Google updates and what they mean for your direct channel

1. AI is becoming a "new front door of travel", but it's still early days

ChatGPT and Claude are starting to send real traffic and real bookings to hotels. But everyone on the panel was careful to keep expectations grounded.

As Brad put it: "It's rapidly becoming a new front door of travel. I think the customers are still exploring… it's early days for the user, it's early days for the hotels too."

Sanjay agreed, and flagged just how volatile the landscape is right now: "Things keep changing, like day to day, sometimes hour to hour, and relatively modest changes that happen on the ChatGPT or the Claude side of the fence can have some really sort of astonishingly big effects for everybody downstream."

He also raised a note of caution about how some AI tools are behaving today, noting that they often surface hotels with "zero context, zero dates, zero additional information". He likened this to "looking at a paper map and circling where all the hotels are and handing it over to you and saying good luck from that point forward."

◆

2. There's already real revenue here, and a benchmark to measure it against

Brad shared a concrete way to gauge whether you're capturing your share. From the properties Agentic Hospitality works with (roadside Midwest and boutique properties in the US), he offered this rule of thumb:

"You should be seeing about a 1:1 ratio of social media ROI to AI surface ROI today. If you're not, you're missing an opportunity already."

The scale varies enormously by property type. As Brad noted: "Some of the brands see $2 to $3 million run rates already with AI surfaces. Boutique properties are seeing as little as $1,500 a month, but it's new and it will pay for itself to invest back into AI."

Meanwhile, tracking data from Hotelrank showed a 102% increase in direct and organic traffic from AI sources in a single recent week, a sign the channel is growing fast, even if absolute numbers remain modest for many properties.

◆

3. Google (not ChatGPT) is still the engine driving AI bookings

This was the central thesis of the session. As Charlie framed it: "Today… ultimately Google is the big driver of AI interactions that lead to bookings or lead to clicks."

The numbers back it up. Tracking data from Peec shows 86.7% of Google searches now return an AI Overview. As Charlie explained, it's not that people are choosing to use Gemini; it's that Google is serving AI answers to the enormous volume of people already searching.

Sanjay offered a fascinating insider perspective on why Google is iterating so aggressively: "My sense is that the way to get a promo at Google right now is to build something that gets into AI Overviews in an aggressive way." That incentive, Charlie pointed out, is exactly what's driving so much test-and-learn behavior and why things change so quickly.

◆

4. Most AI Overviews run on crawled data, not live data

Understanding how AI Overviews are built helps explain how to show up in them. Sanjay traced their lineage back to an old piece of Google technology: "The AI Overviews are basically an outgrowth of some technology Google had almost 15 years ago called web snippets."

Crucially, much of what appears there isn't live: "It is built on top of crawled data, and it is grabbing a snapshot from the internal web index… but it's not live. It is definitely relatively old information." On the "from" pricing that sometimes appears in AI Overviews, he was blunt: "I don't think it's analytically derived… something with no semantic content about a hotel detail page came across, picked a few words off it and popped up on the screen."

Brad added the strategic warning: Google holds the controls. "Google has control, it has the throttle, and what we've seen in these past few months is it can change the needle at any point to link back to suppliers." The encouraging trend is that Google is increasingly showing and citing the source of its content, broken out by paragraph and visible on rollover. This means hotels that provide the best content get the credit.

◆

5. Rich content beats checkbox data, and the OTAs are vulnerable here

One of the most interesting threads was Brad's "shiplap" story. Searching for a stay in Sanibel, Florida, he watched Google bypass the OTAs entirely and analyze the images on individual property sites: "It broke down the images… It found that there was newly installed shiplap in the unit. And that's why it recommended it, because… my wife likes shiplap rooms."

This is the crux of the opportunity for hotels. Google's natively multimodal models are reading images, video and social content instead of just structured fields. That plays to a hotel's natural advantage over OTAs. As Sanjay put it, "Booking.com says that this hotel has a pool is less compelling than the hotel's own website says that this hotel has a pool."

But there's an important caveat on room-level detail. Brad expects Google to "bring the rooms block in… and start showing accurate rooms, so I can prompt for the king room and… when I click book, go to the landing page for the king room." And right now, the OTAs are far ahead on structured room data. Sanjay broke the situation into three points: OTAs have a structured-data advantage built over many years; the nuance (can this room take a crib? is it hard to navigate?) is missing from OTA data; and the effort to add that nuanced detail yourself "is much easier to do now than it was three years ago."

The distinction is important: OTAs win on structured data (kitchenette: yes/no), but direct sites can win on the kind of nuanced, multimodal content that no OTA can replicate, like the shiplap, the view, and the feel of the place.

◆

6. Protect your foundations: SEO still beats GEO quick wins

With so many vendors selling "AI visibility" fixes, the panel's advice was to be skeptical of shortcuts. Brad's message was clear: prioritize effective SEO strategies over quick wins, over so-called AEO (Answer Engine optimization) or GEO (Generative Engine Optimization) fixes.

He urged hotels to protect what already pays the bills: "Focus on what drives the majority of your revenue today. For most businesses, 90, 95% oftentimes comes from Google, their direct website production." He pointed to how the big OTAs are playing it: "They build AI in other channels and marketplaces. They don't touch their moneymaker yet because they don't have to." His suggested low-risk experiment? Have employees book through an MCP (Model Context Protocol) to "try AI in your own infrastructure in a controlled way."

Sanjay reinforced the fundamentals: "The quickest way to lose rank in Google is to have those two things diverge," meaning what a human sees and what the crawler reads. His view on the much-hyped LLMs.txt file: "You shouldn't need an LLM.txt… the information that the AI needs should be trivially parsable off the page."

◆

7. How much does structured data still matter? (The experts disagree)

Not everything was settled, and the panel was open about it. This was the liveliest debate of the session.

Sanjay offered a deliberately provocative take: "I think structured data in some ways is yesterday's grapefruit," meaning it's become less relevant than it once was. His argument: LLMs are trained mostly on prose and aren't always great at parsing rigid structured data blobs. Write a clear paragraph, and the AI will understand it better than a JSON field.

Brad pushed back firmly: "We are using Schema.org's context to extend it into AI surfaces… it is still foundational to being discoverable inside of AI." He noted that structured markup in their MCP payloads actually produced lower token usage, meaning the AI could process the information more efficiently. He also pointed to RV Guha, the inventor of Schema.org, who is actively using structured markup in new "Natural Language Web" initiatives.

The middle ground both circled: training data is prose, but tool usage (like MCP) demands structure, and the right answer may well change again next week.

◆

The bottom line

Charlie pulled the threads together at the close. With Apple putting Gemini front and center for a huge base of iPhone users, paying attention to Gemini matters more than ever. And to be found, recommended and clicked in Google's AI Overviews, the recipe is consistent with everything above: have rich data on your website that gets cited so you earn organic links, run ads on Google surfaces, and make sure your live, accurate prices are showing through metasearch.

Do those things, and you're far more likely to show up and win the click in the AI-powered results where your guests are increasingly starting their journey.

If you'd like to talk about how Triptease can help you show up across Google's surfaces and capture more direct bookings, get in touch with our team.