How the Observatory observes AI-mediated property discovery, constructs conversational prompts, and maps findings to VPR schema.
The Observatory is an observational research project focused on AI-mediated hospitality discovery. We analyze how AI systems currently respond to hospitality queries to increase transparency in conversational discovery. We do not provide recommendations for manipulating AI responses or gaming discovery systems.
We document how properties surface across conversational prompts and identify recurring representation patterns.
We map recurring representation patterns to VPR fields to understand structured data correlations.
We analyze visibility compression inside conversational interfaces—the shift from hundreds of options to a handful of recommendations.
We provide research intelligence that informs VPR creation and optimization decisions.
We do not rank properties or declare winners. We observe what surfaces, not what should surface.
We do not provide recommendations for manipulating AI responses or gaming discovery systems.
Our observations are based on specific prompt sets. AI behavior varies across queries and systems.
We document observed patterns, not predictions of future AI model behavior.
We analyze conversational discovery, not actual bookings, revenue, or conversion data.
A property that appears in an AI response to a conversational prompt. This is an observation of visibility, not an endorsement of quality.
A property that an AI system explicitly recommends or presents as a best match for a user query.
A property, attribute, or signal that appears repeatedly across multiple prompts or AI responses.
A structured data element or property characteristic that correlates with AI surfacing in our observations.
The degree to which observed surfacing patterns correlate with specific VPR (Verified Property Record) schema fields.
The extent to which a property appears in AI-mediated discovery flows versus traditional search listings.
The percentage of AI evaluations where a property is identified as a relevant match for user intent.
The phenomenon where conversational interfaces reduce visible options from hundreds of search results to a handful of AI recommendations.
The Observatory uses timeless language when referring to AI systems to ensure research remains relevant as technology evolves. We avoid hardcoding specific model names in public copy.
We construct representative conversational prompts that reflect real traveler queries. Each scenario includes prompts covering location specificity, amenity requirements, service availability, timing flexibility, and transportation access.
We submit prompts to leading commercial conversational AI systems and collect the properties that appear in responses. Properties are counted, categorized, and analyzed for recurring patterns.
Our research focuses on conversational discovery as a phenomenon, not on specific AI models. We analyze how AI systems in general respond to hospitality queries.
We map observed conversational patterns to VPR (Verified Property Record) schema fields. This helps identify which structured data elements correlate with property surfacing.
Each benchmark includes collection period, last reviewed date, and update cadence. We acknowledge that AI behavior evolves and findings may change over time.
Each benchmark uses categorized prompts representing different traveler intents and contexts.
Prompts are constructed to represent realistic travel planning conversations, not technical queries.
Each scenario represents a distinct travel persona and intent pattern observed in real conversations.
Prompts include decision-context elements that influence property selection.
All data is collected through direct interaction with AI conversational interfaces using the prompt sets documented in each benchmark. We do not scrape AI systems or use automated crawling methods.
Property data (names, locations, amenities mentioned) is derived from AI responses and cross-referenced with publicly available information. We do not claim that all properties mentioned have VPRs in the HomeSelf registry.
For VPR availability status, we reference the public HomeSelf registry at the time of benchmark publication.
Download complete benchmark briefs with all prompt sets, surfaced properties, VPR alignment analysis, and machine-readable markdown format.