Tracking AI Visibility When There Is No Ranking to Track

Every hotel marketer knows their Google rank. Almost none know their AI rank, and the reason is simple: there is no such number. Assistants do not rank; they assemble. Measuring presence in something that has no list takes a different method, and the method is not complicated.

Start with the honest baseline

Rand Fishkin's clickstream work is the number to hold onto: Google sends less traffic than it did, and ChatGPT sends almost none.

That is not a reason to ignore AI visibility. It is the reason to measure it correctly. The value of being named in an assistant's answer is not the click. It is the mention - the traveller now knows the property's name before they search, before they open an OTA, before they compare. Measuring clicks from assistants will always understate the effect, because the effect happens upstream of the click.

The measurement problem

Two things make this harder than Google.

There is no ranking. An assistant does not return position three. It names two properties in a paragraph, or it does not name yours. Presence is binary per answer.

Most assistants send no referrer. When a user does click through from an answer, the session arrives in Google Analytics with no source, no campaign, no referrer. Analytics files it as direct. Industry estimates put 25 to 40 percent of apparent direct traffic on content-heavy sites as actually AI-referred. The channel exists; it is hiding in the wrong bucket.

Three numbers worth tracking

1. Prompt-set presence. The direct measure. Write twenty to fifty prompts a real traveller would ask: where should I stay in [city] for a long weekend; family-friendly hotels near [landmark]; hotels in [area] with free parking; boutique hotels in [city] that allow dogs. Run them monthly against ChatGPT, Perplexity, Gemini and Claude. Record whether the property is named, and if so, alongside which competitors.

The share of prompts in which you appear is your AI visibility. Its change month to month is the only thing that tells you whether the work is working.

2. Referral sessions from assistants that do pass a referrer. Some do, some of the time - chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com. Build a segment in analytics for those hosts. It will be small. It is real, and its growth rate matters more than its size.

3. Direct sessions landing on deep pages. Nobody types a policy page or a long article URL from memory. A spike in direct traffic to a page like that is an AI referral that lost its referrer. Segment direct traffic by landing page, and watch the deep pages.

What the numbers tell you

If prompt-set presence is low and flat, the property's information is not extractable - the fix is content and structure, not marketing spend.

If presence rises but referrals do not, that is expected. The mention is doing its work upstream; check whether branded search and direct bookings moved.

If deep-page direct traffic grows with no other explanation, the assistants are sending people and the analytics cannot see it. That is a good problem.

Two things not to do

Do not report AI visibility as a rank. There is none, and inventing one - a vendor score, an index - hides the fact that the underlying measure is presence in a specific set of answers. Report the prompt set and the share.

Do not judge by clicks alone. That underweights the channel by design, and it will lead to the conclusion that AI visibility does not matter at exactly the moment it starts deciding which hotels get considered.

What to report

Monthly, to whoever asks: the prompt-set share and its trend, the assistant referral sessions and their trend, and the deep-page direct anomaly. Three lines. Anyone can read them, and together they describe a channel that no single analytics view shows.

The first month's numbers will be humbling for most properties. That is the point of taking them.