The Intent-Source Divide: How AI Search Queries Shape Hotel Discovery

A new arXiv study audits Google Gemini's hotel recommendations in Tokyo, finding a 25.1 percentage-point gap in citations between experiential and transactional queries. This 'Intent-Source Divide' suggests AI search may reduce reliance on Online Travel Agencies (OTAs) for discovery.

What Happened

A new research paper, "The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries," was posted to arXiv on March 20, 2026. The study presents an audit of 1,357 grounding citations generated by Google Gemini in response to 156 hotel queries for Tokyo. Its core finding is a systematic pattern the authors term the Intent-Source Divide.

The research categorizes user queries into two types:

• Experiential Queries: Focus on the quality of the stay (e.g., "best hotel for a romantic getaway in Shinjuku," "most stylish boutique hotel in Shibuya").
• Transactional Queries: Focus on booking and logistics (e.g., "cheap hotel Tokyo," "hotel near Tokyo Station with free cancellation").

The audit reveals a stark difference in the sources Gemini cites to ground its answers. For experiential queries, 55.9% of citations came from non-OTA sources (like travel blogs, review sites, magazine articles, and direct hotel content). For transactional queries, only 30.8% of citations were from non-OTA sources, with the vast majority coming from OTAs like Booking.com or Expedia.

This creates a 25.1 percentage-point gap (p < 5 × 10⁻²⁰) in citation sourcing based purely on query intent. The effect was even more pronounced in Japanese-language queries, where experiential prompts drew 62.1% non-OTA citations versus 50.0% for English queries. The authors suggest this is consistent with a more diverse and robust ecosystem of non-OTA travel content in Japanese.

Technical Details

The study is an audit of a production-scale AI system's retrieval-augmented generation (RAG) behavior. When a user asks Gemini a question, the model retrieves relevant information from a corpus of web documents to "ground" its answer, then cites these sources. This research methodology involved:

1. Query Design: Crafting a set of hotel queries for Tokyo that systematically varied intent (experiential vs. transactional) and language (English vs. Japanese).
2. API Calls & Data Collection: Submitting these queries to Google Gemini and programmatically collecting both the generated answers and the accompanying source citations.
3. Source Classification: Manually classifying each cited URL as originating from an OTA (Online Travel Agency) or a non-OTA source (media, blogs, direct, etc.).
4. Statistical Analysis: Quantifying the differences in citation patterns and testing for significance.

The key technical insight is that the AI's grounding mechanism is not neutral; it is highly sensitive to the semantic framing of the query. The model's retrieval system appears to associate different intents with different authoritative source types. This has less to do with the LLM's reasoning and more to do with the underlying information retrieval and ranking systems that feed it data.

Retail & Luxury Implications

While the study focuses on the hotel industry, its findings have direct and profound implications for luxury retail and fashion. The core dynamic—brands versus intermediaries in the discovery funnel—is identical.

The Parallel: From OTAs to Marketplaces
In luxury, the role of the OTA is played by major digital marketplaces (Farfetch, Mytheresa, Net-a-Porter), wholesale partners, and broad-spectrum retail platforms. These intermediaries have built immense power by aggregating demand and charging brands for access to it—through commissions, advertising fees, and prominent placement. Discovery has been, in effect, "rented."

How the Intent-Source Divide Applies to Fashion
A luxury consumer using an AI search engine (like Google's SGE, Perplexity, or future integrated models) will exhibit the same query behavior:

• Experiential / Inspirational Queries: "What to wear to a summer wedding in Capri," "most iconic vintage Chanel bag styles," "quiet luxury workwear inspiration," "best cashmere sweater for travel."
• Transactional / Direct Queries: "Buy black Gucci loafers size 41," "price of Small Lady Dior bag," "Prada Re-Nylon jacket stockists."

This research predicts that answers to experiential queries will be grounded significantly more in non-marketplace sources. These could include:

• Fashion journalism (Vogue, Harper's Bazaar, The Business of Fashion)
• Style blogs and influencer content
• Brand-owned content (campaign lookbooks, heritage stories, material guides)
• Cultural commentary and street style archives

Conversely, answers to transactional queries will remain heavily anchored to inventory-holding intermediaries—the marketplaces and major e-commerce retailers that have the structured product data, pricing, and availability APIs that AI systems rely on for definitive "buy" answers.

Strategic Shift: Competing on the Axis of Inspiration
The study suggests a potential future where AI begins to decouple inspiration from transaction in the consumer journey. The power to shape desire and set trends—the high-margin, brand-building part of the business—could gradually shift back towards entities that control compelling narrative content, not just inventory logistics.

For a luxury brand, this means:

1. Content as a Core Discovery Asset: Investment in high-quality, semantically rich, and authoritative digital content (articles, videos, editorial) is no longer just marketing; it becomes a direct feed into the AI discovery layer. This content must be optimized not just for SEO, but for "AI-GO"—AI Grounding Optimization.
2. Owned Channels Gain Leverage: A robust direct-to-consumer ecosystem (website, app, loyalty program) supported by deep content could see increased AI-driven referral traffic for inspirational queries, capturing consumers earlier in the decision journey.
3. Marketplace Relationship Evolution: The role of large platforms may intensify for fulfillment and last-mile transaction but could become less dominant for initial discovery. Negotiations may increasingly separate "demand generation" fees from "transaction processing" commissions.
4. Localization & Language Strategy: The amplified effect in Japanese indicates a competitive advantage for brands that build deep, culturally nuanced content ecosystems in key languages, beyond simple translation.

The "End of Rented Discovery" is not an overnight event, but the research identifies a measurable technological bias in how next-generation search works. For luxury brands, the imperative is to ensure their stories, expertise, and inspiration are the most groundable sources for the AI that will answer their future customers' most important questions.

---

Originally published on gentic.news