Here's why I created a travel website for robots
A.I. For You
SubscribeSign in
Here's why I created a travel website for robots<br>It’s not as nutty as it sounds. An announcement from Google spells out why
Alex Panetta<br>May 21, 2026
Share
The new homepage of my alternative-travel website, wanderingwell.site<br>I’m going to confess something that sounds unhinged and, in fact, would have sounded unhinged to me until very recently: I built a travel website for robots.<br>In my defense, the bots aren’t the end user. They’re merely the middlemen – digital pawns in service of an eminently human goal: better international tourism. So I’ll get to them in a second.<br>First I’ll explain how I wound up here.<br>I was fortunate this semester to take an AI-powered data analytics course with someone who literally oversaw AI-powered data analytics for the CIA. Thank you, Georgetown.<br>I used the course exercises to advance my own personal project, the alternative-travel app at wanderingwell.site. I explain its mission here.<br>It’s an answer to the explosive growth in international tourism making travel miserable – communities are begging for relief from selfie-snapping hordes; governments are slapping on new tourist taxes; and poll after poll after poll shows travelers getting fed up.<br>Overtourism is fueled by several factors – including online platforms.<br>Internet algorithms feed off popularity. They feast off the math of the attention economy: clicks beget more clicks, fame compounds fame, and unholy masses ultimately pile around the same Instagram shot.<br>That’s why I’m trying to build an anti-algorithm. A formula that does the opposite of rewarding clicks – it actually penalizes clicks. Instead of giving points for popularity, I take points away.<br>I’m collecting a database of beautiful places – less-famous beautiful places. And I’ve kept adding capabilities: images, weather, directions.<br>The latest update adds clustering tools. People can search by theme, not just by destination. So you can look for places in wine country, or with hiking, religious shrines, easy family getaways, foodie towns, or archeological sites.
With thematic clustering, you can search for 18 different types of holidays. Here are some results in the Spa & Wellness category<br>I built an image tool to rate hundreds of thousands of pictures
I’ve also added destinations – I’m up to 12,000 in over 230 countries and territories, and I’ll keep growing and refining the list as new data comes in.<br>It’s all powered by dozens of datapoints for each place.<br>Using the lessons from this past semester, I basically ripped through data for the entirety of Planet Earth. I started with about 5 million place names.<br>Places gained points for beauty and life – proximity to mountains, lakes, rivers, protected architecture, a UNESCO site, walkability, concentration of restaurants and coffee shops, and museums.<br>They also got points for being photogenic. I trained an image classifier to analyze pictures of over 200,000 places, granting scores to each picture, and including the score in my formula.<br>You’d be shocked, by the way, how easy this was. Anyone can do it with the latest LLMs. You basically build a website that lives locally on your desktop; you click through a couple hundred photos, you give each a score, type your reasoning into a box, and it feeds into a spreadsheet. You do this a couple of times, until your AI can predict the score you’ll give.<br>The model learned what I valued: natural features, old or colorful architecture, clustered together in highly walkable communities. The more of each, the higher the score.
Some of the 18 themes travelers can search for at wanderingwell.site
Subscribe
The ‘anti-algorithm’: How places lose points
And then I started removing points.<br>High hotel occupancy? You’re docked. High GDP versus the national average? You get docked. There are wealthy places on my list, but the goal was to steer visitors toward communities that could use more economic activity from travel.<br>Then comes the Wikipedia hammer. If the place has pages, where are the clicks coming from – from the local-language Wikipedia page, or the foreign-language page? I ran the percentages.<br>So if a small Tuscan town is getting a high ratio of clicks in English, it’s a good sign it’s being overrun in July and that hurts the score. The complete formula is here.<br>Now let’s talk about the robots. Or, to use the more accurate technical term, AI agents.<br>This site is an early experiment in agentic travel planning. By wiring its pages to AI research tools – and to something called an MCP server – I’ve made it possible for a personal AI agent to help someone actually research and book a trip.<br>The MCP connector is here. Just give it to a desktop AI app and you’ve got a basic travel agent.
Here’s how I trained an image classifier. I generated the tool in five minutes. Then rated photos. After a couple of hundred, the machine started accurately predicting my score for each picture.<br>Connecting it for...