An AI-Built Booking Site, a Talking Turtle, and Zero Followers

Recently, I wrote that I have two 'faceless' projects I'm working on. One is a credit education/repair platform for women, and the other is for a friend of mine who owns a beachfront condo in Potrero, Costa Rica. I'm building her an AI-built booking site from scratch to see if you can really do the whole thing with AI. Two bedrooms, ocean view, shared pool, steps from the sand. It's been listed on Airbnb for a while, guests love it, reviews are solid, and it's the kind of property that basically sells itself once someone finds it.

This started out because she asked me if I wanted to do a trade/barter (confession: she was my therapist after I lost my husband 23 years ago - I continued working with her for years, she is retired, and is more of a mentor now, but we've become friends). She has a couple of properties in Costa Rica (her having had property there for 20+ years is why I made the leap to move there, sight unseen, for a couple of years in 2021), and asked me if I wanted to create a guidebook for their guests. Of course, I said yes (easier because I know the area), and once I did, I got this wild idea to create a booking site for her to see if I could really push my AI skills and do the entire thing with AI. I knew I could do the site and the technical pieces, but the goal is to create faceless videos: tours, some of Tito the Turtle (more on him in a minute), and promotional social videos.

They're keeping their Airbnb listing (as they should), but this is the longer play. And... no fees (we can connect to the Airbnb calendar so there aren't any duplicate bookings).

But here's the thing about Airbnb: you don't own anything. Not the guest relationship or any listing data (plus, we're going to make sure they show up in Google local for the area). You're paying platform fees on every booking (this used to be shared by the guests, but Airbnb shifted to the hosts paying all the fees in late 2025), and you're one algorithm tweak away from your listing disappearing into page four.

So I made her an offer. Let me build a direct booking website for the property, and let me do the whole thing with AI. I, of course, want a testimonial when it's working (I know it will work; it's just a matter of what that looks like). My thought behind this and my credit app is to create the systems and processes that make this something I can duplicate (if I want to). I love the idea of creating different verticals where I can just do the work without having to be a personal brand (which, of course, I'm keeping).

Not as a hypothetical. Not as a "what if." As an actual, functioning, live-on-the-internet website with a booking system, a local area guide, and an AI-powered concierge chatbot shaped like a 47-year-old sea turtle named Tito.

This is Part 1 of that story.

Can You Build a Vacation Rental Site Entirely With AI?

Yes, and I'm doing it right now. The site is live, the chatbot works, and not a single human developer, designer, or copywriter was involved. However, to be transparent, I'm not completely new to this. I spent 10 years building WordPress sites (with the help of a developer), have studied and written plenty of copy, and have a decent eye for design if I do say so myself. The point in sharing this is that I bring plenty of web experience to this project, so someone without a similar background will have to work a little harder.

Here's what I wanted to know: can you build a legitimate, professional vacation rental site, one that could actually compete with an Airbnb listing, using AI for essentially everything? The copy, the design direction, the local content, the chatbot, the SEO strategy, and the build itself.

I'm not a developer. I can vibe-code my way through every project I'm working on (and I'm getting better at it), but I'm not sitting down to write Next.js components from scratch. What I do have is the background I mentioned, and more importantly, I lived in Potrero for almost two years. I know the area. I know which restaurants are worth the drive, which beach has the calmest water for kids, which road will eat your rental car in the rainy season, and where to find the best casado in town.

That local knowledge turned out to be like pouring gas on a fire.

Why Not Just Stay on Airbnb?

Because you don't own anything. Airbnb is a distribution channel, not a business you control.

I want to be clear: this isn't an anti-Airbnb project. The property stays listed there because Airbnb is a distribution channel, and a good one. People search for it, find the place, and book.

But distribution channels are not the same as owning your online presence. When someone books through Airbnb, the platform owns that guest. You can't email them directly. You can't build a relationship. You can't offer them a discount to come back next year. And as of late 2025, the platform fee I mentioned above? 15.5% on every booking. The old split-fee model, where guests absorbed most of the cost, is gone. That 15.5% comes straight out of your payout before you see a dime.

A direct booking site flips the whole equation: same price for the guest, no 15.5% platform cut, a direct line to the host, and over time, a brand that stands on its own. The challenge is that building one from scratch is a lot of work... or at least, it used to be. 😉

What Does the AI-Built Booking Site Look Like Right Now?

The full site is live with a homepage, a condo detail page, a comprehensive local area guide, and an AI chatbot concierge named Tito. The domain isn't pointed yet (working on a few things with my friend, and we'll have to do the calendar connection together), but everything works.

Here's what's there:

The homepage sets the tone immediately. It doesn't read like a template. It reads as if someone who actually lives (or lived) in Potrero wrote it, because that's exactly where the content came from. The copy talks about the sodas with the good casados, the dive bar with cold Imperial, the fact that Potrero is 14 minutes from Flamingo (close enough to borrow its restaurants, far enough to skip the crowds).

That specificity isn't decoration; it's strategy. When your content sounds like every other vacation rental listing, all "pristine beaches" and "breathtaking sunsets" and "tropical paradise," you sound like a brochure. And brochures don't build trust. Specificity does, especially if someone thinks they're getting one thing and ends up with something else entirely (there are three beaches all within about 10 minutes of each other, and each has different sand).

The condo page walks through the property itself: two bedrooms, two baths, a king bed in the master, twins in the second room, an ocean-view kitchen with a dishwasher and breakfast bar, outdoor dining with a gas BBQ, and beach towels and chairs ready to grab. The details a guest actually needs to picture themselves there.

The local guide is where it gets interesting. This isn't a generic "things to do" page. It's a comprehensive, opinionated guide organized into sections: getting around, beaches, restaurants, activities, wellness, essentials, and a "good to know" section that tells you to shake out your shoes before putting them on. (Scorpions are real, and they like shoes... You've been warned. Fortunately, I never saw one when I lived there).

Every restaurant and business listing includes a WhatsApp link (because that's how everything works in Costa Rica) and a Google Maps directions link calculated from the condo. The kind of practical, on-the-ground details that make someone feel like they already have a friend there waiting for them.

What's an AI Concierge Chatbot, and Why Is It a Turtle?

Tito is a 47-year-old olive ridley sea turtle who serves as the site's AI-powered local guide, and he's live on the site right now. He's the feature that makes people stop scrolling, and honestly, he's been the most fun part of this whole project to build.

Tito was born on Playa Potrero and has lived there his entire life, so naturally, he has opinions about everything: which beach is best for families, where to eat tonight, whether you need a car, what the surf is like, and why you should absolutely do the ATV hidden beaches tour through Flamingo Adventures.

Guests can chat with him directly on the site, ask him anything about the area or the property, and get answers that sound like they're coming from a local who's been around a while. Because that's exactly what his knowledge base is built from.

Here's what makes Tito different from the generic chatbots you've seen on every SaaS site: he has a personality. He's warm but unhurried, speaks like a local rather than a brochure, and has a dry sense of humor ("I don't have shoulders, but I respect the craft" is his line about the local massage therapists).

He never sounds like he's trying to sell you something (and he's super cute; you can see Tito in the video, but I'll share more about him in Part 3, where we talk about the design. Although he might get a bit of a makeover, a friend with young kids said he looks like one of the turtles from Nemo. I'm not interested in a cease and desist letter from Disney, lol.)

That was intentional. The system prompt, the personality design, the knowledge base... all of it was built to make Tito feel like a friend, not a widget. (Which, if we're being honest, is the bar most chatbots fail to clear by a mile.)

[Screenshot of a Tito conversation would go here]

He's powered by Claude (Anthropic's API), and his knowledge base comes from a combination of my firsthand experience living in the area and structured research on current businesses and services. Right now, it's a curated JSON knowledge base. Down the road, it'll move to a pgvector setup in Supabase for proper RAG (retrieval-augmented generation), which is a fancy way of saying he'll get smarter and more accurate as the knowledge base grows (and not something I'd ever heard of before this project, haha).

What Tech Stack Powers an AI-Built Booking Site?

Next.js App Router for the frontend, Vercel for hosting, Claude API (Sonnet) for Tito's brain, and Supabase for the database, with a pgvector upgrade path as the knowledge base scales.

I'm not going deep on the technical build in this post because that's coming in Part 4. But here's the part worth knowing: the whole thing was built using Claude. Not just the chatbot, but the site itself. Claude Code handled the development, Claude generated the content, and I handled the direction and strategy that tied it all together. I also created the design system in Claude Design (I had the logo and colors from the guidebook I created).

More on all of that later.

Why Did the Planning Matter More Than the Build?

Because AI without direction produces generic output. The planning (PRD, voice guidelines, guest personas, knowledge base architecture) is what made the site feel real instead of templated.

Here's the thing most "I built this with AI" posts skip entirely: the planning (as always, zero judgment. You don't know what you don't know, but I promise you that when you spend more time up front on planning, you'll spend less time fixing when building).

Before a single line of code was written, there was a real planning process that included a product requirements document, a full project specification, voice and brand guidelines with two separate voices (one for the site copy and one for Tito), guest persona research, and a local knowledge base architecture with defined topic clusters and data sourcing strategy.

All of it was done with AI, and all of it was essential.

The reason this site doesn't look or feel AI-generated is that the planning was thorough enough that the AI had real direction to work from. Not "build me a vacation rental site." More like: here's the voice, here's the audience, here's what they're worried about (trust, because they don't know the area), here's the content strategy, here's the visual system, now build this specific page with these specific sections. The copy was also written with a list of "don't use terms" that would make it feel like a generic beach rental.

That planning process is Part 2 of this series, and it's the least sexy part and probably the most important one.

What's Coming

This is a five-part series. Here's the road map:

Part 2: The Planning. How the entire project was planned with AI before a single line of code was written, including the PRD, spec, voice skills, guest personas, and knowledge base architecture. This is the "most people skip this" post.

Part 3: The Design. The visual system, from color palette to typography, the decisions behind why it looks the way it looks, and how AI handled design direction.

Part 4: The Build. Where the code actually gets written. How Claude Code handled the development, how the pages came together, and how Tito got wired up and started talking.

Part 5: Going Live + The Marketing Plan. The domain gets pointed, the booking system gets connected, and the marketing engine turns on: SEO, AEO (answer engine optimization), content strategy, social automation, and AI video. The "now it actually has to work" post. 🤷‍♀️

Where We Are Right Now

The site is built, and Tito is live and talking. The local guide is up and gives people plenty to work with, but I'll continue adding to it with more research and content. And the whole thing, every page, every paragraph, every line of code, was built with AI.

But it's not done.

The domain isn't pointed yet, the booking calendar isn't connected, and we haven't started on the marketing engine. There are things that didn't go smoothly (there always are), and I'll get into those in the build post.

If you want to follow along as this AI-built booking site goes from a Vercel preview link to a functioning, revenue-generating, faceless vacation rental brand, stick around. This is just getting started.

What would YOU build if you knew AI could handle the execution? I'd love to hear... reply and tell me.