What happened

On 8 September 2026, Meta, the company behind Facebook, launched an AI agent called Muse in the US. An AI agent is an AI assistant that goes online to look things up, compare prices and place orders on your behalf. You tell Muse what to book, and it opens a browser, fills in the forms and books your flights and hotels, coming back for your approval only before it pays. Muse is available only in the US for now and can’t be used in Taiwan.

Muse books things in two ways. For flights, it books through a company that connects directly into airline systems, without going through a booking site’s comparison pages. Hotels have no such direct line. When a travel-trade publication tested it, Muse opened booking sites such as Expedia and Hotels.com itself and searched and compared the way a person would. Where to book was its own call.

On 22 September, Expedia said it would plug into Muse. According to travel-trade press, US users will be able to book Expedia hotels directly inside Muse. The booking is still handled by Expedia; only the payment step happens inside Muse. No launch date was given. In other words, for hotels, Muse used to go and browse the sites itself; now Expedia is going to plug itself straight in.

The share prices moved first. From a few days before Muse launched to 23 September, shares in Airbnb, Expedia and the parent company of Booking.com, three US booking platforms, each fell by more than a tenth, one of them by nearly a fifth. The day after Expedia’s announcement, all three fell again, while Meta’s own shares went up. Travel-trade press put it down to investor fears that Muse will disrupt the booking platforms’ business, and analysts have estimated that agents could take a bite out of revenue at travel, ride-hailing and food delivery companies. These are the market’s reading and analysts’ estimates. There are no public figures yet on how many bookings agents have actually taken.

Why it matters to you

Booking sites aren’t going away. But customers who use an agent don’t have to open them and choose for themselves; the agent goes in and chooses for them. In effect, the platform has been moved one step back, from in front of the customer to behind the agent.

Until now, someone booking a room opened a booking site and picked from the list the platform had sorted. Where a hotel landed on that list came down to reviews, rates, promotions and whatever visibility it paid for on the platform. Muse hands the ranking to someone else. It browses the platform’s list, but what the customer ends up seeing is the handful it has already picked. Once Expedia is wired in, the customer doesn’t even open Expedia’s site. Expedia becomes something like a shelf behind the agent: the booking is still Expedia’s, but the agent decides which places the customer sees first. How Muse chooses, Meta’s announcement doesn’t say.

An analyst at an investment bank is upbeat about the platforms. In his view, the platforms hold the rooms, run their own loyalty programmes, and hotels are willing to pay for discounts on them, so in this new agent channel they won’t lose out to anyone. That doesn’t contradict the picture above. The shelf will stay; it just no longer decides how the things on it are ordered. Put another way, the ranking you pay for and work for on a platform shapes how customers choose on that platform. It has no say over how an agent chooses.

A lot of businesses in Taiwan get their customers through platforms: guesthouses and hotels through booking sites, restaurants through delivery apps, many brands through e-commerce marketplaces. Extending what has happened in the US over the past few weeks, our inference is that as agents spread to these industries, part of the decision about which platform customers are sent to, and which business they end up with, will move to the agent. Say you run a dessert shop and have climbed into the top few nearby on a delivery app on the strength of your reviews and offers. If customers one day ask an AI to pick their afternoon treat instead, the shops it suggests won’t necessarily follow the delivery app’s ranking.

What to do about it

Start preparing now, and what you’re preparing is making sure AI can find accurate information about you online. Muse can’t be used in Taiwan yet, but the skills you’ve built on platforms, how to push reviews and when to grab a promotion slot, carry little weight once an agent is doing the choosing.

And the way agents book things for customers keeps changing. When Muse launched, it was still booking hotels by browsing websites on its own; just two weeks later, Expedia announced it would plug in directly. Every time that changes, where your customers come from changes with it. So to know where customers end up, you have to look at three things together: what the platforms have changed, which places the agents pick for customers, and whom ChatGPT and similar AI recommend when customers ask them directly. And every time the way agents book changes, you have to look again.

Our judgement is that however agents end up booking, before an AI recommends you it has to have read enough accurate information about you online, such as what you sell, where you are and what other people say about you, and that information is spread across your own website and other sites. We have a service for exactly this, called managed GEO. GEO, put simply, is making sure AI can make sense of you and is willing to mention you in its answers. With managed GEO, our team fixes up your website content, makes what other people have written about you online consistent, and regularly puts your customers’ questions to the major AI engines to see whether they mention you, whether they recommend you and whether they get you right. All you do is sign off on the results.

When a customer asks AI where to buy or where to stay, does it say your name? Book a demo: one of our consultants sets aside time with you and shows you live, playing your customer and putting their kind of question to ChatGPT and Gemini, so you can see on screen whether AI mentions you and where it sends people. You’ll also get your site’s GEO audit score: geoweb.tw’s free website audit, which checks whether AI can make sense of your site and scores it across 12 areas.

Further reading