How I use Reddit and AI to find winning startup ideas (2025 tutorial)

        1. **The core problem** - 90% of AI businesses fail because founders can't identify the 1% idea that makes people actually want to buy

2. **The framework overview** - Generate hundreds of ideas → Validate 10 → Launch at least one

3. **Five validation methods:**
   - **Method 1: Pre-validated ideas** - Using Y Combinator RFPs and VC thesis pages as starting points, then using ChatGPT to personalize those ideas to your background
   - **Method 2: Reddit social listening** - Lurking in subreddits to find complaints about existing products; automating this with tools like Gum Loop
   - **Method 3: G2 reviews for B2B** - Scraping enterprise customer reviews to find feature gaps and positioning opportunities
   - **Method 4: Google reviews for local businesses** - Finding businesses with poor reviews and targeting them with solutions to their specific problems
   - **Method 5: Community engagement and customer interviews** - Joining Facebook groups, providing value first, then recruiting users for interviews

4. **Automation as a time-saver** - Using AI workflows (Gum Loop, Clay, N8N) to scrape and summarize data that used to take hours manually

5. **Product-led growth philosophy** - The Clay example of spending a year talking to users before building marketing automation

---

AI has made it easier than ever to build a business, but 90% of businesses will still fail because of one single thing: people just don't know how to come up with that 1% idea that makes customers actually want to buy.

In this video, I'm breaking down my exact framework for uncovering high-potential AI software ideas. I'll show you how to generate hundreds of ideas, validate 10 of them, and launch at least one—all by leveraging publicly available insights on the internet using AI research assistants that save you tons of time.

I'm Alex. I'm a surgeon and the founder of several AI companies, all of which have scaled to over $1 million in revenue in their first year. I'm going to cover five methods to validate your idea.

---

### Method 1: Pre-Validated Ideas

This is where you start when you have a blank canvas and absolutely no idea what you're doing. It's all about looking at pre-validated ideas.

One of the best places for this—even if you don't want to raise VC money—is Y Combinator's RFP list or the thesis pages on VC websites, like A16Z's AI agent thesis. The Y Combinator team has already researched these trends. VC analysts are constantly scanning the market for emerging themes. This saves you a ton of effort at the beginning.

We can use AI here too. I grab the Y Combinator RFP list, jump into ChatGPT, and add a prompt with information about myself and my interests. For example:

*"My name's Alex. I'm a surgeon. I'm interested in building a business within the medical space. Please use this Y Combinator RFP list to generate me 100 ideas I can apply to medicine that give me an unfair advantage."*

The more specific I am, the better. I might say I want to create something that revolutionizes surgical training, or a voice agent for automating medical note-taking.

Now, the problem with ChatGPT is it's a little basic for what we want. We're not interfacing with actual customers yet, and the ideas tend to be good but vague. But at least now we have a direction to dive into.

---

### Method 2: Social Media Listening

The beauty of forums like Reddit is that you've got people discussing their problems with existing software and industries—all in the public domain.

When I built some of my first companies, I would just live in Reddit subreddits related to medical training or AI and look at the problems people were having. One of my first businesses was around medical exam preparation. I would lurk in subreddits like USMLE or medical doctor communities and look at what people were using and their complaints. They might complain that an existing exam prep company wasn't very good, or they'd ask for help preparing for an exam.

This gave me an idea for a business where if I could create it and market it to those people, I knew they were likely to use it and pay for it. It also gave me insight into how much people were spending on existing tools, so I had a good price point to either undercut or offer better features at a higher price.

That process used to take me hours, but we can automate it now. I use Gum Loop to set up a social media listening workflow. It scrapes conversations from a particular subreddit around a specific topic, uses OpenAI's API to summarize them, and outputs a problem statement list.

Here's another tip: don't just go into generic industry subreddits. Dive into competitor subreddits too. If I was building a microlearning service competing against Kahoot, I might scrape Kahoot's subreddit or Quizlet's subreddit to find customer frustrations and think about how to position my product.

---

### Method 3: G2 Reviews for B2B

Reddit's great, but often those users are individuals who might not be paying customers. What if you want to start a B2B AI company?

One of the best places to go is G2 reviews. This is where enterprise customers review existing products, and those reviews are validated—you know they're coming from paying customers.

I used to go through competitor reviews at Vertie all the time—analyzing what features people liked, what they didn't, and thinking about how to position against those gaps. It took ages manually, but now we can set up an AI automation workflow to scrape and analyze those G2 reviews.

Quick tip if you're not that technical yet: just take a screenshot of G2 reviews or a Reddit conversation, upload it into ChatGPT, and ask it to summarize the problems or pros and cons. Then ask it to help you formulate the killer features and USPs of your own product.

---

### Method 4: Google Reviews for Local Businesses

This method is more geared toward reaching out to customers and potentially building a service business.

We can scrape Google reviews for local businesses using Gum Loop or Clay, particularly looking at businesses with negative reviews. We can also use Glassdoor for employee feedback and look at businesses with low scores, then target them with a solution that helps improve their customer service or employee experience.

We can even use our scraped summarization to send personalized emails. Something like:

*"Hey, I saw your Google reviews are down at the moment and you've got customers complaining about customer service. I wondered whether you'd be interested in an AI tool that can automate this for you. Let me know if you're interested."*

We can link that into our automation by having the AI generate that email using a custom prompt that takes the Google review data and maps it out automatically.

---

### Method 5: Customer Interviews

Now that we're collecting all this publicly available information from social media and Google, it's time to dive into customer interviews.

Go back into communities—Reddit, Facebook groups—wherever people are engaging around your specific topic. Use ChatGPT to find the largest community on Facebook for marketing or sales or whatever your niche is.

You can scrape positive and negative comments using tools like Clay or Gum Loop. But one of the coolest strategies is to actually become part of those communities. Offer helpful advice and engagement without being too business-oriented initially. Then slowly introduce the idea that you're building something for that audience and see if you can organically bring people onto user interviews.

Here's a powerful example: when Adam Robertson joined Clay as their go-to-market lead, he didn't immediately set up marketing automations or push sales. He spent a year hanging around in marketing and sales communities on Facebook, just talking with loads of people to identify their problems and understand how Clay could be optimized. It was all about product-led growth. They've since raised an insane amount of money at a huge valuation because their users love the product and it solves a real need.

In the future, we'll probably be able to use AI tools to automate user interviews and survey generation. But for now, those are my top five methods for validating your ideas—going from that list of potentially 100 ideas on the Y Combinator list, through social listening, down to individual user interviews.

This is something I've used to build $50 million-plus valued companies, and I hope you can put it into practice too.