By Visiting Media Editorial Team (featuring Jascha Kaykas-Wolff, CEO, Visiting Media)
Why AI tools skip right past virtual tours, how we turn visual spaces into buyer answers, and what we learned running a fleet of AI agents across Visiting Media.
Hotels have spent years investing in rich digital experiences so guests and event planners can explore their spaces online. The catch right now is that the AI tools people increasingly use to shortlist hotels and venues can’t interpret any of those visuals on their own.
Visiting Media CEO Jascha Kaykas-Wolff recently sat down with Alex Brooker, founder of Airside Labs and host of the Travel Tech Podcast, to talk through what that shift means for hospitality. Before joining Visiting Media, Jascha spent two decades building internet infrastructure and data platforms across tech and open source. On the show, he and Alex get into how we turn 3D and 360 property captures into something AI assistants can actually understand, what happened when we rolled out AI agents across our own team, and why owning your property data matters more than ever.
Watch the full episode on YouTube: “Can ChatGPT Find Your Hotel? Answer Engines, a Fleet of 17 Agents and One Rule Against AI Slop” (Hosted by Alex Brooker on the Travel Tech Podcast).
1. Making visual spaces readable for AI assistants
A standard king room for one night is pretty straightforward for an AI assistant to recommend. The problem shows up with the spaces that really differentiate a property and drive higher-value bookings: ballrooms, meeting configurations, suite layouts, and rooftop terraces. Most of that detail lives inside virtual tours, 3D walks, and PDFs. Because language models are built to read words rather than walk through a virtual tour, an AI tool putting together a venue shortlist will skip right past a property’s best visual assets.
To bridge that gap, our team built a system that translates visual captures into structured text across our library of property assets. First, computer-vision models look at the property’s 360 and 3D scans, recognize the physical details and dimensions of each space, and write clear, accurate descriptions of every ballroom, lobby, suite, and terrace.
From there, we go a step further. Traditional search is built around short keywords, while AI tools work like a conversation. People ask specific, open-ended questions. So we pair a hotel’s verified property data with AI buyer personas modeled after real hospitality customers, from weekend leisure travelers to corporate event planners. Those personas ask hundreds of practical questions about the space (banquet capacities, room-block setups, ceiling heights, or flooring materials for guests with allergies), and our system publishes the verified answers as structured FAQs that AI assistants can read directly.
“One of our alpha customers in the Bay Area went from not being discoverable in ChatGPT to a top-ranked property in the region on ChatGPT because we converted those visual assets into something the engines could actually read and understand. And the property team didn’t have to do any additional work.”
Jascha Kaykas-Wolff, CEO, Visiting Media (21:00)
2. Hotel teams are running a 24/7 business, so software vendors have to do the heavy lifting on AI
If you work at a software company and sit at a desk all day, keeping up with new AI models every couple of weeks is just part of the job. On the podcast, Jascha calls that constant tinkering “building calluses.”
Hotels don’t have that luxury. A property runs 24 hours a day across three shifts, with hundreds of people on their feet taking care of guests. People move between roles, properties, and brands all the time. A general manager or director of sales can’t ask their team to stop and relearn how a tool works every time an AI model gets updated.
“That level of operational complexity doesn’t afford the freedom to relearn a tool every couple of weeks. We have to make all of the changes taking place in the AI industry opaque to our customers, and run the self-tuning feedback loops on their behalf.”
Jascha Kaykas-Wolff, CEO, Visiting Media (24:36)
3. Virtual staging only works when it respects the real room
When a wedding buyer or corporate planner sends over an RFP, a hotel’s baseline virtual tour usually shows the room the way it looked at 6:00 a.m. between guest turnovers, sometimes on a grey morning. In the past, showing a buyer what that same space looks like set for an evening gala or a golden-hour wedding meant paying for physical staging or waiting weeks for custom CGI.
With AI Staging, a salesperson can restyle a real 360 capture in a few minutes, swapping layouts, decor, and lighting while keeping the actual walls, pillars, and dimensions of the room locked in place. Early on, we found that generic image generators love to ignore physics when you give them a panoramic room: they’ll put banquet chairs on top of fire pits, float guests over swimming pools, or warp the ceiling line. Training the system to respect the real geometry of the room lets a sales team show the buyer what’s possible without ever misrepresenting the space they’re actually buying.
4. Handing out chat logins didn’t work, so we built agents around department leaders instead
Two years ago, we tried what almost every other company tried: we bought ChatGPT and Claude logins for everyone at Visiting Media and told the team to go use them. Very little stuck in people’s day-to-day work.
So we stepped back and rebuilt our approach around a connected fleet of specialized agents:
- Pairing agents with department leaders: Instead of asking every employee to figure out prompts on their own, we paired dedicated agents directly with leaders across Sales, Customer Experience, People, and Operations. Each agent is wired into that team’s tools and data, and they share context with each other behind the scenes.
- Agents in engineering and product: On the product side, coding agents connected to our repositories help turn product specs into working code and review pull requests alongside our engineers.
- A visual dashboard for the whole fleet: Nobody wants to dig through raw text logs to see what an agent is doing, so we had the agents build a simple visual web page (we call it the “HTML Brain”) that shows what’s running, how the agents are talking to each other, and where anything needs attention.
5. One rule against AI slop, and why owning your data still wins
Once our teams started using agents every day, the volume of internal docs took off. Pretty soon, people were sending 30-page AI-generated pre-reads ahead of meetings.
To keep us honest, Jascha put one simple rule in place for the whole company:
“If you are using an agent to produce content and communicate on your behalf, the one rule is that you have to read the output before you send it to someone else. You have to have enough comprehension to have a live conversation about it. It isn’t a check for risk. It is a check for comprehension and brand experience.”
Jascha Kaykas-Wolff, CEO, Visiting Media (37:22)
At the end of the episode, Alex asked how hospitality and software companies protect their business when AI models keep getting faster, cheaper, and easier to copy. For Jascha, it comes back to owning your data. Software interfaces change and generic AI tools are easy to clone. The lasting advantage is owning and organizing the real, verified data about your physical spaces and your customers, rather than handing that asset over to someone else’s platform.
Episode timestamps and highlights
- 01:47 / From early internet infrastructure and open source to hospitality
- 05:59 / Why every business needs to own its customer and property data
- 10:14 / How Visiting Media captures, enriches, and distributes property visuals at scale
- 16:47 / Why AI assistants skip past virtual tours unless the visuals are translated into text
- 19:06 / Using AI buyer personas to anticipate real planner and guest questions
- 21:00 / How a Bay Area hotel went from invisible in ChatGPT to a top regional recommendation
- 24:36 / Why busy 24/7 hotel teams shouldn’t have to relearn AI tools every two weeks
- 29:48 / Why company-wide chat logins stalled out, and how pairing agents with team leaders worked
- 34:56 / The HTML Brain: making agent activity easy to see without reading code logs
- 37:22 / The one rule against AI slop: read it before you send it
- 42:28 / AI Staging, the golden-hour Hawaii wedding, and keeping the real room geometry locked
- 48:07 / Search keywords vs. AI conversations: answering buyer questions before they ask
- 52:41 / Where AI agents are headed over the next six months
- 58:13 / Why owning your verified property data matters long-term