The short answer
The OTAs are not being disintermediated by AI travel planners. They are buying and building their way into owning the AI booking layer instead. Expedia Group acquired Layla, the Berlin-based conversational AI trip planner, in July 2026 (Skift, July 31, 2026). Booking Holdings is running at least five consumer-facing AI experiments, with Priceline’s revamped agentic assistant Penny cited by executives as the standout early performer on satisfaction, engagement and conversion (Skift, September 14, 2026). OpenAI, after testing direct transactions, stepped back in March 2026 and refocused ChatGPT on discovery — routing live pricing and checkout through partner apps, starting with Expedia and Booking.com. And Google’s AI Mode hotel booking launched on August 27, 2026 with ten partners: five chains, and five OTAs.
The consequence for hotels, tour operators and DMCs is simple and uncomfortable: AI-assisted demand is defaulting to OTA inventory. Every AI-influenced stay that closes through that inventory pays 15–25% commission instead of landing in your direct channel. The counter-play has two fronts — be citable in AI discovery and be transactable direct. Skip either front and the OTA counterattack wins your bookings by default. If you want a structured read on where you currently stand on the first front, a travel AI audit is the fastest starting point.
The deal that deserved more attention
When a $20-billion-plus travel group buys an AI startup, the industry press normally reacts loudly. Expedia’s acquisition of Layla on July 31, 2026 got a few days of coverage and then disappeared into the earnings cycle. It shouldn’t have.
Layla is exactly the kind of product that was supposed to disintermediate the OTAs: a conversational AI that plans trips with live pricing and books them. Expedia bought it. Layla keeps operating, but its team now works alongside Expedia Group’s brands — Expedia.com, Hotels.com and the rest (Skift, July 31, 2026). The strategic logic is visible in Expedia’s own framing: having admitted that an all-in-one AI concierge “isn’t practical for now,” the group is focused on what Skift’s newsletter called “armies of specialized agents” — and it now owns a consumer-facing AI planner that feeds its checkout.
This is the move every hotel marketing team should study. The OTA response to AI was never going to be passive. It was going to be acquisition.
Booking’s five experiments and the Penny bet
Booking Holdings took the opposite route: build, test, iterate. Skift’s September 14, 2026 analysis counts five live consumer AI experiments across the group’s brands, and describes the Booking-versus-Expedia divergence as “a live natural experiment for the industry” — with Agoda’s turn rumored next.
The early signal matters more than the experiment count. Priceline’s Penny, the agentic assistant the brand rebuilt around conversational booking, is the standout performer — executives cite improved customer satisfaction, engagement and conversion even on a small early sample (Skift, September 14, 2026). Translation: when an OTA puts a booking-capable AI inside its own funnel, the funnel converts better. The OTAs aren’t hedging against AI. They are using AI to strengthen the exact middle layer — discovery-to-checkout inside one app — that direct-booking strategies have spent a decade trying to bypass.
Meanwhile the demand side keeps moving. Hospitality Technology’s forthcoming 2026 Guest Engagement Study (September 2026) finds that 17.1% of travelers now cite generative AI platforms — ChatGPT, Google AI and Copilot — as their primary discovery channel when selecting where to stay. And 37% of travelers already use AI models embedded inside online travel sites to plan and book trips (NYU Tisch Center and BCG, March 2026). Read those two together: travelers discover in AI, and a large share of them transact in AI that OTAs embedded in their own storefronts.
Why the AI layer defaults to OTA inventory
Three structural decisions in 2025–2026 shaped where AI-assisted transactions land, and none of them favored independent properties:
1. OpenAI retreated to discovery. In March 2026, Skift reported that OpenAI stepped back from handling bookings directly inside ChatGPT, refocusing the product on discovery and leaving the transaction to partner apps — a shift welcomed by investors in Expedia and Booking Holdings. Live prices in ChatGPT arrive through the Expedia and Booking.com connected apps, which as of September 2026 are not yet available in the EU or UK. When the AI layer doesn’t transact, it hands the checkout to whoever integrated first — and that is not your booking engine.
2. Google built the booking button with OTAs at the table. Google AI Mode’s hotel booking (launched August 27, 2026, US) lets travelers complete a stay inside the conversation via Google Pay. The ten launch partners: Booking.com, Expedia, Priceline, Hotels.com and Trip.com — plus Hilton, Marriott, IHG, Choice and Wyndham. The hotel or OTA remains merchant of record, but no independent hotel and no regional operator was in the first wave. We covered the details in what Google’s AI Mode booking means for independent hotels.
3. AI assistants prefer transactable inventory. An agent that can verify live availability, total price and cancellation terms through an API will use it. OTA APIs exist; most independent hotels’ websites do not expose one. We walked through the protocols behind this in agentic booking: when AI agents stop recommending and start reserving.
| AI surface (Sept 2026) | What the traveler sees | Who completes the transaction | What a property typically pays |
|---|---|---|---|
| ChatGPT | Recommendations; live prices via connected Expedia/Booking partner apps (US; EU/UK pending) | OTA partner app | 15–25% commission |
| Google AI Mode (US) | In-chat hotel list, room pick, Google Pay checkout | Hotel or OTA merchant of record — 10 launch partners only | Commission when OTA-routed; chain terms for the five chains |
| Gemini | Google Flights/Hotels data surfaced in answers | Google Hotels → OTA or direct links | Varies by routing |
| Booking.com app (group AI, incl. Penny on Priceline) | Conversational planning inside the OTA funnel | Booking Holdings brands | 15–25% commission |
| Expedia (Layla + specialized agents) | Conversational planning with live pricing | Expedia Group checkout | 15–25% commission |
The commission math nobody is running
Hotels track OTA commission on bookings they can see. Almost none have modeled OTA commission on AI-assisted bookings — the channel growing fastest.
Illustrative math for a 40-room independent hotel with a €180 ADR: if AI-assisted discovery sends 25 extra stays per month through OTAs instead of direct, at an average two-night stay and 18% commission, that is €9,000 in monthly booking value and €1,620 per month — €19,440 per year — in commission paid on demand AI generated. Scale the AI share toward the 17.1% of travelers already calling generative AI their primary discovery channel, and the annual figure grows with it. Tour operators and DMCs face a sharper version of the same math: experience marketplaces charge 20–30%, and as we documented in the tour marketplace blind spot, several keep listings invisible to AI crawlers while charging you for the placement anyway.
Commission was always the price of demand you couldn’t reach. AI-assisted commission is worse: you can reach this demand directly — it is asking about you by name in a chat window — and you still pay the toll if your direct channel isn’t the machine-readable, transactable option.
The counter-play: win two fronts, not one
Front one: be citable. AI assistants still need somewhere to learn that your property exists, what makes it distinct, and what a stay actually includes. That is a first-party content problem — structured facts, schema markup that machines can parse, fresh seasonal detail — and it is the foundation of AI visibility for hotels. Operators and DMCs should treat their own site as the canonical record of their product, not a brochure that redirects to a marketplace. A travel AI audit will tell you which priority prompts you already appear in — and which competitors own instead.
Front two: be transactable. Citable but unbookable means the agent hands the traveler to an OTA anyway. The practical checklist: a modern direct booking engine, machine-readable availability and total-price pages, clean cancellation terms, and — as the agentic protocols mature — a path for AI agents to book direct. The goal is not to beat Priceline’s Penny at conversation. It is to make your direct inventory the easiest verified option when an assistant compares options.
Then measure the right thing. Not “AI traffic” in aggregate, but AI-influenced direct bookings versus AI-assisted OTA bookings. That ratio is the single number that tells you whether the OTA counterattack is taking share from you or from your competitors — and whether a dedicated travel GEO agency engagement or an in-house program (see pricing) is the right next move.
FAQ
Are OTAs losing ground to AI travel planners? No — the 2026 evidence points the other way. Expedia acquired AI trip planner Layla in July 2026, and Booking Holdings is running five AI experiments with Priceline’s Penny delivering improved conversion on early samples (Skift, September 14, 2026). The OTAs are embedding AI inside their own funnels rather than losing the AI customer to standalone assistants.
Do AI travel planners book through OTAs or direct with hotels? Today, overwhelmingly through OTA rails. OpenAI refocused ChatGPT on discovery in March 2026 with transactions flowing to partner apps; Google AI Mode’s in-chat booking launched with five OTAs and five chains as partners on August 27, 2026. Direct, agent-bookable inventory for independent hotels exists only where properties expose machine-readable availability — which most still don’t.
How much commission do OTAs take on AI-assisted bookings? The same commission as any other OTA booking — typically 15–25% for hotels and 20–30% for experience marketplaces. What changes with AI is volume distribution: as 17.1% of travelers make generative AI their primary discovery channel (2026 Guest Engagement Study, September 2026), the share of demand paying that toll rises with it unless the direct channel captures it.
What should an independent hotel do first? Two moves, in order: fix citability (structured first-party facts, schema, freshness — start with a travel AI audit), then fix transactability (direct booking engine, machine-readable rates and availability). Visibility without a bookable direct path just warms travelers up for an OTA checkout.
Is this only a hotel problem? No. Tour operators and DMCs face a harsher variant: several major experience marketplaces both charge 20–30% commission and block AI crawlers from listings, so AI planners can’t see the inventory you pay to list. Your own site is the only machine-readable record of your product — make it canonical.
When do independent hotels get a direct AI booking button? No date exists. The agentic commerce protocols are open, and payment networks are building agent-to-merchant rails now, but the launch calendars so far favor chains and OTAs. Treat 2026–2027 as a preparation window: the properties that are citable and transactable when direct rails open will take the first wave.