AI is changing everything about how travel gets found and sold.

My read on the shifts underneath it, and what it costs an operator to miss them.

The feed is the storefront

Discovery for considered travel purchases has migrated to social feeds. Two-thirds of Gen Z travelers already use social media for trip inspiration, and 41% of consumers overall research destinations there. The traveler meets the trip in a swipe, not in a search box.

2 in 3 Gen Z travelers use social media for trip inspiration

The working paper

Data Axle consumer survey, fielded by Dynata, March 2025, n=1,000

Paid and organic have converged

The ad and the post now share a form: the same short vertical video, judged the same way. TikTok’s own creative analysis found 63% of the highest click-through videos hook the viewer inside the first three seconds. What differs is only who paid for the reach.

3s to earn the next second of attention, paid or not

TikTok for Business creative analysis

Nobody wants the blank box

Even before AI, free text search was not how travelers preferred to shop. A blank box hands the whole problem to the average user, and it is overwhelming. Filtered attributes, visual browsing and structured choices won through years of testing because travelers prefer to be guided. A proposal beats a prompt.

The working paper

Free text search has never converted particularly well; the optimized booking flow exists for a reason

The first click is where it breaks

The creative promises a specific trip. The click lands on a generic booking surface, and everything the ad said about the trip is discarded in the handoff. Personalizing the ad without personalizing the destination is the industry default.

What the click discards

Documented in the working paper’s survey of Meta Travel Ads and dynamic creative platforms

The window outlives the identity

A hotel guest decides in about 36 days. A cruise guest can take up to 18 months. Browser-resident identity forgets in as little as 7 days, so the technical memory of the traveler expires long before the decision is made.

52× longer than the browser remembers, at the cruise extreme

The working paper

SiteMinder booking lead times · CLIA advisor surveys · Safari’s 7-day identity cap

Posting regularly is infrastructure

Feeds reward accounts that publish consistently, and AI summaries assemble whatever material exists. In Buffer’s analysis of over 100,000 accounts, those posting at least weekly for 20 of 26 weeks earned about five times the engagement per post of inconsistent accounts. A brand that goes quiet is summarized by its gaps.

5× engagement per post for highly consistent accounts

The working paper

Buffer, analysis of 100,000+ accounts over 26 weeks

Follow-up speed is a conversion lever

Demand that surfaces in the feed resolves with whoever answers first. Choice Hotels reported that responding to group RFPs 30% faster lifted conversion by 250 basis points. The pattern is not new: in Harvard Business Review’s study of 1.25 million leads, firms responding within an hour were seven times likelier to qualify the lead.

+250 bps conversion from answering 30% faster

The Choice Hotels note

Choice Hotels, Q1 2026 earnings call · Harvard Business Review lead response study

Working papers

Longer arguments, with the evidence and the figures that go with them. Each one reads in full here and downloads as a PDF.

Figure 1 Two reservation books, twelve months, opposite problems Each square is inventory. Watch when it sells. The hotel has nothing to forecast from until the demand is already there. The ship has nothing left to price by the time costs move.

Hotel60% booking within 12 daysAccor, March 2026

Last-minute share30 to 35% of all bookingsfrom 15 to 20% pre-pandemic

Cruise85% of a year sold in advanceCarnival, March 2026

The two grids hold the same amount of inventory and sell roughly the same amount of it. Only the timing differs, and the timing is what every commercial model in travel is built on.

Figure 2 The report that was accurate every day it was useless The same pacing report, read at five points before arrival. Nothing on it is wrong. It simply cannot see the buyers, because they have not arrived yet.

This is the operational cost of the compression, and it is why yield management run on a weekly review cadence struggles at a twelve-day window. The decision points that matter now sit inside the period the historical curve has least to say about.

The booking window is diverging, and both directions break the same model

31 Jul 2026

Hotel booking windows are compressing toward twelve days while cruise windows stretch past a year. Compression removes the pacing signal revenue management is built on, commitment removes the ability to reprice, and the identity horizon underneath both is measured in days.

Read the paper

Figure 1 The visitor who did all the work, and left Two sessions on the same travel site, running side by side. One arrived from an assistant after describing the whole trip. One arrived from a search result. Watch which one books.

The counterpoint belongs in the same frame. That conversion gap has narrowed by nearly 70% since Adobe began tracking it in October 2024. If it closes entirely, the argument in this paper weakens considerably, and it is the number I would watch before any other.

Figure 2 The empty field and the proposal Both surfaces are asking the traveler to choose. One asks them to supply the choice set from memory. One shows it to them.

A proposal is assembled from things the operator already holds: live inventory, the real fare, its own photography, and a record of what this person engaged with. The traveler's job becomes reacting to something concrete, which is the task the evidence says they perform well.

Nobody wants the blank box

31 Jul 2026

The industry is building conversational interfaces on the premise that travelers want to express intent in free text. The evidence from two decades of booking-flow optimization says they want to be shown options and to verify them, and the context that would make free text work sits with the assistants, while the operator starts at zero on every session.

Read the paper

Figure 1 Someone searches your name The assembled answer builds from whatever exists, which includes a review written under previous ownership, a headline from this week, and a forum thread. Your own page waits underneath it.

Branded queries carry the one encouraging exception in the search data, gaining around 19% in click-through where an assembled answer appears, which fits the reading that someone searching your name has already decided and the summary confirms them. That exception holds only while the summary is favorable.

Figure 2 The renovation the answer never heard about Capital goes in and the product changes. The material the answer is built from does not, because the old material has had years to accumulate links, reviews and repetition.

This is the line most renovation budgets are missing. Improving the physical product no longer guarantees the improvement propagates into how the market understands the property, and the material that describes the old version does not expire on its own.

The record you no longer write

31 Jul 2026

Assembled answers are becoming the operating description of a travel business. They are built from material written across ownership cycles, from a news cycle the brand does not control, and from forums that are being deliberately polluted. Consistency and continuous publishing are what the assembly reads.

Read the paper

Figure 1 The same person, four times, and the site never knows them twice One cruise decision runs February to August. This is what the operator's own site is holding on the person making it, visit by visit.

Vacation rentalunder 30 days to decide4 expiries

Hotel36.4 days to decide5 expiries

Cruiseup to 18 months to decide77 expiries

Safari deletes all script-writable storage after seven days of browser use without a return visit to the site, and caps cookies set on an ad landing page at 24 hours when the arrival URL carries click identifiers. Even at the hotel end of the range, where the window is five weeks rather than eighteen months, the record expires five times over before the booking is made.

Figure 2 What the click discards The post is about one specific trip. The page it opens is about the whole catalog. Everything the creative established stops at the boundary between them.

Meta Travel Ads renders the catalog row into the creative and optimizes delivery against it. Feed-driven creative platforms generate variations from the same feed. Both personalize the ad. Neither personalizes the destination, and a customer data platform holds traits and segment memberships rather than which creative began this and what it depicted.

Figure 3 What the surface knows, across one consideration window The same five touches either decay to nothing between visits or accumulate. Read the right-hand edge: that is the state the booking surface renders from.

Without a thread, each touch is read in isolation and the record of it expires before the next one.

With the thread, the engaged creative is stored structurally at the first touch and is still there at the last.

Without a thread, a second view of the same sailing looks identical to a first.

With the thread, history is retained rather than last-value, because recurrence is signal.

Without a thread, re-identification depends on device state that the window outlives.

With the thread, the token in the email link resolves server-side to the record.

Scoping to the trip rather than the person keeps concurrent intents from contaminating one another. A person record holds only durable attributes, and a new intent inherits selectively: embarkation preference and physical constraints carry, prior destinations and stale budgets do not.

Carrying creative semantics across the travel consideration window

24 Jul 2026

Discovery for considered travel purchases has migrated to social feeds, while purchase resolves days to months later at a booking surface. The consideration window exceeds the lifetime of browser-resident identity, and the semantics of the initiating creative are discarded at the click.

Read the paper

What I'm building

Ondra

An AI-powered travel marketing platform that turns social media into a sales engine.

S StayStrategy

An AI-native performance marketing agency for independent travel and hospitality operators.

Writing

Applied AI A knowledgable skeptic between content and reader Pricing and revenue architecture Carnival's committed forward book cuts both ways Distribution and the direct channel Etihad's loyalty response is sized for the wrong risk Travel technology The tab problem is real, the conclusion is wrong

All writing

Get in touch

I am always glad to talk with people working on the commercial side of travel, whatever the context.

Email me Set up a meeting