What the paper studied
Grounded in Social Presence Theory and the Stimulus–Organism–Response (SOR) framework, the authors ran one field study (restaurants) and three scenario-based online experiments (hotels, railways, airports) to test how an AI service robot's conversational style — emotional (warm, empathetic, enthusiastic) versus rational (neutral, concise, task-oriented) — influences guests' purchase intentions. They examined engagement as the mediating psychological mechanism and robot persona realism (humanlike vs. less humanlike appearance) as a moderating boundary condition.
Key findings
- **Emotional > rational for purchase intent.** Across all four studies, emotional conversational styles produced significantly higher purchase intentions than rational styles (Field Study: M = 5.27 vs. 4.51; effects replicated in hotels, railways, airports).
- **Engagement is the mechanism.** Engagement partially mediates the link from conversational style to purchase intention — emotional cues drive cognitive, affective, and behavioral involvement, which then drives bookings.
- **Style must match robot appearance (the key insight).** Robot persona realism flips the optimal style:
- **Highly humanlike robots** → emotional style produces higher engagement (M = 5.3 vs. rational 4.6).
- **Less humanlike / kiosk-style robots** → *rational* style produces higher engagement (M = 5.3 vs. emotional 4.8).
- **Moderated mediation is significant.** The indirect effect of style on purchase intent through engagement is roughly 2.7× stronger with humanlike robots (0.64) than with less humanlike ones (0.24).
- **Real environments amplify effects.** The field study showed a larger effect (F = 15.01) than the online replication (F = 6.45), suggesting that in-context immersion strengthens emotional cue impact.
- **Identical content, different framing.** All stimuli held service information constant — only linguistic style varied — meaning the effects come from communicative framing, not what was said.
Why it matters for hospitality
Hotels, airports, restaurants, and OTAs are deploying chatbots and service robots at speed, often defaulting to either a polite-warm tone or a clipped-efficient tone without testing which fits the embodiment. This paper says the choice is not aesthetic — it directly affects bookings, and the wrong pairing actively hurts engagement. A humanlike concierge robot or avatar should sound warm and empathetic; a tablet kiosk or text-only chatbot should sound concise and task-focused. Pairing a stick-figure chatbot with overly emotional language reduces engagement just as much as pairing a humanlike avatar with cold transactional text.
Practical takeaways
- Audit every guest-facing AI touchpoint (front-desk robot, concierge avatar, booking chatbot, kiosk, in-room voice assistant) and classify it as **higher** or **lower** persona realism.
- For **humanlike avatars or anthropomorphic robots**: write scripts that are warm, empathetic, enthusiastic, and personalized — emojis and conversational affect markers help.
- For **kiosks, text-only chatbots, and minimalist UIs**: write scripts that are direct, neutral, task-focused, and information-dense — skip the warmth padding.
- Treat **engagement** (immersion, time-on-task, willingness to continue) as a leading KPI for AI service tools; it predicts purchase intent.
- Avoid expectancy violations: an emotionally gushing kiosk or a robotically terse humanoid will both underperform.
- For high-realism robots, test for the uncanny valley (perceived eeriness) — the congruence benefit only holds when persona realism stays in a non-aversive range.
- Build adaptive scripting that escalates from rational/task-focused to emotional/relational when the interaction shifts from informational lookup to a decision-laden moment (booking, upsell, complaint).
- Treat emotionally persuasive AI as a regulated design surface — disclose AI use, protect privacy, and avoid manipulative framing as part of responsible-AI adoption.