How Competing Platform Actors Frame a Disruptive Technology: A Dyadic Analysis of Generative-AI Risk Disclosure by Hotels and Online Travel Agencies

Research Overview

Hotel executives publicly describe generative AI (GenAI) as their route out of dependence on online travel agencies (OTAs), while OTAs treat the same technology as a threat to be absorbed. Disclosure studies that pool structurally different firms into one cross-section cannot see this opposition. This study reads both sides of the hotel–OTA distribution relationship together and asks whether the two frame GenAI differently in their mandatory risk disclosures, and how those frames change over time.

Research Questions

1. RQ1 (themes): Do hotel operators and OTAs frame GenAI through different themes?
2. RQ2 (convergence): Does firm-specific GenAI language persist, or converge toward common boilerplate?
3. RQ3 (stance): Is GenAI framed as a threat or as an opportunity, and how does that stance evolve?

Data

• Sources: Risk-factor sections of annual reports filed with the U.S. SEC (Form 10-K Item 1A; Form 20-F Item 3.D)
• Firms: Four hotel operators (Marriott International, Hilton Worldwide Holdings, Hyatt Hotels, InterContinental Hotels Group) and three online travel platforms (Booking Holdings, Expedia Group, Tripadvisor)
• Period: FY2019–FY2025, one annual report per firm-year
• Corpus: 49 filings (42 on Form 10-K, 7 on Form 20-F); 75 strict GenAI passages (51 OTA, 24 operator)

Methods

• Keyword-anchored extraction of GenAI passages (e.g., “generative AI”, “large language model”, “agentic”)
• Thematic coding under a five-theme codebook, re-coded by an independent model-assisted stage (84.0% agreement, Cohen’s κ = 0.79) with every disagreement adjudicated
• Cross-firm cosine similarity of AI-related language, with a fixed-content control, to track convergence toward boilerplate
• Reproducibility-tested stance coding (threat vs. opportunity); topic modeling used only as exploratory corroboration
• Python 3 with scikit-learn and gensim

Key Findings

Dyadic Differentiation (RQ1)

• Rival-capability framing is exclusive to the OTA side: 11 of 51 OTA passages versus none of 24 operator passages
• Operator language is dominated by deployment-liability and compliance content (58% versus 28% on the OTA side)

Convergence Toward Boilerplate (RQ2)

• Cross-firm similarity rises across FY2023–2025 at fixed content volume: firm-specific language converges toward common legalistic phrasing within about three reporting cycles

Stance Intensification (RQ3)

• Threat framing accounts for roughly two-thirds of stance-bearing passages in FY2023 and almost ninety percent by FY2025, while opportunity framing stays flat

Implications

• Rivals’ annual filings are public, auditable competitive intelligence: OTAs now name supplier-direct and AI-mediated discovery as threats—precisely the opening hotels pursue
• Once AI risk language becomes universal, its mere presence carries little diagnostic value; what a firm says that its peers do not is the informative residual
• The single seven-firm dyad yields illustrative rather than generalizable findings

Publication Details

Journal: Journal of Theoretical and Applied Electronic Commerce Research (SSCI) Volume: 21, Issue 10, Article 339 (published 26 September 2026) DOI: 10.3390/jtaer21100339 Article: MDPI article page Authors: Yonghee Kim (first author), Shi-Shiuan Wang, Sungjin Yoo (corresponding author) Open Access: Creative Commons Attribution (CC BY) license