The AI Booking Shift: What Token Economics, GEO and Agentic Workflows Mean for Your Direct Revenue

We have spent the last few years discussing the velocity gap in hotel commercial strategy. The 14.9-second mobile load time. The OTA margin erosion. The video-first booking journey. These are real, urgent, structurally significant problems – and they still need fixing. But there is a new layer sitting on top of all of it that most hoteliers are dangerously under-prepared for. It is the AI evolution challenge. And it does not simply compound the problems we already have; it rewires the entire commercial game.

If you are only thinking about AI as a chatbot on your website or a dynamic pricing tool in your RMS, you are at least two years behind the curve. The conversation has moved – fast.

We know the velocity gap. YouTube has overtaken Netflix in daily viewing time. The distance between inspiration and reservation intent has collapsed. But here is what has changed in the last twelve months: that collapse is no longer happening through a browser. It is increasingly happening through an AI agent.

As of early 2026, around 37% of travellers use large language models integrated into travel websites to plan and book their trips. That figure will not stay at 37%. Industry forecasts suggest that by 2028, more than half of all bookings will involve an AI agent at some point in the shopping journey. Once that shift becomes the majority, hotels will no longer compete solely for human attention. They will compete for AI attention – and that is a categorically different battle.

Here is the nuance that most commercial teams are missing: an AI agent does not browse. It does not experience load times the way a human does. But it does make decisions based on the data it can access cleanly, quickly, and confidently. If your rates are not structured for machine readability, if your availability feed is not AI-accessible via an open protocol like MCP (Model Context Protocol – think of it as a USB-C connector for AI systems), and if your content is fragmented across siloed databases with no clear signal hierarchy, then the AI agent will simply route the booking elsewhere. To the OTA that has already invested in exactly this infrastructure.

Do not underestimate what that means commercially. A neutral AI agent has no inherent incentive to route through a high-commission OTA – unless the direct channel is harder to access. Most hotel direct channels currently are. That is the opportunity and the threat sitting in the same breath.

This is where the conversation gets genuinely technical – and genuinely important. I want to spend time here because it is almost entirely absent from mainstream hospitality discourse.

When an AI agent processes a travel query and retrieves information to generate a recommendation, it consumes what are called tokens – units of text processed by the model. Every word in, every word out, every tool call, every retry, every cross-reference costs tokens. And tokens cost money.

Here is the counter-intuitive reality of 2026: token prices have fallen dramatically – down nearly 99.7% from 2022 peak levels by some measures – yet enterprise AI bills are rising sharply. Why? Because lower unit costs have encouraged vastly more complex, multi-step workflows. Agentic loops, retrieval augmentation, tool calls, orchestration overhead, vector databases, data egress, observability layers – these hidden costs sit outside the model invoice and are largely invisible to the commercial teams deploying them. The result is that many hotels building or licensing AI tools are looking at a visible model cost that looks manageable, and an invisible total cost of AI ownership that is significantly higher.

What does this mean for hotel distribution? Two things.

First, OTAs have a structural token-cost advantage. Booking Holdings is reinvesting $700 million in AI infrastructure in 2026 alone. Their agentic tools already cut average cost per booking by approximately 10% in 2025 by reducing handling time and contact rates. When an AI agent queries Booking.com, it gets a clean, structured, optimised API response – a low-token, high-confidence interaction. When it queries a mid-market hotel’s direct website, it often gets a slow, unstructured, JavaScript-heavy page that requires far more processing to extract the same information. That processing overhead makes the direct option less efficient for the AI. Less efficient means less likely to be recommended.

Second, the economics of AI-powered tools on the hotel side are often misunderstood. Revenue managers and commercial directors are being sold AI platforms with headline pricing that does not reflect the true total cost. The actual spend includes orchestration costs, integration with PMS and CRS systems, vector database hosting for conversational memory, retry loops when the model is uncertain, and engineering time for maintenance. Teams that benchmark only the model invoice are systematically underestimating their AI investment – and therefore underestimating the business case required to justify it.

Understanding token economics is not a technology team problem. It is a commercial strategy problem. Budget holders need to ask harder questions.

Search engine optimisation took years to become a hotel boardroom priority. Generative Engine Optimisation – GEO – has perhaps eighteen months before it becomes non-negotiable. And most hotels have not started.

OTAs currently account for over 50% of AI citations when a traveller asks an LLM to recommend a hotel. That is not a coincidence; it is the direct consequence of years of structured data investment, content standardisation, and API availability that OTAs have made while hotels were focused elsewhere. An AI citation that does not include a brand mention occurs 61.7% of the time – meaning the AI recommends a property type or a location experience, but routes the conversion through an intermediary. That is the billboard effect, reimagined for the LLM era.

The levers for GEO are not entirely different from what good digital commercial practice already demands: accurate, structured, consistent data across every platform the LLM crawls; current property descriptions with verified amenities; schema markup that signals clearly to AI systems what your product is and what it offers; and, critically, live rate and availability feeds accessible via open protocols. Without these, the AI describes you incorrectly, routes guests to OTAs, or shows rates from eighteen months ago. Any one of those outcomes is a direct booking lost.

The transition from a legacy siloed model to an Integrated Commercial Engine was already the central thesis of where hotel commercial strategy needed to go. The AI evolution does not change that thesis – it makes it more urgent and adds a fifth dimension.

Sales, Marketing, Distribution, and Revenue Management must be fused. We have established that. But in a world where AI agents are inside the booking journey, there is now a fifth workstream that must sit alongside the other four: AI Connectivity and Content Infrastructure. This is not a technology project. It is a commercial strategy priority, because it directly determines whether your property is visible, bookable, and preferred by the systems that an increasing number of travellers are delegating their decisions to.

What does this look like in practice? It means your commercial team owns the question of machine readability as actively as it owns rate strategy. It means GEO sits on the same agenda as SEO. It means your booking engine vendor is assessed not only on conversion UX for humans, but on its API architecture for AI agents. And it means your content – property descriptions, amenity data, rate structures, package terms – is treated as live commercial infrastructure rather than a marketing department filing exercise.

The legacy model operated on the assumption of a human decision-maker moving through a predictable, structured funnel. The 2026 reality is a hybrid funnel where AI systems filter, recommend, and increasingly execute at speed. Your infrastructure needs to serve both audiences simultaneously.

The velocity gap, the technical friction, the video-first discovery shift, the OTA margin war, and now the AI evolution – these are not five separate problems. They are five symptoms of a single structural misalignment between legacy hotel infrastructure and the speed of the modern booking environment. The action plan addresses them as one.

1. Fix the foundation before anything else. Your 14.9-second mobile load time is not a web team problem – it is a commercial strategy failure. No AI strategy, video strategy, or direct booking strategy delivers ROI on a broken foundation. Core Web Vitals, mobile-first architecture, and sub-3-second load times are the baseline from which every other initiative should be measured.
2. Conduct an AI visibility audit. Run your own property name through ChatGPT, Claude, and Gemini. What does the AI say about you? Is the amenity data accurate? Does it route to your direct booking or to an OTA? Is the rate information current? This takes thirty minutes and will tell you more about your GEO exposure than any vendor pitch.
3. Assess your machine readability. Does your booking engine expose a clean, structured API that an AI agent can query for live rates and availability? Is your property data consistent across all platforms the LLMs index? Have you explored MCP-compatible connectivity with your PMS or channel manager? If the answer to any of these is “I don’t know,” that is where the commercial conversation needs to start.
4. Pressure-test your AI tool investment. If you are already using AI-powered tools – chatbots, revenue management assistants, guest messaging platforms – request a full cost breakdown that goes beyond the model licence fee. Ask specifically about orchestration costs, integration maintenance, retry logic, and data infrastructure. If your vendor cannot provide that breakdown, you are likely underestimating total cost of ownership by a material margin.
5. Deploy the OTA counter-strategy with structural discipline. The break-even math is unchanged: shifting just five bookings per week from OTA to direct generates approximately £5,500 in additional annual margin per room. Price-match widgets, exclusive direct perks, and opt-out from OTA-funded discount programmes like Booking Genius Level 2 and 3 remain the tactical priority. The AI layer adds one more reason to win this battle: the more direct bookings you drive, the more first-party data you accumulate – and first-party data is the fuel for every AI tool you will deploy in the next three years.
6. Build your video presence for AI, not just for humans. LLMs are increasingly using YouTube and Reddit as citation sources. Video content that describes your property with specificity – real guest experiences, authentic narratives, verifiable detail – is not just social media strategy. It is GEO strategy. The content that AI crawls and cites is the content that ends up in the recommendation that a traveller receives. Invest accordingly.
7. Move to real-time budget allocation. Fixed annual and quarterly marketing budgets are structurally incompatible with an AI-driven commercial environment. LLMs and AI agents do not wait for Q3 approval cycles. If your brand is not visible and optimised at the moment an AI agent makes a recommendation, you lose that booking – permanently, not temporarily. Fluid, real-time allocation is not a finance team aspiration; it is a commercial survival mechanism.

The parallels to the early internet era are instructive. Hotels that dismissed online booking in 2000 spent the next decade paying for it through OTA dependency. The AI adoption curve is moving significantly faster. The hotels investing in AI connectivity now – in GEO, in machine-readable infrastructure, in structured data, in MCP-compatible booking architecture – are building the direct booking advantage of 2028 and beyond.

The good news is that a neutral AI agent, properly informed, has no incentive to favour an OTA over a well-connected direct channel. The commission economics are clear. The margin maths are compelling. The technology standards are open. What is required is commercial leadership willing to treat AI connectivity as a revenue priority – not an IT project, not a future initiative, not something to revisit at next year’s planning cycle.

The velocity gap was always about infrastructure failing to keep pace with the speed of the traveller’s decision. The AI evolution has simply made that failure more visible, more immediate, and more commercially consequential. Fix the foundation. Build the connections. Understand the economics. The engine is ready. The question is whether your commercial strategy is.

Main Sources:

AI in the booking journey / agentic AI

https://www.altexsoft.com/blog/agentic-ai-vs-ota-hospitality/

https://www.siteminder.com/r/agentic-ai-for-hotels/

Token economics

https://www.navyaai.com/reports/ai-cost-report-token-prices-vs-ai-bill

OTA AI investment / Booking Holdings

https://finance.yahoo.com/news/booking-cfo-touts-9-9-233553480.html

GEO / OTA dominance in AI citations

https://www.click2convert.com/insights/do-otas-influence-ai-seo-llm-hotel-bookings/

MCP / AI connectivity for hotels

https://www.hospitalitynet.org/explainer/4130211/ai-terms-for-hoteliers-mcp-geo-agentic-ai-and-llms-explained

https://www.mylighthouse.com/resources/blog/ai-terms-for-hoteliers

AI adoption and ROI data https://otelciro.com/en/news/hotel-ai-adoption-82-percent-2026 https://hoteltechreport.com/news/ai-in-hospitality