this is a way that you can find differentiators in your content leveraging the power of AI to you know, think of a bunch of different angles

28:14that otherwise would be difficult for you to do manually. Exactly. And I think the hooks Another thing that I would like to add is that the information that we get, like if you see over here is something like um while you sleep, how key automated keyword tracking uncovered a $50 uh 50K dollar opportunity for brand X. I think we need to clearly understand that these are like filler or uh exaggerated data points. Sure. So we should These are some ideas. Right? Like these are some ideas that we can take inspiration from and not blindly take that content and then do it. When we build content and

28:54when we are going to write prompt for that, we should be extremely clear set up guardrails and say don't take this content just like that. There are some hallucination. You define all of that. Avoid this if if it is not provided. If there is no data that you have with the input that is provided do not take that data is something that we need to instruct the LLM model so that it doesn't hallucinate when it is creating

29:18content. Yep, makes sense. So yeah, so I think then we are again like using a set node to get that into only the hooks into a different place. We are merging that and yep taking it to the sheet where you can see the hooks and angles for all of it. Awesome. Fantastic. Yeah, I mean, again what's cool about this is that you're going to one subreddit, one topic, and you're able to like totally expand on that one thing to create, you know, tons of content ideas with multiple hooks and angles and and stuff. So, you know, somebody might be like, "Okay, this is great getting all this from from

30:08Reddit, but what what's the use case?" You know, well, let's say you need ideas for ads, you know, one of the ways you succeed in advertising is by testing a lot of different ads. And, you know, while everyone else is scraping their competitors ad library and recycling hooks and angles that everyone else is using, the person that's going out there and understanding like the conversations that are trending in communities is going to have a bunch of fresh ideas that others have not already taken

30:44advantage of. So, that's one. So, I think that, you know, looking looking at at channels like this really give you an opportunity to create content that that stands out from the crowd. Yep. And the angles are really really good. You know, when you fish through it and I'm excited to basically like build on top of it where, you know, taking a specific angle, come up with multiple content ideas for different channels, and build content on top of it. Like

31:13it's going to be interesting. Yeah, absolutely. And, you know, another way that angles are useful would be to highlight, let's say, in your website copy or something like that. You know, there might be a hidden angle that really makes a potential customer, you know, kind of light up. And that might be the thing that pushes them over the edge to to convert on your product

31:34or your service or something like that. So, I think if you're in a competitive market or a crowded market or something like that or your copy and your content might feel like me too, this is the type of research that you should be doing. And, you know, this is the use case and and the benefit of doing something like this. Again, if you were to try to do this manually, it would take you just

32:00can't do this manually. Yeah. Yeah, and I think one more thing that I would like to add, James, since it is Reddit, right? Like when you mentioned about website angle and things like that, imagine the pain point you can capture of the target audience around the subreddit where people are talking about that, you know, you understand what is ideally like missing in the market, potentially missing in the market. What people are facing, like multiple people talk about, "I've used Gumroad. I've used Any.do. I've used this. I've used that." And when you see, you would see an underlying pain point when they discuss saying that, "Hey, I

32:33tried to use Any.do, but it did not really work for me because of the learning curve." Right? And when you are building a product or something or a service around it, you can basically capture the pain point from there. And like you said, for the website that you write a copy, this will be like a like literally like a golden goose for you to basically like pick that and just put it drop it into your website so that it converts. Right. Right. Yeah, I'm I'm also excited to see how with future

33:03iterations we can leverage comments. You know, maybe there's a way to, you know, look at sentiments and extract more information to dial in on the customer language or the pain points or, you know, something something along those lines. I'm excited. I think the next thing that I'm going to work on is look into the comments and see what we can extract from there and how we can use it. Like multiple use cases, even like, you know, we can ask AI to basically, you know, how would you map these pain points for different content assets and where can be used and see what it is actually like giving us

33:40the suggestions. Yeah. And, you know, I've been asked a lot like, "How do I grow on Reddit?" Like and, you know, it's hard. Like people don't really know how to interact with or engage in a Reddit community unless they're like already a Reddit power user. So, when you're coming from the outside as like a marketer or something like that to the platform, it can feel, you know, very intimidating. Like, "Ooh, I don't want to get roasted on here. I'm I'm scared to say anything, you know?" So, I think that another valuable use case would be just giving people a bit of a co-pilot in terms to to guide them on how they

34:21should interact with a certain community on Reddit. Like, how are commenters talking? What's the tone? How are they resonating? You know, are they asking questions? Are they adding a lot of value? What topics are are things to avoid and what are topics and discussion points within threads to go deeper into? Absolutely. Yeah, that's a great angle

34:48to look at. Cool. Well, this is really helpful. Great deep dive into what we would consider kind of like a starter workflow, but also a lot of other, you know, nuggets around how you can utilize this information and take it to the next level. So, thanks for showing this to us. Thanks, James. Thanks for sharing some interesting ideas. I think I'm going to definitely implement in the next

35:12workflows. Awesome. Sounds good, Guru. Until next time. We'll talk soon. Talk soon. Bye. Bye.