The direct answer: Activities are where AI travel planning is strongest at the top of the funnel and weakest at the bottom — and that combination quietly redirects tour operators’ demand into commission channels. Among travelers who use AI for travel planning, the single most common use is discovering activities and restaurants (65%), ahead of finding destinations (56%) and building itineraries (55%) (TakeUp AI Research, April 2026). Meanwhile, only 17% of travelers have used AI to book an activity, even though 69% have used AI to plan or book a trip overall, and one in three trip-planning sessions now starts with AI (GetYourGuide/Arival research, 2026). That is the 65/17 gap: AI answers decide what travelers do, then hand the transaction to whatever bookable link sits closest to the recommendation — usually a marketplace taking 20–30% commission (Samba HQ OTA commission tracker, August 2026). For an operator, the strategic response is not to fight AI discovery — it is to be the named business in the answer and to make your own site the path of least resistance from that mention to a confirmed booking.
Why activities are the strangest category in AI travel planning
Hotels get compared on price and location. Flights get compared almost exclusively on price. Activities are different: the traveler asks an open question — “what should we do in Porto with two kids and a rainy Tuesday?” — and the assistant answers with a curated shortlist, often with no price comparison at all. The recommendation is the product.
That is why discovery is the top AI use case in travel: 65% of AI travel planners use it to find activities and restaurants, the highest of any category measured (TakeUp AI Research, April 2026). And the habit is mainstream now — one in three trip-planning sessions starts with AI (GetYourGuide/Arival, 2026), and 37% of U.S. travelers use AI to plan trips (Allianz Partners survey, reported by Hotel Dive, July 23, 2026).
But the bottom of the funnel behaves differently. Only 17% of travelers have booked an activity through AI (GetYourGuide/Arival, 2026). The assistant inspires; the transaction happens later, on a page the traveler trusts — and the traveler typically gets there by searching or clicking the name the AI gave them. GetYourGuide’s own product chief frames it plainly: “AI’s strongest role today is inspiration, not final transaction” (GetYourGuide Spring 2026 release, April 22, 2026).
Here is the part most operators miss: the inspiration step is not neutral. When ChatGPT or Gemini answers “best food tour in Lisbon,” it names three to five businesses. The traveler doesn’t choose from the universe of Lisbon food tours — they choose from the AI’s shortlist. If you’re not in the answer, you don’t exist for that traveler, no matter your reviews, your ranking inside Viator, or your Google Ads budget. We’ve documented the same mechanics for hotels in what moves AI hotel recommendations — but activities have one crucial difference that makes the stakes higher, not lower.
The difference that raises the stakes: the money
For hotels, an AI answer that sends the traveler to Booking.com costs the property its 15–20% OTA commission on that stay. For activities, the commission economics are materially worse:
| Booking channel | Typical commission (2026) | Notes |
|---|---|---|
| Viator | 20–30% (25% most common) | Boost programs push visibility in exchange for higher tiers |
| GetYourGuide | 20–30% by country | 30% is the common starting rate for new operators; volume negotiates down |
| Klook | 15–25% | Lower rate, thinner Western coverage |
| Airbnb Experiences | Flat 20% | |
| Google Things to Do links | ~0% | But the click still goes through Google’s surface first |
| Your own website | 0% | The only channel where the margin stays with you |
(Sources: Samba HQ OTA commission tracker, August 18, 2026; automate.travel operator analysis, June 7, 2026.)
Run the math on a real booking. A €75 per person food tour for a family of four is a €300 sale. On Viator at 25%, that’s €75 gone before you’ve paid a guide, bought the tastings, or covered insurance. Do that forty times a season and you’ve handed the marketplace a five-figure sum — for demand that, in a growing share of cases, originated in an AI answer where the marketplace’s brand and reviews were simply more citable than yours.
And this is the structural trap of 2026: marketplaces have made themselves the easy citation for AI assistants. Since April 2026, Tripadvisor and Viator content flows directly into Claude, ChatGPT, Perplexity and Alexa+ itineraries, so a marketplace page with 4,872 reviews is increasingly the source an assistant reaches for when it names a tour — we broke down that partnership layer in Tripadvisor and Viator are now inside Claude, ChatGPT and Alexa+. GetYourGuide has a ChatGPT app of its own (Spring 2026 release, April 22, 2026). The marketplaces are not waiting for AI to commoditize them; they are becoming AI’s inventory layer. Every operator whose citation presence is weak is, functionally, voting to make the marketplace their storefront.
How an AI answer actually assembles a “things to do” list
Understanding what the engine does with your question tells you exactly what to fix. When a traveler asks “what should I do in Seville,” the assistant, broadly, does four things:
- Entity resolution — it maps “Seville” to a place and assembles a candidate pool of activities, tours and attractions from its index and live sources.
- Source synthesis — it weighs what established publishers, review platforms, tourism boards and operator pages say, looking for consensus about what’s worth doing.
- Constraint matching — it filters against the traveler’s stated constraints: kids, budget, rain, mobility, two hours spare.
- Citation selection — it names a short list, usually three to seven items, with one-line reasons.
You cannot buy your way into step 4 the way you can buy a top ad slot on Google. You win it by being legible at every step: an entity the model can resolve, a source it can verify, a page that states constraints clearly, and a consensus of mentions across the web. That is generative engine optimization for activities — and it is almost entirely unclaimed territory. Most operators’ sites fail at step 1 (no structured data tying the business to its destination and activity type), step 3 (nothing on the page says “runs in rain,” “ages 6+,” “2 hours”), or both.
The operator playbook: seven fixes
1. Publish constraint-annotated tour pages. For each flagship experience, state in plain text: duration, age range, rain plan, physical intensity, languages, group size, price, what’s included. This is not web copy hygiene — it is the raw material the engine uses for step-3 constraint matching. A page that never says “suitable for kids 6+” cannot be recommended to a family, no matter how good the tour is.
2. Mark up with schema — properly. TouristAttraction, TouristTrip, Offer and LocalBusiness structured data are what turn your page from prose into an entity. Validate coverage across every experience, not just the homepage. Schema is boring, unclaimed, and disproportionately effective in activities because so few operators do it.
3. Build the itinerary-shaped content layer. AI answers assemble itineraries, so give them itinerary-shaped raw material: “48 hours in X with kids,” “rainy-day plan for X,” “one week in X: the local’s version.” These pages get cited, and citations mention the named operator. The average 2026 trip is ~6 nights and roughly 1 in 12 is multi-city (MonkeyTravel 1.1M-search report, July 27, 2026) — content shaped like real trips wins real citations.
4. Make your review footprint citable. Fresh, dated, detailed reviews on your own Google Business Profile and major platforms — with owner responses — are the consensus signal engines synthesize. GetYourGuide now surfaces AI-generated review summaries above its activity reviews (April 22, 2026): the synthesis layer is coming for reviews everywhere, and thin review profiles summarize badly.
5. Own the in-destination repeat of the question. 51% of travelers ask AI for real-time recommendations during the trip (Phocuswright, March 2026, as we analyzed in the second visibility window) — and same-day activity bookings are the highest-margin demand an operator gets. Keep availability and prices current on your own pages; stale data quietly disqualifies you from same-day answers.
6. Convert the mention. When a traveler who saw your name in an AI answer arrives, you have seconds. Fast site, calendar-style availability, instant confirmation, no forced account creation, and a booking link that works from a phone in a foreign country. If your direct path is harder than the marketplace’s one-tap checkout, the 65/17 gap swallows the margin anyway.
7. Measure citations, not vibes. Track across engines: are you in the candidate pool at all, how often are you named for your priority prompts, and who else is on the list. Measure smoothed over weeks — AI rankings churn heavily week to week, as our 11-week hotel volatility study showed (July 2026), and one-off spot checks mostly measure noise. Then tie citation share to direct-booking volume, the way we outline in measuring AI-driven direct bookings.
What this looks like as a program
A focused operator program is small: an audit of which engines currently name you (and which never do), a fix sprint on schema and constraint content for your top experiences, an itinerary content layer of four to six destination pages, and a monthly citation-and-booking measurement loop. Palmtree runs exactly this for tour operators — the operator-side work is detailed on AI visibility for tour operators, the baseline starts with a travel AI audit, operators who want it handled end-to-end can look at our travel GEO agency engagement, and everything is scoped on pricing. Hotels have spent two years learning this lesson; activities is the category where the lesson is still cheap.
FAQ
Do travelers actually book tours through AI assistants yet? Mostly no — that’s the 17%. Only 17% of travelers have used AI to book an activity, versus 69% who have used AI to plan or book travel at all (GetYourGuide/Arival, 2026). The booking still happens on a website or app; what moved into the assistant is the shortlist decision.
If AI doesn’t book, why should a tour operator care about AI visibility? Because the answer names the businesses the traveler will consider. If an assistant recommends three cooking classes in Rome and yours isn’t one, you lost that customer before they ever saw a booking page — and the winner is whoever was most citable, not necessarily best. Discovery is the #1 AI use case in travel (65%, TakeUp, April 2026).
What commission do Viator and GetYourGuide charge in 2026? Viator typically 20–30% with 25% the most common standard; GetYourGuide 20–30% depending on country and volume, with 30% a common starting rate for new operators; both run boost programs that trade higher commission tiers for visibility (Samba HQ, August 2026; automate.travel, June 2026).
What is the single highest-leverage fix for a small operator? Structured data plus constraint-annotated experience pages. Together they make you resolvable, verifiable and constraint-matchable — the three gates an AI answer applies before it ever names you. It’s cheap, it’s yours, and almost no operator has done it.
How do I measure whether this is working? Track named mentions across engines for a fixed set of prompts (“best X tour in Y,” “what to do in Y with kids”), smoothed over 4–8 weeks, then correlate with direct bookings and inquiry volume from your own site. Citation share that doesn’t move bookings means your conversion path, not your visibility, is the bottleneck.
Sources
- GetYourGuide Spring 2026 release and GetYourGuide/Arival research (April 22, 2026)
- TakeUp AI Research, AI travel planning usage (April 2026)
- Samba HQ, OTA Commission Rates 2026 tracker (August 18, 2026)
- automate.travel, Viator vs GetYourGuide operator analysis (June 7, 2026)
- MonkeyTravel, 2026 Travel Planning Report, 1.1M searches (July 27, 2026)
- Allianz Partners / Hotel Dive, U.S. AI travel planning survey (July 23, 2026)
- Phocuswright, traveler AI usage (March 2026)