Will AI Replace Tour Operators? The Data

By Tessar Napitupulu, Founder & CEO, PT Arfadia Digital Indonesia, GEO pioneer in Indonesia since 2023. More from Tessar.

Depending on which survey you read, somewhere between 28% and 56% of travelers now use AI in some part of planning a trip, and only 2% trust it to book one without a human checking first. Those two numbers are not contradictory. They describe two different jobs, one AI has already taken over and one it has barely started, and the confusion between them is why the "will AI replace tour operators" debate keeps producing opposite answers from equally credible sources.

Why Do AI Travel-Adoption Statistics Vary So Much?

Because they are measuring different things and get quoted as if they were measuring the same thing. Phocuswright's own tracked figures show AI-for-travel-planning usage moving from roughly 28% to 39% on a "used AI specifically to plan a trip" basis. A separate, broader "used AI at any point in planning, booking, or during the trip" basis shows figures moving from roughly 33% to 43% to 56% over a similar window. Both tracks are real and both are commonly cited, but chaining them together into a single "28% to 56%" trend line, as often happens in industry commentary, describes a track that does not actually exist. The honest statement is narrower and less dramatic: a meaningful and growing share of travelers use AI somewhere in the journey, the exact share depends heavily on how the question is worded, and no single number captures it cleanly.

Does High AI Usage for Planning Mean Low Trust for Booking?

Yes, and that gap is the single most important number in this entire debate. A Skift survey published in March 2026 found only 2% of leisure travelers are currently willing to let an AI system book a trip without a human reviewing it first. Put the planning-adoption figures next to that 2% and a consistent picture emerges: travelers have adopted AI enthusiastically for the research and comparison work, and have not remotely adopted it for the actual transaction. That is not AI failing to deliver on autonomous booking. It is travelers making a specific, rational trust distinction between "help me think" and "act on my behalf with my money," and that distinction has not moved much even as planning adoption has climbed steeply.

What Is Actually Happening on the Agentic-Booking Side, Then?

Real infrastructure is being built, even if consumer trust for autonomous booking remains low. Priceline launched "Penny," an AI agent handling parts of the trip-planning and booking flow. Sabre, PayPal, and Mindtrip announced integrations aimed at agentic, permissioned commerce, where an AI agent can execute a purchase within limits a traveler has explicitly authorized. Expedia's "Romie" and Booking.com's AI trip planner both extend further into the booking step than a year ago. None of this describes AI booking trips today without a human in the loop. It describes the plumbing for agentic commerce being laid down in 2025 and 2026, ahead of the trust curve rather than in response to it.

What Does Disintermediation Actually Mean for a Tour Operator's Business?

The generic-information layer of a tour operator's value is genuinely at risk. If a traveler's only need was a list of things to do in a destination and an estimated budget, an AI system already answers that competently, and that specific service, once a real reason for a traveler to visit an operator's inspiration content, is eroding. What is not eroding, and shows no sign of eroding based on any data available in 2026, is the execution layer: a guide with local weather judgment who reroutes a hike when conditions turn, an operator who can rebook a missed liveaboard connection at 11pm, someone who takes actual responsibility if a permit falls through or a driver doesn't show. An AI system can draft a compelling day-by-day plan. It cannot be present, accountable, or liable when the plan meets reality.

Layer of the trip	AI's current capability	Where the operator's value holds
Inspiration and generic information	Strong, and improving quickly	Weakening; this layer is genuinely at risk
Itinerary drafting and comparison	Strong for generic plans	Holds for constraint-specific and expert-curated plans
Booking and payment	Early pilots exist, trust is very low (2%)	Strongly holds for now
On-ground execution and risk-handling	None	Fully holds; this is the durable core of the business

Which Kinds of Trips Are Most Exposed, and Which Aren't?

Not every operator faces the same degree of risk, and the difference tracks closely with how much of the trip's value sits in information versus execution. A simple city-break itinerary, three days in Yogyakarta with the standard temple-and-food circuit, is close to fully commoditized information an AI system already assembles competently, and an operator whose entire offering is that itinerary with a markup is genuinely exposed. A multi-day liveaboard diving trip with permit logistics, weather-dependent routing, and safety decisions made in real time on the water is a different category almost entirely: the itinerary is a small fraction of the value, and the parts that matter, judgment, accountability, and physical presence, are exactly what an AI system cannot provide regardless of how good its underlying travel knowledge becomes.

The same logic applies to technical trekking, climbing expeditions, remote-area logistics requiring local relationships and permits, and any trip type where something going wrong has real consequences. The more a trip's value depends on things that can only be verified or handled in person, the more durable the operator's position, almost independent of how capable AI planning tools become.

What Does Travel's Own History With Disintermediation Already Show?

This is not the industry's first disintermediation cycle. Online travel agencies did to traditional storefront travel agents in the 2000s and 2010s roughly what AI planning tools are now doing to generic information provision: they made the commoditized part of the job, browsing options and comparing prices, available directly to consumers without an intermediary. Traditional agents who competed on providing generic information mostly did not survive that transition. Traditional agents who repositioned around complex itineraries, corporate travel management, high-touch luxury planning, and destinations requiring genuine specialist knowledge largely did, and many still operate profitably today, decades after OTAs were supposed to make them obsolete. The AI shift looks structurally similar: it is commoditizing the same layer OTAs commoditized two decades ago, one level further down, and the operators most likely to navigate it well are the ones already positioned around complexity and execution rather than generic information provision.

Where Does MaiA Fit Into This Picture?

Indonesia's own government entered this space directly when the Ministry of Tourism launched MaiA on 28 November 2025, an AI trip-planning assistant embedded in the Wonderful Indonesia platform. That a national tourism authority built its own AI planning tool, rather than treating AI planning as a foreign platform problem to react to, is itself a signal worth reading carefully: the generic-information layer of trip planning is now considered infrastructure the government wants to own and shape, not a niche consumer behavior. For operators, that raises the bar on what "generic" content is worth building at all, since a government-backed, nationally promoted tool is a difficult competitor to out-rank on broad destination information. It simultaneously reinforces the argument made throughout this piece: the defensible ground is execution and specificity, not the kind of general destination information a well-funded national platform is now built specifically to provide.

How Should an Operator Actually Respond to This, Practically?

Two moves matter more than any others. First, stop competing with AI on generic information, and instead make sure the operator's specific, structured content, licensing detail, and constraint-specific expertise are what AI systems cite when a traveler's question gets more specific than "what should I do in Bali." That is the GEO strategy covered in our piece on Instagram out-citing OTAs in Indonesian AI search. Second, build the clean, structured, schema-backed data foundation now, even while consumer trust in agentic booking is low, since that data foundation is exactly what determines whether an operator is selectable when an AI agent does eventually get permission to transact on a traveler's behalf. Waiting until agentic booking is mainstream to start structuring data means starting from zero at exactly the moment competitors who started early have already compounded an advantage.

What Should an Operator Track to See This Shift Coming, Rather Than After It Arrives?

Three leading indicators are more useful than waiting for booking volume to move, since booking volume is a lagging signal that only confirms a shift well after it has already happened. First, the share of inquiry-form submissions that reference an AI tool by name, "ChatGPT recommended you" or "I asked Gemini and found your page," which is a direct, low-cost signal of AI-referred demand that most booking systems do not currently capture unless someone deliberately adds a field for it. Second, citation presence on a fixed panel of the operator's own core queries, tracked weekly across the major platforms, since a decline here often precedes a decline in direct organic traffic by weeks or months. Third, the ratio of assisted conversions to last-click conversions in analytics, since a rising share of AI-influenced-but-not-AI-attributed bookings is exactly the kind of shift a traffic-only dashboard misses entirely.

None of these three require new technology investment. They require adding a field to an inquiry form, running a prompt panel manually or through a monitoring tool, and looking at an analytics report differently than most operators currently do. The cost of tracking this is close to zero. The cost of not tracking it is finding out about a real shift in demand six months after competitors who were tracking it already adjusted.

So Is the "AI Will Replace Tour Operators" Framing Wrong?

As a binary claim, yes. As a description of a real shift in which parts of the trip AI now handles, no. The honest version, supported by the actual data rather than either extreme, is that AI has already absorbed a meaningful share of the generic-information layer, is nowhere near absorbing the transaction layer given a 2% trust figure, and has not touched the execution layer at all. An operator's realistic threat is not replacement, it is irrelevance in the narrow generic-information slice of the business that competitors structuring for AI citation are already claiming, while the execution-layer value that actually justifies an operator's existence remains untouched and, if anything, more valuable as the generic layer commoditizes around it.

---

Frequently Asked Questions

Should we be worried about AI trip planners taking our customers?

Worried about losing the generic-inspiration customers who were never going to book a full-service package anyway, somewhat. Worried about losing customers who need execution, risk management, and local expertise, no, based on any data currently available.

When will AI actually be trusted to book trips autonomously?

No credible forecast puts this at scale before the late 2020s at the earliest, and trust shifts like this tend to move slower than the underlying technology. Treat 2026 and 2027 as an infrastructure-building period, not a booking-automation period.

Is it worth building schema and structured data now if AI booking isn't mainstream yet?

Yes, for two separate reasons that don't depend on agentic booking maturing: it directly improves AI citation for planning and trust queries today, and it positions the operator to be selectable once agentic booking does mature, rather than starting the data build from zero at that point.

Which travel-adoption statistic should we actually use in our own marketing or planning?

Whichever one matches the specific claim being made, cited with its exact basis stated. Citing "up to 56% adoption" without specifying which measurement basis that figure uses is the exact conflation this article warns against.

Does this analysis apply equally to leisure and business travel?

The trust and adoption figures cited here are leisure-travel-focused. Business travel shows somewhat different dynamics, generally faster institutional AI adoption for expense and policy compliance, but similarly low trust for fully autonomous booking without an approval step.

Are simple package tours doomed, then?

Not doomed, but structurally more exposed than complex or execution-heavy trips. An operator built entirely around simple, generic packages should treat this as a real signal to either add genuine differentiation and specificity, or accept a shrinking margin as that layer commoditizes further.

How is this different from previous "technology will replace travel agents" predictions that didn't fully pan out?

The OTA-era prediction was directionally correct, it just took a specific form: it replaced generic information provision and simple booking, not the entire industry, and specialist operators adapted and survived. The realistic read on AI is the same pattern one layer deeper, not a repeat of a prediction that turned out to be entirely wrong.

---