The Tour Operator Is Being Rehabilitated
For two decades, the travel industry wrote them off. The AI layer is bringing them back.
We ran 148 agentic shopping sessions across ten of the largest UK-facing travel platforms — Booking.com, Expedia, Trip.com, TUI, Agoda, Hopper, Jet2, Trivago, Skyscanner, and Tripadvisor — on ChatGPT, Gemini, and Perplexity.
99 in 100 recommendations bypassed all ten entirely.
The brands capturing the sessions were Jet2holidays, EasyJet Holidays, Kuoni, Thomas Cook, and First Choice. Operators the travel industry largely believed were being marginalised by OTA growth.
They weren't optimising for AI. They were just built the right way.
The architecture problem
OTAs built their dominance on search logic: broad inventory, price comparison, ranking algorithms that rewarded scale. An AI agent doesn't work like a search engine. It doesn't return 400 results ranked by price. It reasons toward a recommendation.
When a UK family of four asks ChatGPT for a week in Greece in August under £2,500 with kids' clubs and free cancellation, the AI applies an implicit checklist. Is this end-to-end? Are transfers included? Is ATOL protection documented? Is the package pre-assembled?
OTAs score poorly on every one of these because their model is modular. The AI layer prefers pre-assembled. The specialist tour operator wins not because it optimised for AI but because its product shape matches what the agent is looking for.
Search rewarded inventory. AI rewards curation.
Three ways to fail, one way to survive
The study identifies four distinct failure classes. Understanding which one you're in determines whether your remediation investment works at all.
Booking.com, Expedia, and Trip.com face a Reasoning Gap. The LLM knows them, acknowledges them as valid candidates, and then chooses against them. They are in the comparison set and losing. The fix is structural data signalling package equivalence — ATOL flags, transfers-included signals, kids-club provision — in crawlable positions, not buried in booking flows.
Hopper and Agoda face Categorical Mis-classification. The LLM has classified Hopper as a flight aggregator app and Agoda as an Asia-inventory platform. They are filtered out before comparison begins. No editorial investment fixes a failure that occurs upstream of evaluation.
Trivago, Skyscanner, and Tripadvisor face Pre-booking Displacement — a failure class we hadn't anticipated before this study. The AI treats them as discovery tools, not booking destinations. They appear in the reasoning chain as the search step before the agent names somewhere to actually book. They never appear as captors. The fix requires re-architecting as a commit-stage destination, not a search tool.
TUI is the only brand that survives. It wins 1 of 15 agentic sessions and is the only brand to record a positive mean RCS. The source-diet analysis explains why: TUI has twice the brand-owned and brand-editorial presence in the agent's reasoning chain compared to any other audited platform. That's not a deliberate AI strategy. It's a legacy content footprint that happens to match what the agent wants to read.
The risk for TUI is drift. That advantage erodes the moment competitors invest in the same layer.
The number that should alarm every travel brand
Travel editorial content — the specialist, expert-authored material that should be driving holiday recommendations — accounts for 0.6% of the AI's source diet across 3,209 annotated probe turns.
The agent making a multi-thousand-pound family holiday recommendation is drawing on a substrate that barely includes travel expertise. General web content accounts for 45%. Retail commerce pages account for 21%. The industry that built its credibility on specialist knowledge has almost no presence in the layer that now makes the first recommendation.
The PSOS finding destroys the visibility narrative
We measured first-prompt visibility — PSOS — across all ten brands. Booking.com leads the cohort with a score of 50. It records 0% own-commit rate in 148 agentic sessions.
TUI sits 9th of 10 on PSOS with a score of 41. It is the only brand to capture any agentic commits.
Visibility does not predict commercial wins at the decision stage. The commit decision rides on source-diet — specifically on brand-owned and brand-editorial content in the agent's reasoning chain. Not on mention frequency, not on citation share, not on share of voice.
This is the finding that makes every citation monitoring dashboard look like the wrong instrument.
What this means for the category
The travel industry is earlier on this curve than financial services. 43% of decision-stage probes produced no winner — the AI hedged rather than committed. That uncertainty window is unformed market share. The brands that build the right signals now will own it when the agent starts committing consistently.
The specialist operators winning today didn't plan for this. They happened to be built in a way the AI reasoning layer understands. That is not a durable advantage. It is a head start.
The OTAs that understand which failure class they're in and invest in the right remediation tier have a window to close the gap. The ones that continue optimising for search-layer metrics — citation share, share of voice, AI mention rate — will watch that window close.
The full State of Agentic Travel Booking 2026 covers all ten brands across four failure classes, a complete source-diet breakdown, platform divergence analysis across ChatGPT, Gemini, and Perplexity, and a brand-by-brand remediation roadmap. DM for the report.