The short answer
AI travel planners are structurally biased toward OTAs, and loyalty is the one asset they cannot price into the comparison. When US travelers let AI help book a hotel in 2026, only 38% of those transactions went to the hotel’s own channel — 44% landed on an OTA and 8% were completed inside the AI tool itself, a share that will grow as Google rolls out agentic booking (NPS Prism/Dynata survey of 2,287 US AI travelers, published August 14, 2026; Skift, August 7, 2026). Four in ten of those bookers said AI changed which hotel they chose — it did not just narrow the search, it altered the outcome.
The mechanics are simple and unforgiving: AI builds its shortlist of two or three hotels from what it can observe from the outside — price and availability. That is exactly the surface OTAs are built to present. A loyalty program pays off only inside the brand’s own channel, long after the shortlist is made. So the recommendation happens in AI, the transaction happens on an OTA, and only the stay happens with the hotel.
The strategic consequence is the inverse of the panic narrative: loyalty and first-party guest data are not obsolete in the AI era — they are the moat. The chains that make an AI agent come to them for stay history, preferences and member pricing keep the agent dependent on the brand. The hotels that let that data leak onto platforms get compared on price forever. This article lays out the 2026 evidence, what Marriott and IHG are already doing, and five moves independent hotels, tour operators and DMCs can run this quarter. If you are new to how assistants pick properties at all, start with AI visibility for hotels — what follows assumes you know the recommendation layer and are now defending margin.
Where AI-assisted hotel bookings actually land
The NPS Prism/Dynata study (n = 2,287 US AI travelers, 1,846 converters, 332 hotel-intent converters; published August 14, 2026) is the clearest public dataset yet on the question hotel revenue teams keep asking: when a guest uses AI, who gets the booking?
| AI-assisted booking destination | Share | What the hotel loses |
|---|---|---|
| Hotel’s direct channel | 38% | Nothing — guest relationship, data and margin intact |
| Online travel agency | 44% | 15–25% commission, guest email, preferences, upsell path |
| Inside the AI tool itself | 8% | Commission and the merchandising steps that carry margin |
| Other / not stated | ~10% | Mixed |
Two adjacent findings make the table worse before it gets better:
- The OTA share is won on visibility, not preference. AI shortlists two or three options built from price and availability — the two dimensions OTAs are engineered to be compared on (NPS Prism, August 14, 2026). Nobody chose Booking.com; the machine just reads it best.
- Even direct-path bookings arrive thinner. AI compresses search-to-reservation and skips the merchandising steps in between. 23% of travelers who booked hotels through AI reported they felt they missed out on extras and add-ons — the upgrades, packages and spa slots where hotel margin actually lives (NPS Prism, August 14, 2026).
Meanwhile, trust in the machine is real and compounding: three-quarters of AI users said the results met their expectations, and the more they use it, the more they trust it (NPS Prism, August 14, 2026). A guest who trusts AI to pick a good hotel has, as the researchers put it, little reason to stay loyal to any single brand — unless the brand gives the AI a reason it can read, or gives the guest a reason AI cannot see.
Why AI is structurally OTA-biased (and always will be)
This is not a bug in ChatGPT or Gemini. It is an information asymmetry:
- OTAs publish machine-readable everything. Rates, availability, room types, cancellation terms — normalized across a million properties. An AI assistant comparing “boutique hotel in Lisbon under €200, free cancellation” is reading OTA data structures natively.
- Most hotel websites still publish marketing. A brand story, a gallery, a booking widget the crawler cannot operate. The assistant cannot query your member rate, because members-only pricing lives behind a login it will never have.
- Loyalty value is invisible by design. Points, tier upgrades, late checkout for elites — none of it is externally comparable. It pays off after channel choice, which is precisely why it cannot influence an AI shortlist made before channel choice.
The result: a decade of direct-booking investment (rate parity work, booking-engine UX, “book direct” campaigns) is being routed around by assistants that compare the comparable. We documented the platform side of this in The OTA AI Counterattack; the loyalty angle is the defensive half of the same fight.
And the clock is speeding up: Google confirmed in August 2026 that agentic hotel booking — completing the reservation inside the AI platform — is now in testing (Skift, August 7, 2026). Every point of share that moves from “OTA via AI” to “AI tool itself” is another step away from the enrolled member and their first-party data.
What the big chains already know: loyalty is the agent-proof asset
Marriott is treating AI as a way to pull travelers deeper into Bonvoy rather than a threat to it, rolling out natural-language search across its own site and app — moving the conversational experience inside the channel where loyalty lives (NPS Prism, August 14, 2026). IHG is going further on data: CEO Elie Maalouf confirmed on the Q4 2025 earnings call that IHG is building one unified loyalty and CRM platform on Salesforce, with an AI-powered version coming in 2026 — a single first-party data spine designed to anticipate what a guest wants next rather than react to what they did (Hospitality Net, July 6, 2026).
Read the strategy underneath both moves: make the loyalty dataset the thing an AI agent has to request. Stay history, stated preferences, tier status and its attached benefits are data an agent can use only by coming through the brand’s own channel. Fund the parts of your program the agent cannot route around — not the parts it can find anywhere.
Five moves for independent hotels and small chains
You do not need Bonvoy scale to run the same logic. You need a defensible dataset and an AI-readable surface.
1. Fix the machine-readable layer first. If your rates, inventory and content are not clean and structured, you are cut from the AI shortlist before loyalty ever gets tested — a guest cannot be loyal to a hotel the assistant never names. Structured data, consistent entity facts and quotable answers are the entry ticket; a travel AI audit tells you exactly where you stand on the prompts that matter in your market.
2. Publish the member-rate delta, not just the existence of member rates. AI cannot see behind your login, but it can read a page that says “members save 10% and get late checkout — compare rates.” Make the loyalty advantage externally legible in plain, crawlable text, and the shortlist starts pricing it in.
3. Turn stays into structured preference data. Every direct checkout should capture three fields: room preference, stated dislikes, occasion. That is the dataset an agent — or your own AI features — needs to personalize, and it exists only because the booking came direct. This is the first-party moat in its smallest viable form.
4. Attack the add-on gap deliberately. Since 23% of AI bookers feel they missed extras (NPS Prism, August 14, 2026), pre-arrival upsell flows are no longer a nicety — they are margin recovery on AI-routed bookings. Trigger them at confirmation, not at check-in.
5. Reinvest the commission you save, measurably. Every point of AI-era share you defend direct saves 15–25% of channel cost on that booking. Route part of it into the visibility work itself — the audits, content and monitoring that keep you in the shortlist — and measure AI-driven bookings honestly, as we laid out in Measuring AI-Driven Direct Bookings.
Tour operators and DMCs: same moat, different shape
If you run experiences or destination programs, “loyalty” means repeat-client and trade relationships: operator preferences, group history, supplier terms. The AI-bias problem is identical — assistants compare what OTAs and marketplaces make comparable, and your margin lives in everything they cannot see. The defense is also identical: an AI-readable commercial surface (structured itineraries, clear inclusions, quotable availability) plus a first-party client dataset no marketplace can aggregate. If you want that handled as a system rather than a side project, that is literally what a travel GEO agency engagement covers, scoped against pricing that maps to direct-booking outcomes, not traffic.
FAQ
Do AI travel planners recommend hotels with loyalty programs? They can, but loyalty status is invisible to them — AI shortlists are built on price and availability, typically two or three options, which favors OTAs (NPS Prism, August 14, 2026). Loyalty influences the booking only if its value is published in machine-readable form on the hotel’s own pages.
What share of AI-assisted hotel bookings goes direct? 38% direct, 44% to OTAs, and 8% completed inside the AI tool itself (NPS Prism/Dynata, n = 2,287 US AI travelers, August 14, 2026). The in-tool share is expected to grow as Google’s agentic booking testing matures (Skift, August 7, 2026).
Is a hotel loyalty program still worth building in 2026? Yes — arguably more than at any point since OTAs began. Loyalty and its first-party data are the assets AI cannot compare away; Marriott and IHG are re-platforming their programs specifically for the AI era (NPS Prism, August 14, 2026; Hospitality Net, July 6, 2026). Independents can run the same logic with lightweight preference capture and member-rate transparency.
How do hotels get into AI recommendations at all? Make rates, inventory and content clean, structured and machine-readable, keep entity facts consistent everywhere the assistants read, and publish quotable direct answers. That recommendation-readiness layer is what a travel AI audit measures first.
Why do AI bookings carry less margin per stay? AI compresses the path from search to reservation and skips merchandising steps; 23% of AI hotel bookers said they felt they missed out on extras and add-ons (NPS Prism, August 14, 2026). Pre-arrival offers are the recovery lever.
The question for 2026 is no longer whether AI decides which hotels get booked — four in ten AI bookers say it already did. The question is whether, when it decides, it can see anything about you that an OTA cannot.