Can ChatGPT Find Your Hotel?

From the Travel Tech Podcast newsletter for episode 39, Can ChatGPT Find Your Hotel? Answer Engines, a Fleet of 17 Agents and One Rule Against AI Slop.

A hotel can hold billions of pixels of its own ballroom and still be invisible to ChatGPT. This week's guest says his team took one Bay Area property from undiscoverable to first in its region, and he can explain exactly how. Specifics like that are rare in an AI pitch, which is why I wanted him on the show for our 30,000+ readers across aviation, travel tech, and AI.

He spent more than 20 years in the infrastructure of the internet, from Yahoo in the late 90s to helping build the Microsoft Store, then ran marketing at BitTorrent and Mozilla, where the work centered on who controls information online. A meal with a hospitality founder pulled him into the industry two and a half years ago, and he now runs the company that founder built. Apple Podcasts, also streaming on Spotify and YouTube.

Hotels hold their richest data in a format answer engines can't read, and they don't have the staff to keep up with engines that change every few weeks.

The gap between a 360 tour and a ChatGPT answer

That tension has a structural cause. A midsize property can run hundreds of staff across three shifts under one GM, in an industry where people move between properties, so nobody is positioned to relearn a toolset every couple of weeks. The asset hotels paid to create is just as constrained: capturing 5K video and LiDAR can mean shooting at 6 a.m., because a full resort can't take rooms offline, and the result sits in a form LLMs don't read.

AI raises the stakes because planners increasingly start in ChatGPT or Claude, and they ask in full sentences rather than keywords. Visiting Media converts its visual data into descriptions, has synthetic buyer personas ask hundreds of questions of it, and uses a council of agents to answer them, which my guest says took that Bay Area alpha customer to number one on ChatGPT. He is far more cautious about customer facing automation, and his company's rule is that a person reads agent output before it goes to anyone else.

The other half of the conversation stayed with me just as much. Jascha runs a fleet of about 17 agents across a 140 person company, and the lesson he took from it is that handing everyone a chat license didn't change how anyone worked. Agents built for each department head did, along with one rule he introduced when agent output turned into a flood of AI slop: read it before you send it.

Jascha Kaykas-Wolff, CEO of Visiting Media, joined me to talk through the pipeline from pixels to answers, the difference between SEO and AEO, how his agent fleet is built, AI staging for RFP responses, and why he sees data as the moat.

Key Takeaways

• Visual data needs a text layer: Answer engines read written content rather than immersive assets, so a 360 tour only aids discovery once it becomes accurate descriptions and FAQs.
• AEO rewards predicted questions: SEO leans on structured data and authority, while answer engines respond to content that anticipates conversational, multi constraint queries.
• Ground truth constrains generative output: AI staging had to be taught that chairs can't sit in the water or on a fire pit. My take: every regulated deployment has to solve the same ground truth problem.
• Give agents to departments, not seats to people: Company wide chat licenses didn't change how Visiting Media worked. Agents built for each department head, with their own skills but shared data and the ability to talk to each other, did, and those agents are now being rolled out to individuals.
• Make the agents' work visible: Jascha isn't an engineer, so he had his fleet turn its Markdown files into an HTML "brain" that shows fleet health, the inbox the agents use to talk to each other, and any loops that have broken. For non technical leaders, he calls it the biggest unlock.
• One rule for the slop: Once agents were in daily use, 30 page briefing documents full of Claudeisms flooded the company. The fix is one rule: read the output before you send it, well enough to discuss it. It's a check for comprehension, not risk.

Alex Brooker, Host of the Travel Tech Podcast, Founder of Airside Labs