The real problem with TikTok GO for hotels has nothing to do with TikTok

When TikTok launched GO last May 12th, the partner list already said a lot: Booking.com, Expedia, Trip.com, Viator, GetYourGuide, Tiqets. Six companies that have spent years investing in building machine-readable inventory pipelines. Airbnb wasn't there, and that absence sparked quite a bit of debate about platform strategies.

It's a legitimate conversation, but it risks distracting us from the truly uncomfortable one. Boštjan Koželj writes with clarity in an opinion piece on PhocusWire that gave me a lot to think about: before asking whether TikTok GO matters for their distribution strategy, a hotel should ask whether its inventory is structured in a way that these systems can actually read it. And for a large part of the industry, says Koželj, the answer is no.

The central point of the article is that Booking.com didn't land at the TikTok GO launch by accident. It had already begun testing hotel sales on TikTok in August 2025, months before the official launch, and that test wasn't primarily about conversion optimisation — it was about validating infrastructure: verifying that its inventory pipeline could feed a new channel at scale without having to rebuild anything from scratch. The principle underpinning Booking's entire architecture, according to Koželj, is that inventory must be readable by any system, through any interface, at any moment. It's not a technological advantage in the traditional sense; it's a strategic posture adopted years ago that keeps paying off with every new channel that opens up.

Airbnb, by contrast, has made a different and equally coherent bet: Brian Chesky publicly argues that conversational interfaces are architecturally wrong for travel — too text-heavy, incapable of direct manipulation, poorly suited to comparing thousands of options. Airbnb isn't absent from TikTok GO because it missed the train; it's building what it believes is a better interface, with Ahmad Al-Dahle — Meta's former head of generative AI — leading technology.

The historical comparison Koželj offers, and which I find the most useful part of the entire piece for anyone working in Italy, concerns France. When French regulators removed Booking.com's rate parity clauses in 2014, a natural shift towards direct booking was expected. It didn't happen. A 2022 European Commission study found that 79% of hotels still weren't differentiating prices between OTAs, and the competitive landscape had remained substantially unchanged from 2016, despite years of regulatory intervention. The lesson, Koželj writes, is not that direct booking is impossible — it's that price was never the real reason guests booked through OTAs. The reason was architecture, speed, familiarity, the frictionless experience Booking had built into its own flow from discovery to transaction.

Here I'll allow myself a personal reflection, because over the past year I've been working on this front concretely with my own hotel. I had a complete structured-data package built in JSON-LD for the Oasi website, with the explicit aim of making machine-readable what until recently was designed only for human eyes. Koželj draws a precise distinction between two technical levels that, read after having done this work first-hand, feel even more concrete to me. The first is JSON-LD (JavaScript Object Notation for Linked Data), the structured data format that search engines, AI assistants and platforms like TikTok GO use to understand what a hotel is, what it offers, how much it costs and whether it's available on a specific date. A hotel without correctly implemented JSON-LD isn't penalised in the next generation of travel search — it simply doesn't exist for those systems. The second level is the Model Context Protocol, or MCP, which goes further and allows AI agents to connect directly to inventory systems to retrieve availability and prices in real time, in some cases completing bookings without the guest ever visiting the website.

Here, though, I want to be fully honest, because not everything Koželj proposes carries the same urgency for a small Italian property. JSON-LD is work within reach of any independent hotel, even a small one, with a modest investment and a concrete return already today. MCP is an entirely different story: it requires a property management system with open APIs, technical expertise that small hotels rarely have in-house, and — frankly — for now the volume of bookings generated through these channels in Italy is still minimal (I'm saying this because it's something I activated at the start of the season). Koželj gives the example of HomeToGo, which already has an active MCP server with ChatGPT integration, but that's a player of a completely different scale from ours. The risk, if you don't set priorities in the right order, is chasing the new buzzword instead of fixing the foundations.

There is finally a third level Koželj discusses, and it's the one that touches me most closely in day-to-day management work: the hotel's internal knowledge — operational procedures, service protocols, property specifications — which in most properties exists as a collection of Word files, PDFs and institutional memory scattered across shared drives. Vector databases that index this documentation and make it queryable in real time by AI agents are, according to the author, the way any property can start making AI genuinely useful without depending on staff having to rummage through memory for how to handle a particular type of request. Here too I'll add a practical note: before thinking about a vector database, most properties I work with would already benefit enormously simply from writing down their procedures in one place — procedures that today exist only in the heads of whoever has been there the longest.

The figure that closes the article is the one that should give everyone the most pause: independent hotels' dependence on OTAs increased in 2025, with the booking share rising to 63.4% from 61.3% in 2024. The direction is unambiguous.

The lesson I take home from this piece, for anyone running an independent property in Italy, is not to be distracted by the name of the channel of the moment — whether it's TikTok, a chatbot or whatever platform appears six months from now. The right question isn't whether a specific channel will matter, but whether your infrastructure will be ready to be part of it when the volume arrives. JSON-LD can be done today, high priority. MCP and vector databases can be watched with interest, but without the anxiety of having to be first. The serious work — the kind that actually counts — gets done quietly before the traffic arrives, not after.

FAQ

Q: Why isn't TikTok GO the real problem for independent hotels?

A: Because the real issue isn't whether a hotel is present on a specific channel like TikTok GO, but whether its inventory is structured in a way that's readable by the AI systems that increasingly mediate between traveller and property.

Q: What is JSON-LD and why does a hotel need it?

A: It's the structured data format that search engines and AI assistants use to automatically understand what a hotel offers, at what price and with what availability. Without correctly implemented JSON-LD, a hotel risks being simply invisible to these systems.

Q: Is it worth investing in MCP right away for a small Italian hotel?

A: Not with the same urgency as JSON-LD. MCP requires complex technical infrastructure and, for now, generates minimal volume in Italy. It makes more sense to prioritise basic structured data and evaluate MCP further down the line.