Your Travel App Is Connected to the AI. It's Still Getting Skipped.
TravelThe travel industry spent months integrating with AI chatbots. Skift’s testing found the integrations mostly work — but only if the chatbot chooses to use them.
What the Test Found
Skift tested connected travel apps across ChatGPT and Claude’s connector ecosystem, using Booking.com, Expedia, and Viator. The headline finding: Claude proactively surfaced relevant connectors in context, routing queries through connected apps with minimal friction. ChatGPT’s behavior was less consistent — in multiple cases it initially told testers it could not use connected, authenticated apps, eventually working only after additional prompting.
Expedia confirmed the ChatGPT behavior is a known platform bug. But the broader problem is structural, not a single bug: even when ChatGPT eventually used a connected app, it initially bypassed it. For a travel brand, that initial bypass is the referral that doesn’t happen.
The commercial model divergence adds another variable. OpenAI is layering advertising into ChatGPT — Booking.com and Expedia ads appeared beneath competitor results in testing. Anthropic has committed to staying ad-free in Claude. For travel brands, the difference matters: in OpenAI’s model, appearing prominently may depend on paid placement in addition to (or instead of) integration quality.
Why It Matters
For travel brands, the old visibility playbook was: rank on Google, appear in metasearch, optimize for OTAs. A connected travel app inside an AI chatbot looks like a new referral channel with better conversion potential — users are already expressing intent in natural language, asking for exactly the product you sell.
The problem Skift identified flips that optimism: connection does not guarantee usage. If the chatbot’s model or retrieval logic deprioritizes your integration, you’re invisible — not because you didn’t integrate, but because the chatbot decided not to invoke you. This is a new category of distribution risk with no existing mitigation playbook.
The finding also has implications for how brands approach multi-AI distribution. Claude and ChatGPT behaved materially differently in the same scenario. A brand that optimizes connector behavior for one platform is not automatically optimized for the other.
The So What
Travel brands need to add “AI referral testing” to their distribution monitoring stack alongside SEO rankings and OTA positioning. Integration is table stakes; what matters is whether the chatbot invokes your app when a relevant query arrives. Testing your own connector behavior — and monitoring it over time as model updates change retrieval patterns — is the new technical SEO for the AI era. The Skift finding that a well-connected brand still gets skipped is the clearest evidence that integration work is not the end of the distribution problem, it’s the beginning of a new one.
Source: Skift
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →