Introduction

Thank you, Reggie. Yes, so I'm going to quickly introduce myself and go through my experience. It's really just a personal story.

Yes, I made the slides with AI. I think that's the great of this. And yes, as Reggie said, we really met only 48 hours ago.

ago, so I'm generally here just to tell my story and how I used LLM in the past five months to build a really, like, I'm trying to build a product while working full time, and I just launch it.

And the good thing is that I'm not an engineer. I'm not technical. So as I said, I'm not technically.

I just try to leverage as much as I can, like, my expertise in some companies, in processes, and using the right technology.

From CRM and RevOps to Building Software

I am CRM and revenue operation lead in Zurich. So finally, the video before your spreadsheet, CRM, yeah, that's what I do in my life. I try to help company not having their CRM in a spreadsheet.

But I always wanted to build a software. This was always like something interesting for me because I like the idea of coding something programming something and see the immediate impact of like an effect of what I'm doing and I I always love this but for various reason I ended up in doing data analytics and CRM

How AI Changed the Motivation Curve

something clicked when this AI era started a little bit more than two years ago and this is my experience and with it basically what AI enabled me to do and it's this basically creating something that works and see the progress of it like in in few little steps that give me a sort of dopamine

aid effects like I don't know how many of you played role -playing games where where you level up with your experience. Sorry for that. So that's what the AI does for me.

Before, I would start a blog, a side project, and I would see the results in months. By then, my motivation would just go down. And I would not have the consistency to keep going without seeing little results.

And that I think what AI is great for me, that as soon as you do something, you can validate immediately and you can see like, oh, wow, this is working. So let me do a little step for it, a little step. So I try to visualize this with this 0 to 100 seems like a big milestone. And now I feel like I'm doing small wins every day. I see the progress and it's like leveling up in a video game.

so while I was building my product I was seeing okay I build a page it works now I'm building another page now another screen another feature everything works everything is saved this store like it's real and it's getting compounded every day I have this dopamine effect that motivates me and then it compounds and

so now I want to tell a little bit the story and do a quick demo on what I I managed to build in these five months.

The Problem: Research Is Moving from SEO to LLMs

And so I need to step back a little bit in my research process, which is, how do I research thing?

And I think you guys heard all this switch to SEO, search engine optimization, into generative engine optimization. And basically, so that was the trigger for me.

Why AI-Driven Website Traffic Is Underreported

I started seeing in the company I currently work for that AI website traffic is only 1 % according to kind of official analytics but actually if I look at my search process I'm not spending only 1 %

of my time in searching brands tools products hotels asking things only 1 % so this 1 % doesn't reflect the reality of at least my way of researching searching information.

And so I asked myself, is there maybe an opportunity to help companies looking at this dark traffic?

“Dark Traffic” and Why It Matters

So you search for something, then maybe one day later or a few minutes later, you don't go through the click funnel that is usually what marketing attributes to a source. You go in a different tab, you maybe check the day after.

so there is so -called like dark traffic you land on our website but that website doesn't know you found that product on chat gpt because there is no way to track in this at the moment so i try to solve this problem and to basically build something that help companies understand why they are getting traffic from ai and how they can improve it because i don't know if someone of you is in this

this field, but turns out that this traffic converts a little bit more, like five times more on average.

And there is also this effect of people searching in ChatGPT for something, and they don't even land on the website, because they have gathered all the info already.

Before, I would usually search on Google, blogs, I don't know, a couple of articles, Trust Pilot, Trip TripAdvisor, like at least 10 sources, and then I would always land on their website now. And now

the LLM gives you all this information if you do web search mode in one answer, or maybe a little bit more if you iterate. So sometimes you don't even need to go to the website.

And imagine from the website perspective, that company doesn't even know you have landed to their website, site, but your content was crucial for ChatGPT to show the right information to the right user.

So the intent is not capture, but actually the user is really interested in your product. You just don't know.

So obviously, the first thing I ask myself, surely someone built this, because it's always like this. for anybody of you will probably try to find found a product I think it's always like someone else might have thought about it so it's true like obviously

there are different tools and products that are already in the market but I tried anyway because I wanted to build it and also I think it's the right timing so I launched this month this tool and it seems I'm getting some

traction I just put some numbers but it it's not the purpose of the slide but

The Solution: Measuring AI Visibility and Recommendations

what does it do it's called lonely and it measure often AI model recommends your brand when people ask questions and basically it tracks that dark traffic by by combining what the chat GPT bot is looking at your website and what the real visitor is doing, combines with the time kind of alignment, a little bit of source tracking, a little bit of inferred model. And then it tells you an estimation of AI traffic.

Plus it queries LLM models and search how your brand is positioned so it gives you recommendation on how to optimize

What LLMs Reward: Structured Data and Authority Signals

your website the most important thing in this is structured data LLM loves structured data they don't love videos so humans we love graphic I love design I my decision -making process of buying something is very much on the how it it looks, how it feels. But that's not how LLMs apparently think. They want structured data, because they need to fast access the information of your website and gather the right content.

It's basically SEO for AI.

So I want to quickly show how it works, just to showcase how I could build this, starting from the end result.

Product Walkthrough

Free AI Visibility Report

So basically, Basically, this is my website. And the first thing I built is a sort of free AI visibility calculator, where I can basically put a domain.

I'm going to put on shoes, because I guess you all know this brand. And this gives you a sort of report that everybody interested could run for free,

no need to sign up. and the idea is that the the tool checks multiple sources obviously where is chachi PT Gemini perplexity Claude with industry queries of on shoes so in this case like on shoes a hundred percent visibility in running shoes because if

you search for running shoes hundred percent of the time any LLM would would recommend you on in the answers. But it also shows how the competitor shows up.

And it calculates some sort of positive, negative, neutral sentiment, the generative engine optimization score, and how many mentions your brand has in those queries. This is a benchmark.

It's not the entire billion queries users run, but it's to give a brand a sort of first AI visibility report. And then what's important in this, I realize it's the brand authority.

LLM have one goal, and it's they don't want to hallucinate content. This is super important for whoever is working in the field. I realize the more they recommend bad quality content and sources, sources, the less the users like and the less the user's trust is basically gone.

So their way of citing a source is to measure how much that source with domain rating, brand authority, how many mentions. YouTube and Reddit seems to be the most important sources, which is fascinating to me.

I don't think a lot of companies invest in Reddit presence or YouTube, but that counts a lot for LLMs. And also reviews.

So the idea of this brand report is that it gives you a glimpse of what the platform can do. It's static.

Logged-In Dashboard: Analytics + AI Traffic Estimates

And then when you log in in the platform, you basically, it's a sort of Google Analytics plus AI visibility. ability.

It gives you the usual traffic KPIs in this simple, clean dashboard. And you can also toggle AI traffic only.

It would show you what's the AI traffic value and share of your traffic. Now, this is real. It's my website using my tool, which is amazing, I think.

So this is basically, it's good for me, it's very nice for me to see that AI bots are crawling more and more my website and basically I can see here every single log of who is visiting my website and if it's coming from AI or not.

How It Was Built (Step-by-Step, Non-Technical Approach)

and so how I build this now I want to stay really in the 20 minutes and share how I managed to build this even being non -technical and this is a pure like step -by -step experience I want to share and the main important thing is I spend

Research Workflow: Perplexity and Claude

I spend a lot of time in researching the problem. And usually, I use Perplexity for this. I'm not sponsoring Perplexity. But I just feel it's the fastest access to reliable information.

So I don't know if some of you use deep research in Perplexity. I think it's the fastest and most reliable. But it's also a little bit not so human -friendly as Claude. so I also use Claude a lot.

Once I have my research of the problem, for example, is GeoScore important for companies? What is GeoScore? Okay, yes, it's this way of optimizing website. Okay, I know my problem, what do I do?

I research again how to build it. So what are the data sources that make sense when calculating this. So what are the tools? What is the backend?

And turns out there's so many tools launching every day.

So my way of working is always to give a context of the current day, like as of January 26, because LLMs tend to be lazy and propose you what's in their training knowledge and not what's available now and you've seen that a lot of things have been launched in the past two years so if you don't force them to tell you what's now they're gonna recommend you something that it's already hold because in one year ago we already have older things usually once I know what I want

to build let's say back -end tables this logic data source I write it down in a a system of record.

Sorry, I don't know what happened.

Keeping Context with a System of Record

1And my system of record is an issue tracking tool. I don't know who uses this, but it could be Notion. It could be a note. Doesn't matter, but it's something that I think should have a date, should have a status, and a clear what needs to be done, why, and where. there.

And this enables me to basically work well with the context of the information I want to give to my coding agent.

I struggled months in finding a way to keep the context because I don't know how many, but my biggest struggle with LLM is that context is like your memory. You can do small tasks and the context will be enough, but when you try to to build a software, you always need your LLM, coding agent, whatever, to know what's behind that software. And you cannot just simply copy -paste every time that prompt, because it would be impossible.

So my system of record, which is issue tracking, linear, I can use Jira or Jira, Notion, whatever, whatever, it's something that is task, date, status. And you have to keep that updated.

Because then, when the next time I'm going to do something, I'm going to tell, hey, go to my history of all my issues. What's relevant for this feature? What was done? What was wrong? Why? And see the files that need to be changed and do it properly.

And this would, like, this approach, I figured it out after three, four months, has reduced the number of, I call this moment, flying into a wall dramatically. Because I couldn't find another way to keep my context memory relevant and updated.

And this is not something LLM can do. You have to do it. You can obviously use LLM to document stuff. stuff. But you need to be diligent in always following this approach, or at least that's what worked for me.

Tooling: Coding Agent, Integrations, and Git History

Then I obviously used a coding agent. I didn't use Vibe Coding App because simply I thought for my use case it was too much complicated and I wanted to learn how a proper development framework works.

So I use Cursor. It's a coding agent that That is basically for developers here. It's like VS Code, but with AI integrated.

I think VS Code also has it now. And I connect all my tools into the coding agent. So if I have a database, the coding agent has access to the database.

If I use perplexity for doing research, I connect my coding agent to perplexity, because then it's way better if I let the agent do the research because if I do it myself I have to copy paste things around context is lost

while the agent can do the research if you connect with another agent I know could sound complicated but I am here to just give my what I've what I've done

then also for people will build software here we all use git github I guess the The commit history is also like my best friend. Every time I don't even remember what I've done, simple. You just ask your agent to check the commit history.

Everything is indexed. Everything is versioned. So the file you modified one month ago is going to be in the system.

You can tell your LLM agent, AI, to go and search for what happened in that file. So it's important to go and version every work you do.

Shipping Mentality: Break Things, Fix Fast, Keep Learning

And then, I mean, I obviously broke things many times. I mean, I wouldn't recommend this to any enterprise software developer. Don't do this.

But in my project, I realized if I break something and I fix it in 30 minutes, It's OK. It's way more effective than break my head in five different development environments,

test one million instances, and then lose three days.

What I find fascinating in this era is that when you... That's exactly what I want to share, that AI didn't build things that I think I couldn't build before, but made it 10 times faster.

In another sliding doors, another world without AI, I maybe could have built this, but it would take me 10, 20 times more and probably 100 times more money.

Conclusion

So that's basically my main story here, that I tried to find a problem, I was consistent in doing it with a kind of proper methodology, And then I wasn't shy or too worried about building something ugly at the beginning and break things because I think you can fix it very quickly. And if you fix it in your working session, it's fine. Nothing is going to happen.

And it's all about learning.

Q&A

Yes, if there are questions, I hope I was not too long. I put just a QR code if you want to connect with me on LinkedIn. Any questions?

You know, for someone to...