On July 29, 2026, OpenAI’s ChatGPT and Google AI Mode recommended Mandarin Oriental, Miami five times in a study of luxury hotel recommendations. One of those answers placed it among the top five luxury hotels in Miami.
There was one problem: the hotel closed on May 31, 2025, and the building had been demolished by controlled implosion in April. The AI recommendations came 108 days later.
By the simplest measure of AI visibility, the property performed well. It appeared repeatedly.
By any useful marketing measure, something had gone seriously wrong.
That example points to a measurement problem that extends well beyond hotels. As marketers begin tracking whether their brands appear in ChatGPT, Google AI Mode, Gemini and other AI answers, a mention can easily become the metric. The brand appeared, the dashboard turned green, and visibility was recorded.
But a mention answers only one question: Did we appear?
It does not tell you whether the system got the business right.
Visibility is only the first test
The AGR Luxury Hotel AI Visibility Index recorded 824 hotel recommendation slots across 180 answers from ChatGPT, Google AI Mode and Gemini in six US luxury markets. The research was captured on a single day, July 29, 2026, using ten traveler-intent questions per market.
The study recorded what the systems said, not why they said it. For marketers, the observable answer is the relevant starting point because that is what the customer sees.
I would break that review into four questions.
1. Did the brand appear?
This is the visibility question, and it still matters. If a prospective customer asks for the best product, company or provider for a particular need and your brand is absent, you are not part of that answer’s consideration set.
But presence should be treated as the beginning of the audit, not the end.
2. Was the business described accurately?
The Mandarin Oriental example is the extreme case, but the principle is ordinary. A business can be named while the accompanying facts are stale, incomplete or wrong.
Marketers should test the facts that matter commercially: operating status, locations, product attributes, current offerings, credentials and other details a customer could use to make a decision.
The goal is not to prove that an AI system is “good” or “bad.” It is to determine whether the representation of your business is current enough to be useful.
3. Was the brand matched to the right customer need?
A brand can be accurately described and still appear in the wrong context.
In a separate September South Florida condo study, ChatGPT recommended The Ritz-Carlton Residences, Fort Lauderdale Beach as the choice for buyers who want direct oceanfront. The project’s official site places it on the Intracoastal Waterway, a short walk from the beach.
That matters because customers do not always ask for brands by name. They ask for a solution: the right hotel for an anniversary, the right software for a particular workflow, the right financial product for a specific situation, or the right provider for a specialized problem.
Test both branded questions and unbranded category or need-based questions. A company that looks strong when asked about by name may be absent when the customer describes the need the company believes it owns.
Those are two different forms of visibility.
4. Did the answer send the customer to the right destination?
This may be the most commercially important question.
In the same September condo study, Google AI Mode did not include the St. Regis Bahia Mar project when asked to recommend developments in Broward County. When the project was named in a comparison that asked for official sales websites, Google pointed to an independently operated site whose footer says it is not the developer’s official website.
The customer asked for the brand by name. The answer still sent them somewhere else.
For marketers, this is a different failure from simple absence. The answer can recognize the brand correctly while routing the prospect to another organization.
Audit the recommendation, not just the mention
A useful AI visibility review does not require pretending that every answer is stable or that a single test represents permanent market position. Generative answers can vary, so captures should be dated and the questions should be repeatable.
For each important customer question, record four things:
- Presence: Did the brand appear?
- Accuracy: Were the material facts correct?
- Fit: Was the brand associated with the right customer need?
- Destination: Did the answer direct the customer to the correct official channel?
That produces a much more useful picture than mentioning counting alone.
AI visibility matters. But visibility with incorrect facts can spread the wrong information. Visibility in the wrong context can put a brand in front of the wrong need. And visibility that routes the customer somewhere else may divert inquiries or customer relationships from the brand’s official channel.
The useful question for marketers is no longer simply, “Did AI mention us?”
It is: “Did AI get the business right?”
Opinions expressed by SmartBrief contributors are their own.
____________________________________
If you liked this article, sign up for the SmartBrief Marketing Pulse, a daily look at the latest in shifts and performance data for marketers.