1. Two Numbers That Don't Tell the Same Story
As of early 2026, roughly 37% of travelers report using a conversational assistant β a large language model built into a travel website or app β to plan a trip: exploring a destination, comparing neighbourhoods, getting a feel for budget, ruling out options that don't fit. That's a high number, still climbing, and it no longer surprises anyone who has watched the sector for the past two years.
The second number is a lot less dramatic, and that's exactly why it deserves attention. Asked whether they would let an AI agent complete a travel purchase on their behalf without a human double-checking the transaction, only about 2% of travelers say yes, according to one recent industry data point. A separate survey from Expedia Group puts a similar figure at around 8%. Methodologies differ, samples differ, but the order of magnitude doesn't move: somewhere between two and eight travelers out of a hundred are willing to hand the final click β the one that commits a credit card β to an AI agent without supervision.
Placed side by side, these two numbers describe a drop-off of roughly five to eighteen times between AI usage at the research stage and AI trust at the payment stage. That gap, more than either number on its own, is the useful signal here for a hotelier in 2026.
2. Why the Gap Is So Wide
The explanation isn't complicated once you state it plainly: asking for advice and committing money are two different psychological acts, and AI doesn't hold the same trust status in each.
Using an AI assistant to explore destinations or compare hotels feels like asking a well-informed friend, or consulting a particularly responsive travel guide. The downside of a bad suggestion is minor and reversible β if it doesn't fit, you keep looking. Perceived risk is low, which explains why adoption at this stage has moved so fast: it barely requires a behaviour change, just a faster research channel.
Paying, on the other hand, commits a credit card, a named reservation, fixed dates, sometimes strict cancellation terms. A mistake at this stage β wrong date, wrong room type, double charge, currency error β costs real money and real time to fix. It is not surprising that travelers want, at this specific moment, to look at the screen themselves one more time rather than delegate the act entirely to an automated agent, however capable it might be.
3. What This Says to Those Who Fear Agentic AI
Part of the hotel industry has spent months worried about a specific scenario: autonomous AI agents that would research, compare, and book stays end to end, without a human ever visiting the hotel's own website, seeing its brand, or clicking its own booking button. In that scenario, the hotel becomes an interchangeable inventory feed, invisible behind the agent's interface.
Current data says that scenario is not happening at scale β not yet, and not at the pace some feared. We covered on this blog ChatGPT pulling its integrated payment for travel bookings, a strong signal that even the players best positioned to push full automation stepped back from the same reality: travel is a purchase category where trust in fully autonomous checkout remains particularly low, well below other e-commerce verticals where agentic purchasing is moving faster. Paying for a trip involves too many sensitive variables β dates, occupants, cancellations, disputes β for the majority of travelers to hand it to an unsupervised agent, today or in the near future.
For the hotelier worried about losing control of the guest relationship to an invisible AI agent, these numbers are reassuring without being an excuse for inaction. The human channel β the final click, the card, the confirmation β remains firmly in the traveler's own hands in the overwhelming majority of cases. What has changed isn't who presses the pay button. It's who shapes the decision before that button ever appears.
4. What This Says to Those Ignoring AI
The other half of the industry makes the opposite mistake: taking comfort in the idea that since almost nobody actually books through AI, there's no point investing in AI visibility. That reasoning is looking at the wrong number.
The 37% of travelers already using AI to plan a trip is not a marginal footnote β it's already more than a third of the market, and the trend is climbing, not flat. That research phase is where a hotel gets presented, compared, recommended, or simply left out. If an AI system never mentions your property when a traveler asks for a hotel in your area with a specific feature, you lose that share of demand before the payment question ever comes up β whether that payment eventually happens through AI or, as is overwhelmingly the case today, directly on your own website by a human.
In other words: the fact that AI almost never completes the transaction does not mean AI has no effect on the transaction. It shapes the shortlist, the comparison, the first impression β and that influence converts into bookings, even when the final click is human and happens somewhere else entirely.
5. The Strategy These Two Numbers Point To
This data points to a fairly clear allocation of effort, and it matches what we've recommended throughout this blog: invest heavily in the AI-assisted research phase, and keep a fast, trustworthy booking flow on your own website ready for the moment a human actually clicks to pay.
Concretely, that means continuing to take seriously the fundamentals already covered here: complete and accurate Hotel schema, a robots.txt file that explicitly allows AI crawlers to access your content, an llms.txt that clearly signals how your content should be used, clean and unambiguous content structure that AI systems can extract without error. That is exactly where the battle is being fought right now, and it is a winnable battle through concrete technical work rather than speculative bets.
At the same time, it means not easing off the booking flow itself: page speed, a clear availability calendar, a simple payment form, an immediate and reassuring confirmation. That is where the human traveler lands after being convinced during the AI-assisted research phase, and that is where the final conversion happens β still, and for a while yet, in the hands of a person, not an agent.
6. Don't Build for a Future That Hasn't Arrived
Faced with any new technology, it's tempting to position ahead of its hypothetical maximum adoption: build deep integrations for fully agentic payment right now, in case the market flips overnight. Current data suggests caution on this specific point. Two to eight percent trust at payment is not a temporary technical friction about to dissolve within a few months β it's a trust friction, slower to shift, and travel remains one of the categories where that friction is strongest.
That doesn't mean ignoring the topic: it's reasonable to track how it evolves, and stay ready to adapt if trust rises meaningfully. But it does mean not over-investing today in agentic payment infrastructure at the expense of the fundamentals that already have a measurable effect right now: being well cited during the research phase, and delivering a flawless booking experience the moment a human takes back control. That's a solved and solvable problem. The other one isn't yet, and doesn't need to be for your property to perform.
The AI research phase is being won or lost right now. AIscore scans your hotel website across 91 signals β schema, bot access, content structure, llms.txt β and shows you whether your property is visible at the exact moment travelers are planning with AI. Free, no sign-up required.
Make sure you're visible in the research phase
AIscore analyses your site in 30 seconds and shows you exactly where you stand.
Scan your hotel now β