The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries

Peiying Zhu; Sidi Chang

cs.IRMar 20, 2026

The End of Rented Discovery: How AI Search Redistributes Power Between Hotels and Intermediaries

Authors: Peiying Zhu, Sidi Chang

Organizations: Blossom AI San Francisco, CA, USA

Abstract

When a traveler asks an AI search engine to recommend a hotel, which sources get cited -- and does query framing matter? We audit 1,357 grounding citations from Google Gemini across 156 hotel queries in Tokyo and document a systematic pattern we call the Intent-Source Divide. Experiential queries draw 55.9% of their citations from non-OTA sources, compared to 30.8% for transactional queries -- a 25.1 percentage-point gap ( $p < 5 \times 10^{-20}$ ). The effect is amplified in Japanese, where experiential queries draw 62.1% non-OTA citations compared to 50.0% in English -- consistent with a more diverse Japanese non-OTA content ecosystem. For an industry in which hotels have long paid OTAs for demand acquisition, this pattern matters because it suggests that AI search may make hotel discovery less exclusively controlled by commission-based intermediaries.

Figures & tables

Category	T non-OTA%	E non-OTA%	Gap (pp)	95% CI	$\chi^{2}(1)$	$p$ -value
Budget	15.5%	42.5%	27.0	[17.5, 36.5]	26.00	$3.4\times 10^{-7}$
Rating/Quality	22.1%	46.7%	24.6	[13.6, 35.5]	16.78	$4.2\times 10^{-5}$
Business	40.1%	68.5%	28.4	[18.1, 38.7]	25.81	$3.8\times 10^{-7}$
Convenience	40.1%	59.2%	19.0	[9.3, 28.8]	13.29	$2.7\times 10^{-4}$

Table 1. Non-OTA Citation Rate by Query Category and Intent

Segment	OTA%	Non-OTA%	$N$
EN Transactional	68.2%	31.8%	274
EN Experiential	50.0%	50.0%	386
JP Transactional	70.0%	30.0%	333
JP Experiential	37.9%	62.1%	364

Table 2. OTA and Non-OTA Citation Rates by Language and Intent

Source Type	EN	JP
Travel blog	22.9%	1.2%
Editorial curation	22.5%	12.3%
Hotel direct	19.3%	23.6%
Travel agency	0.0%	12.9%
Travel media	8.9%	3.7%
UGC	6.4%	8.6%

Table 3. Non-OTA Source-Type Composition by Language (% of Non-OTA Citations)

Language	T-rate	E-rate	Difference	95% CI	$p$
English	8.0% ( $n$ =274)	8.3% ( $n$ =386)	+0.3 pp	[ $-$ 4.0, 4.5]	0.90
Japanese	9.3% ( $n$ =333)	12.6% ( $n$ =364)	+3.3 pp	[ $-$ 1.3, 8.0]	0.16

Table 4. Hotel Direct Citation Rate (% of All Citations) by Language and Intent

Hotel	Cited	FAQ	Area Guide	Blog	Access	Unique Content	SAD Score
Kimpton Shinjuku	Yes	3	3	2	2	3	13/15
Kadoya Hotel	Yes	3	3	2	2	3	13/15
Super Hotel	Yes	2	2	2	2	2	10/15
Park Hotel Tokyo	Yes	1	1	1	1	3	7/15
Tokyu Stay	Yes	1	0	0	2	3	6/15
TRUNK (Hotel)	Yes	0	2	1	1	2	6/15

Table 5. Search-Answerable Depth Audit (14 Hotels)

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

ID	Language	Category	Intent	Query
q0157	EN	Budget	Transactional	Cheap hotel in Chiyoda
q0176	EN	Budget	Experiential	Good value hotel with local charm in Chuo
q0194	EN	Rating	Transactional	Best rated hotel in Chuo
q0212	EN	Rating	Experiential	Hotel with exceptional service and atmosphere in Chuo
q0229	EN	Convenience	Transactional	Hotel near Chiyoda station with easy access
q0169	JP	Budget	Transactional	新宿で安いホテル

Appendix

Table 6. Sample Queries from the Paired Design

Category	Type	Examples
OTA	Global OTA	booking.com, expedia.com (incl. regional variants), hotels.com, tripadvisor.com
	Japanese OTA	jalan.net, rakuten.co.jp, ikyu.com, jtb.co.jp
	APAC OTA	agoda.com, trip.com, traveloka.com
	Meta-search	kayak.com, trivago, skyscanner, hotelscombined
Non-OTA	Hotel direct	kimptonshinjuku.com, superhotel.co.jp, princehotels.com
	Editorial curation	wanderlog.com, myboutiquehotel.com, icotto.jp, ozmall.co.jp

Appendix

Table 7. Source Classification Taxonomy

Variable	M1 Coef	M1 OR [95% CI]	M1 $p$	M2 Coef	M2 OR [95% CI]	M2 $p$
Intercept	$-$ 1.429	0.24 [0.18, 0.32]	$<$ 0.001	$-$ 1.241	0.29 [0.21, 0.40]	$<$ 0.001
Intent (experiential)	1.080	2.95 [2.34, 3.71]	$<$ 0.001	0.782	2.19 [1.57, 3.04]	$<$ 0.001
Language (Japanese)	0.283	1.33 [1.06, 1.67]	0.015	$-$ 0.053	0.95 [0.67, 1.35]	0.769
Category: Rating	0.039	1.04 [0.74, 1.47]	0.827	0.037	1.04 [0.73, 1.47]	0.833
Category: Convenience	0.694	2.00 [1.48, 2.71]	$<$ 0.001	0.690	1.99 [1.47, 2.70]	$<$ 0.001
Category: Business	0.944	2.57 [1.87, 3.54]	$<$ 0.001	0.945	2.57 [1.87, 3.54]	$<$ 0.001

Appendix

Table 8. Citation-Level Logistic Regression—

P(\text{non-OTA Citation})

Variable	OR [95% CI]	$p$
Experiential	2.19 [1.85, 2.59]	$<$ 0.001
Japanese	0.95 [0.79, 1.14]	0.566
Rating	1.04 [0.87, 1.24]	0.680
Convenience	1.99 [1.71, 2.33]	$<$ 0.001
Business	2.57 [2.18, 3.03]	$<$ 0.001
Experiential $\times$ Japanese	1.77 [1.40, 2.24]	$<$ 0.001

Appendix

Table 9. Query-Level Quasibinomial Regression (

n=156

queries)

Test	Statistic	Value	$p$
Core finding
Chi-squared (intent $\times$ source)	$\chi^{2}(1)$	84.23	$4.4\times 10^{-20}$
Unadjusted OR (non-OTA, E vs T)	OR [95% CI]	2.84 [2.27, 3.56]	—
Cramer’s V	$V$	0.249	—
Mann-Whitney U (query-weighted)	$U$	4,721	$<10^{-9}$
Category-level

Appendix

Table 10. Complete Statistical Test Summary

E-Commerce Search

E-commerce search studies how to match shoppers’ queries, preferences, and product signals with useful items in online marketplaces.

See all 42 papers

Whose hotel does the AI recommend? An algorithm audit of reputation signals in LLM-assisted hotel selection
Jun 15, 2026Mirza Samad Ahmed Baig, Syeda Anshrah Gillani, Asher AliLLM Auditing Recommender Systems
Fandaqah, Al Khobar, Saudi Arabia. · Hamdard University, Karachi, Pakistan.
Does Rank Still Matter? Position Bias When AI Agents Shop on Our Behalf
Aug 24, 2026Davood Wadi, Yu MaAgentic Search Agentic Commerce
Desautels Faculty of Management, McGill University, Montreal, Quebec, Canada
Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain
Jul 8, 2026Qiaoni Shi, Kai Zhu, Kai GuInformation Retrieval
Bocconi University, Milan, Italy

Jun 15, 2026J/K move · Enter open · S save

Whose hotel does the AI recommend? An algorithm audit of reputation signals in LLM-assisted hotel selection

Mirza Samad Ahmed Baig, Syeda Anshrah Gillani, Asher Ali

Fandaqah, Al Khobar, Saudi Arabia. · Hamdard University, Karachi, Pakistan.

LLM Auditing Recommender Systems Large Language Model-Based Recommendation Position Bias

Open paper AskPDF arXiv

Travelers increasingly ask large language model (LLM) assistants which hotel to book, making these systems gatekeepers of property visibility -- yet what moves their recommendations is undocumented. We conduct a pre-specified algorithm audit using a randomized choice-based conjoint: across personas, prompt templates, and twelve open-weight and proprietary models, assistants choose among five hotels whose guest rating, review volume and recency, management response, chain affiliation, price, eco-certification, and list position are independently randomized. We estimate the average marginal component effect of each signal on the probability of recommendation. Guest rating and price dominate (a top rating raises selection by 31.6 percentage points; a high price lowers it by 30.0), reproducing human valence-and-price primacy but over-weighting eco-certification and ignoring management response. List position -- a content-free artifact -- shifts recommendations causally, worth about $12 per night. Stated reasons track revealed weights imperfectly. The findings ground generative engine optimization and the accountability of AI infomediaries in causal evidence.

Abstract

E-Commerce Search

Explore similar work

Whose hotel does the AI recommend? An algorithm audit of reputation signals in LLM-assisted hotel selection

Does Rank Still Matter? Position Bias When AI Agents Shop on Our Behalf

Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain