How I Use AI For Market Research
May 22, 2025
Parker maps out how he’s using AI for market research to reshape his community offering, test tooling, and build a scalable content engine. Practical, punchy notes on what he’s trying, what’s working, and what he’s loading into the AI stack next.
AI tooling and experiments#
- Augment: asked it to produce an implementation plan from a highly detailed PRD; results influence landing-page work (Astro) and overall IA.
- VS Code: using the editor as a strategy workspace to draft and refine plans.
- Postgres plugin (Microsoft): new plugin with schema visualization, database explorer, file navigator, query history, and IntelliSense—aims to bring DB visibility into the IDE.
- ConvexH: attempted to auto-build the entire app; interesting pattern guidance, but/**
- Strong opinions on backend patterns if you’re TS-based; less ideal if your stack is Python.
- TanStack tooling and early beta explorations: experimenting with how TanStack patterns fit into the app/landing-page workflow.
- Astro for landing pages: ongoing choice for front-end architecture.
- Echo (YouTube metadata automation SAS): automates generating subtitles, metadata, and parts of the video script into a production pipeline; current gaps are thumbnail/YouTube interactions and end-to-end UX.
- Notebook LM: slides into a core theme—useful for scaling coding research and pre-planning R&D work instead of prompting blindly.
Market research and ICP strategy#
- Repositioning the community: three subsegments under a single AI-for-developers intent.
- Subsegments: tech-adjacent (dipping toes in), developers on teams who want to level up, and business owners who see ROI from AI skills.
- The aim is a single strong ICP with nuanced targeting, not a generic “everyone who codes” approach.
- Research approach:
- Name-generation prompts and origin-story prompts to study company names.
- Compare and model communities beyond school-based ones.
- 3-tier community model:
- Tier 1: online community
- Tier 2: marketplace with partner distribution and member discounts
- Tier 3: IRL events
- Evergreen content pillars + channel mapping + automations:
- Build a content engine where actions feed back into growth and retention (action 1 → action 2 → action 3+).
- Metrics focus:
- MOM growth and churn as the north star; LTV/CAC is less critical for his YouTube-driven model.
- Target churn: sub-1% if possible; current churn in high single digits due to broad initial ICP.
- Offer design via problem mapping:
- Reddit scraping to identify adjacent problems and craft offers around them.
- Emphasis on riffing and iteration—prompts are thought partners, not one-size-fits-all.
Content strategy and execution#
- Evergreen content pillars and a clear content engine wheel.
- Channel mapping and automations that support value delivery and retention.
- Not just “Parker the news guy”—aiming to be a builder who marketing-savvy; alignment of content with the evolving ICP.
- Thumbnails and templating:
- Experimenting with template thumbnails to enable programmatic generation and more variety beyond Parker’s face with text.
Community experience and engagement#
- Discord automation MVP:
- V0: index two channels, notify members when relevant topics appear; onboarding and post-event roundup.
- V1: add Q&A via DMs and more targeted sentiment prompts; future: smarter prompts and role-based notifications.
- Aiming to elevate membership experience by delivering timely, role-aware information and easier access to important updates.
AI in the software development life cycle (SDLC)#
- Notebook LM is a core example of using AI for upfront research and planning:
- Reduces “blind prompting” by providing structured research and rationale.
- Helps with coding tasks by bridging R&D, architecture, and feature delivery.
- The broader view: AI-supported research, architecture, bug-hunting, and feature development—not just code generation.
Takeaways (actionable)#
- Define a precise ICP with three subgroups and tailor messages to each, then test and refine.
- Use Augment to convert detailed PRDs into concrete implementation plans; treat prompts as iterative collaborators.
- Build a structured content engine: evergreen pillars, channel mapping, and automations to compound impact.
- Design a three-tier community model early: online, marketplace with partner distribution, and IRL events.
- Run a minimal Discord automation MVP to index key channels and notify members; plan DM-backed Q&A in later iterations.
- Use Echo to streamline YouTube metadata workflows; expect to iterate on thumbnails and upload integration.
- Leverage Notebook LM for research-heavy tasks to reduce blind prompting and speed up decision-making.
- Keep churn in the single digits and monitor MOM growth; constantly tighten the ICP to reduce leaky funnel effects.
Links#
- Augment
- VS Code
- PostgreSQL Extension for VS Code (Microsoft - schema viz, db explorer, file navigator, query history, IntelliSense)
- Convex
- TanStack (and related tooling)
- Astro
- NotebookLM
Transcript
Okay. Has a man ever cooked? Can never tell if I'm supposed to play music during these because I get super hype, but then I don't know, maybe I'll get a little too hype. People might they might fall off while you're here. We're doing good things today here on this lovely Thursday, May 22nd. And yeah, I've been using some stuff to do some stuff. What does that mean? Trying out augment right now. And it's bananas. Bananas. Yeah, I asked it to give me an implementation plan off the back of one of my PRDs. And this PRD, it's maybe the most detailed thing you ever seen. It's got a bunch of objectives in it. So, as part of someone who's running a community on the side, there's things you want, right? Every business you have metrics that you care about. For me, it's churn and growth. And I need to stand up a landing page, but then it just went bananas on it and it's trying to do the whole thing in Astro. I'm just like, "All right, I'll let you go." Like you I'm not going to put the dog down over here. Come on. Anyways, yeah. So, this is what we're going to talk about today. It's a bunch of AI strategy things that I'm doing, things that I'm cooking, how I'm feeling. Been using VS Code. Whoa, crazy, right? But it's pretty awesome. And it's got a lot of stuff in it that I spoke about yesterday. I don't have a Postgress database set up, but if I did, Microsoft just cooked something that lets you basically see everything in your Postgress DB. Let's just look at what's I think it's Kiwi Copplestone. Is this There's no Why did I think mixture? I was going to say there's no way it's his name, but yeah, Microsoft cooked this week. New Post plugin with schema viz, database explorer, and the file navigator, query history, IntelliSense. Boom. Pretty sick. basically this superbase thing right in there. Also, I checked out convexh. You watch one Theo video, you're like, maybe I'll check out something he talked about cuz he knows a thing or two. But I totally broke it when I was trying to make it build me the whole thing that I want. I want my entire app built in one shot. Couldn't hang. It's pretty interesting. If you didn't have a backend written in Python, if you had a back end that's written in Typescript, this is awesome because it's very opinionated about the pattern that you need to use. And I don't know, something about that. I think it's kind of loosey goosey in the React world and it's not always great. Like it's very tempting to want to check out Ruby on Rails or something and then you go use it. But no hate on there. But it is the idea of having not batteries included but just some patterns not opinions like strict patterns is awesome but it wouldn't make sense to use something like react query cuz it handles caching it handles real time it handles a whole lot of stuff and I think they're just building more and more into it like I checked out their O and that's still in beta but I'm still Tanstack boy tanstack start and for landing page be using Astro So, and I'm curious to see where augment goes with this. It's just kind of keep cooking. It has an enhanced prompt thing that I've noticed, which has been pretty helpful. And you buy like 500 credits and it's per message. So, I think this same cost if I just hit the API with this is Claude 4 right now. I don't even know what it would cost. Like, I keep making this joke. Hey honey, we can't send our kids to college cuz I just opened a new Claude account. Anyways, we've been filibustering as they say. Hey, screenshots or thumbnails. Go check out this thing. Yeah, we've been filibustering a little bit. That means to beat around the bush. Now we're in the bush. Nope, that's me breaking down augment. This is old thumbnail. Here we go. So, yeah, strategy. So, I was repositioning the community that I'm building and this is the same pattern that you could use if you're launching any product. Our community, our initial customer profile has three subsets, but it's all under the same intent, which is, hey, there's this AI thing. Hey, I should be really effing good at it if I'm going to be developing. So within that subset, that amazing very descriptive thing, it's typically people ages 18 to 44 that have disposable income. They're either working at technology company, they want to get better, they are currently in a development team and they want to get better. So those are slightly different. The first one is they're tech adjacent. They're developer adjacent. They're dipping their toe further into the pool. And then the third one is actually business owners that recognize the ROI they can get on gaining these skills. And I'm not marketing towards vibe coders, which is ironic, but I think they'll find it, but they're serious. There's a trick where people like, "Oh, I made something in bolt." And then they jump into the deep in the pool and they're like, "Ah." Anyways, I was using a bunch of research tools so I could better position what the offer is. And so I was like, I don't really like this name. Now I like it again. but was using some prompts around finding the best names of companies and their origin stories. And that was a cool thing just to see how people came up with it. And then I was like, forget it. And then I went on to finding examples of communities that aren't just school communities. So I think there's a distinction between a company that is independent, that is a community that has multiple tiers to it. So tier one's online. Tier 2 is a marketplace where we partner as a company with other companies. They get distribution through the membership. We get discounts that basically discounts the cost of the memberships from the members. And as you get more members, you get more leverage with the brands and the brands are still happy. So that's step two. And step three is IRL events and actual physical world things. So this is a multipleyear thing and you're just watching it in the earliest days. School is a platform for beta testing, figuring out what that ICP is, all that. That said, needed to do some research to figure things out. So I have this context of the evolution of the product, where things are going, what are evergreen pieces of content, how can I figure out a content engine that makes sense and is essentially on this wheel of actions where when I do action one, it helps action two and it compounds into action three, four, five. And I built that by coming up with a basic product or content strategy and it looks something like this. So you have evergreen content pillars, you have channel mapping, you have automations around the thing. So I'm not just Parker the news guy because I don't want to be a news guy. I want to be a builder that knows how to market really well cuz those two together are sick. So I came up with this content strategy by essentially riffing inside of VS Code. And then once I got done with that, I brought it into a plan. And that plan is inclusive of a bunch of different things with the branding, the definition, engagement ideas. So ways that you can pe people actually gaining more value out of this. And that plays into retention and churn because as a subscription product, you are always focused on that, right? You want to worry about or you want to focus on basically what is your LTV to CAC ratio. That's the lifetime value versus the cost of acquisition. For me, it's zero. like I don't have to worry about that because I just make YouTube videos. But then there's other things that are more important which are what is your MOM or month-over-month growth and you can do that in terms of MR as a percentage or you can do it as like the raw number of members. But yeah, growth and membership while keeping a churn rate sub 1%. That's like the northstar. That'd be crazy. I've noticed that the churn is in high single digits because the way that I marketed early it was it's for everybody and then people come in and they're like I don't know what's going on. There's code here. So you tighten that up and then it fixes that leaky bucket and that's part of the ICP. That's part of figuring things out. And then the engagement ideas of what are the things that we could do? What is it that I would want if I were paying $100 a month for? And the $100 a month price is less. It's basically inelastic if you're dealing with people that are already in the workforce. They're not just trying to get their first money online or whatever. So this also includes an MVP, also includes different engagement initiatives that were taken from what I spoke about earlier from this deep research. And when you look at communities, you have things like Pelaton. You have a bunch of different examples that are across the different areas of really any industry. And this was more naming stuff. I had fed in the history of names. More examples, more examples, more examples. So, side hustle dev, indie CTO, technical PM. This is me kind of riffing through all the different positioning and then ultimately that can turn into an offer. So, what is the offer? What are the strategic me recommendations? And that's what I think people miss is you don't have one prompt sizefits all. It's riffing. It's using it as a thought partner, asking questions, understanding how to take examples of product messaging and marketing messaging that worked on you and then reflecting it and saying, "These are the things that I liked. Here is my psychoraph. Here's who I am as a person. I think there's going to be more people like me. Let's create a shadow audience or an audience that replicates people like me or those that have the same intent and desire. So that's what I was doing with all this marketing stuff. I also found one that's doing Reddit scraping. So figuring out all the different areas of subreddits that are adjacent to what I talk about, which is development, AI, business, SAS, a little bit of marketing here and there, and then finding problems there. And then you build offers around those problems. That's problem mapping. And let's see, template thumbnails. Yeah, I talked about that. I'm just trying out different template thumbnails because I also understand that as I start to programmatically generate these, I want to have more options than just here's Parker's face and some text with a little bit of a wrapper around it. Something more interesting. So, what else is going on? Discord automations. So the existing solutions do not cut it and I want to solve a problem in order to make membership experience elevated where if you were to not be in here for 3 days and you came back and you went into let's say like AI news like what is going on cuz these things are just lit up right all day. And that's awesome if you're active, but if you come back, it would be awfully nice if it had an index in there and it was like, "Oh, I'm talking to Yoda and Yoda knows all." And here's Yoda. And you can give me a roundup of what's going on and what was important. And then based on the roles that people have, they can pick and choose which notifications they want. So I tried to pick one up off the shelf. It was not very good. In fact, it was awful. So I built my own first version. It doesn't do hardly anything, but the V0 of it or the MVP will be that it can watch and index two channels and I'll notify everybody that this is literally a thing that happens and both the onboarding and then an announcement postfacto. And then it'll run it scheduled task. V1 or the thing after V0 will be for it to be able to answer questions. So if something like what happened in this channel yesterday and then it'll go into a DM and break that out so not everybody's looking at it and then later the sentiment on prompts and technologies. So haven't really thought this one out but I think it's interesting. Now on to Echo. This is if you haven't watched this channel this is a SAS product that I'm building for myself where I can go and I hit the end record button which is actually not there. It's on my mouse. But when I do that and I upload it to the website, then it immediately starts all these processes for the 15 or so different pieces of metadata that are methods in the YouTube API that then generate the subtitles. Generate it's essentially taking all the best parts from the script, which is what I've been using, and just does it with generative AI. So yes, this is using generative generative AI, but it doesn't include some of the parts like the thumbnail and also the YouTube integration is very janky. Meaning I have to I can't actually use the app. I use the website and then the modality of actually getting it to YouTube means you have to take all the assets, generate them separately, paste them into a different place, and then bring them back and restyle them. And I just want to streamline, that's all. And I don't see them doing thumbnails in the next month. So hopefully by then I'll be on to shorts and then on to blogs and all the things because you're going to multiply Parker everywhere and then it won't take much time for me to do this. I just want to spend time writing SAS and then doing this part, not the stuff around it. We spoke about this. We Oh, yeah. Lastly, I made a really good video that I'm excited about with Notebook LM and how you can use it to scale up. I think there's a lot of things there's a lot of areas basically in software development that people are trying to figure out how to use AI for. So, I think the most obvious one is coding. Everyone knows this one. It's the actual execution. But then research is a big one, right? You have R&D, research and development. You have bug hunting, you have architecture, there's all these areas essentially, and I'm trying to plug the holes in the boat of the AI SDLC or AI software development life cycle. And so then you have coding could also just be broken out to feature dev. You have new project. That's not even one because it's technically architecture. But for the case here, I look at it and I'm like, this one needs more help. And so I found that doing the research and the homework ahead of just going and blindly prompting cuz you know the blind leads the blind in a way is that notebook LM is just absolutely perfect for it. That's it for this video. If you liked it, make sure you like it. If you learned just even half of one thing, you have to subscribe. Yeah. See you.