Why standard analytics miss AI traffic
When a traveller reads a ChatGPT recommendation mentioning your hotel and then types your name directly into Google or visits your website directly, their visit is logged as “direct” in your analytics. There is no referrer. The booking platform records the booking; the AI that triggered the consideration isn’t visible anywhere in your standard reports.
This is the measurement problem with AI search. You cannot see what you cannot attribute. And most hospitality analytics setups are not designed to surface this signal.
Method 1: Direct query testing (manual, high signal)
The most reliable way to measure AI search visibility is to test it directly. Set up a regular testing cadence: once per week, ask the following queries in ChatGPT, Perplexity and Google AI Overview:
- “Best [category] hotels in [your city]”
- “Where to stay in [your neighbourhood] in [your city]”
- “Good [your specific attributes, e.g. family-friendly, boutique] hotel near [nearest landmark]”
- “[your hotel category] in [your city] under [your approximate price point]”
Record the results: which properties are mentioned, how they are described, whether you appear, and if so how you are described. Do this weekly and track the trend.
This method requires 20-30 minutes per week and gives you direct, unambiguous data. It is the most useful signal available.
Method 2: Dark traffic analysis (indirect, trend signal)
In Google Analytics 4, look at your “direct” traffic over time. A sustained increase in direct traffic, particularly from sessions that show high engagement and go on to book or enquire, is a possible indicator of AI-driven consideration.
This is an indirect signal. Direct traffic increases for many reasons. But combined with Method 1, a correlation between appearing more in AI queries and rising direct traffic provides a useful validation.
One practical test: track direct traffic in the weeks before and after a significant technical implementation (adding schema.org, adding llms.txt). A measurable uptick in high-engagement direct visits 4-8 weeks after implementation is a reasonable signal that the implementation is working.
Method 3: Branded search volume (Google Search Console)
When AI systems recommend a hotel by name, some guests will then search the hotel’s name in Google before booking. Track your branded search impressions and clicks in Google Search Console over time.
An increasing trend in branded search, without a corresponding increase in paid advertising or traditional marketing, is a signal that more people are hearing about your property from a source outside your control: word of mouth, editorial, or AI recommendations.
Building a report ownership will believe
Property ownership and management teams are skeptical of metrics they can’t directly tie to revenue. When presenting AI visibility data internally, use this structure:
Baseline. Record your current AI visibility score (how often you appear in a defined set of weekly test queries) before any work begins.
Activity log. Record every technical implementation: date, what was done, who did it.
Trend line. Show the AI visibility score and the direct traffic trend side by side over 12 weeks after implementation.
Booking correlation. If your PMS or booking system can show direct bookings with no referrer attribution, track these alongside the AI visibility score.
This structure shows a story: before implementation, during implementation, after implementation. It gives skeptical owners something concrete to evaluate.
What to expect from a well-executed AI visibility programme
In the first 4-6 weeks after a complete technical implementation (entity consistency, schema.org, llms.txt, FAQ pages): improved accuracy in how AI describes your property when it does mention it. Fewer wrong or vague descriptions.
At 8-12 weeks: beginning to appear in queries you were previously invisible for, particularly long-tail queries with specific location or attribute intent.
At 6 months: a measurable trend in AI visibility score, and a plausible signal in direct traffic or branded search volume.
These timelines assume consistent implementation and no major changes to the property’s review profile or web presence. They are based on observed patterns, not guaranteed outcomes. AI systems change how they work. The methods described here are current as of mid-2026.