Royal Caribbean Cruises To An Almost AI Disaster

In 2025, Alex Rivlin, was arranging transportation for an upcoming Royal Caribbean cruise when he searched Google for the company’s customer service number.

The AI Overview gave him an answer. The person on the other end of the number sounded legitimate. They understood cruise terminology. They knew how the shuttle service worked. They quoted a plausible price.

Rivlin handed over his credit card details. The number was fraudulent. He was charged $768 and later discovered additional unauthorized transactions.

For an in-house SEO team, the important question is not only how the scam worked. It is this: When a customer asks an AI system a question about your brand, who is actually answering?

The Royal Caribbean incident was not simply a LLM inventing a phone number from nothing. The system appears to have retrieved false information from the web, treated it as credible enough to use and presented it inside an interface that looked authoritative.

The scammer did not need access to Royal Caribbean’s systems. They only needed to influence the information environment around the brand. That should worry more than the security team; it should worry SEO.

Traditional search results make the evaluation process visible.

A user sees a list of pages. The domain name, title and description provide clues about which result might be legitimate. A suspicious support site can still deceive people, but the user is at least looking at an identifiable source.

AI answers change that presentation layer.

They retrieve information from several pages, reconcile it and compress it into one response. That is useful when the information is accurate. It also means a weak or malicious source can inherit the authority of the entire interface.

A fake phone number on an unfamiliar webpage may look suspicious. The same phone number inside a polished AI answer can look verified.

This is not only a theoretical difference. Research into AI Overviews shows that the sources selected for generated answers can differ substantially from the pages shown in traditional organic results. One 2026 study found that almost 30% of cited AI Overview pages did not appear in the accompanying first-page results. The same study found that 11% of the individual claims it examined were not supported by the cited pages.

That does not mean every AI answer is unreliable. It means teams cannot assume that monitoring ordinary rankings tells them what information is being used inside generated answers. The retrieval system is related to traditional search, but it is not identical to it.

Royal Caribbean did not publish the fraudulent number.

That distinction may matter to lawyers and platform engineers. It matters less to the customer who was trying to contact Royal Caribbean. From the customer’s perspective, the journey began with the brand.

They searched for the brand. They believed they had reached the brand. They lost money while trying to buy a service connected to the brand.

The resulting complaint, support request or social post will still contain the company’s name. That is how brands can absorb the consequences of information they did not create. The immediate costs may include lost transactions, fraud complaints, chargebacks and support escalations. The longer-term cost is harder to measure: customers become less certain that the company’s legitimate contact channels can be trusted.

This is why AI brand visibility is not only a discovery metric. It is part of reputation infrastructure.

There is an easy way to overstate this problem.

Fake support numbers, impersonation sites and poisoned search results existed long before AI Overviews. Scammers have always tried to place false information where people are likely to find it.

Google itself has warned users about fake customer support, fraudulent travel sites and other search-related scams.

Generative AI did not create the underlying attack. It changed how the result is delivered.

With a traditional search result, users are more likely to notice that they are leaving the official domain. With a generated answer, the information has already been extracted from its original context. The user may never inspect the source.

Google says its anti-spam protections are designed to keep scam information out of AI Overviews and to show official support numbers where possible. It has also said it is continuing to strengthen those systems.

Those protections matter. Brands should not pretend platform safeguards do nothing. They should also not treat those safeguards as a substitute for maintaining their own information. Security systems reduce risk. They do not remove the need for a clear source of truth.

Fraud is the clearest version of AI brand hijacking because the intent and harm are obvious.

The broader competitive risk will be harder to classify.

Brands already bid on competitors’ names in paid search. They publish “[Company A] versus [Company B]” pages. They create alternative lists. They sponsor affiliates and review sites. They optimize content designed to intercept a buyer before that buyer reaches a competitor’s website.

Trying to influence AI-generated answers is a natural extension of those strategies. Consider the questions buyers ask close to a decision:

A competitor does not need to invent a blatant lie to influence these answers. A competitor can publish the clearest page explaining what “enterprise-ready” means, then choose criteria that favor its own product. It can emphasize a rival’s weakest contract terms while presenting its own most flexible plan. It can produce an integration page that is technically accurate but conveniently incomplete.

If that content is relevant, current and easy to retrieve, an AI system may cite it.

Controlled research into AI answer engines suggests that topical relevance is one of the strongest factors affecting which retrieved source receives the first citation. Explicit price information and recent timestamps also appear to help, while formatting changes alone have much less effect.

The practical implication is not that competitors have discovered a secret AI trick. It is that useful, specific and current content has an opportunity to define the answer.

A competitor appearing in an AI answer about your product does not prove manipulation. It may have better comparison content. Its documentation may be clearer. It may have more current pricing information. Independent publishers may cite it more often. Its website may answer the actual question while yours offers a vague positioning statement and a demo form.

The uncomfortable explanation may be that another company has done a better job of explaining your market than you have. That is not hijacking, it is competition.

Treating all unfavorable visibility as manipulation will make the monitoring strategy less credible. It will also distract the team from gaps it can fix.

The brands most exposed are not necessarily small or unknown; they are brands with ambiguous information.

A company may have strong brand recognition and still make its support number difficult to verify. It may have different phone numbers on regional pages, old details in partner directories and important support information hidden inside an application. Its product documentation may be several releases behind. Its pricing page may avoid giving direct answers.

Its integration information may be buried inside a JavaScript-heavy interface or a logged-in knowledge base. Its official website may say less about the product than affiliates, reviewers and competitors do.

When that happens, an AI system still has to answer the question. It searches the available information environment. If the official source is unclear, inaccessible or incomplete, another source has more room to become the answer.

No brand can own every generated answer about itself.

Platforms control retrieval, ranking, synthesis and safety policies. AI answers can vary between locations, products and minor changes in wording. Research has also found that AI Overview results can be less consistent across repeated runs and sensitive to small query changes.

Third parties are allowed to criticize your company. Competitors are allowed to compare products. Publishers are allowed to reach conclusions you dislike. Scammers can create new pages faster than most companies can remove them. Total control is not a realistic operating goal. That does not justify ignoring the channel.

Brands can make critical facts easier to retrieve and verify. They can make official support routes unmistakable. They can correct inaccurate listings. They can publish specific product documentation instead of relying on brand language. They can build credible references beyond their own domain.

For critical brand facts, the official website should not require detective work. Contact details should be presented on clear, indexable pages. Essential information should not live only inside PDFs, images, chat widgets or authenticated portals.

Phone numbers, URLs, policies and location information should be consistent across the main website, help center, business profiles, app store listings, partner pages, social profiles and industry databases.

This sounds basic because it is basic; that does not make it optional.

SEO teams often look for sophisticated AI visibility tactics while the company still has three different support numbers attached to the same service.

The systems generating answers do not benefit from organizational explanations. They do not know that one page belongs to an old regional team, another is maintained by customer support and a third is controlled by a distributor.

The Royal Caribbean story looks like a customer service scam. For SEO teams, it should also look like an infrastructure warning.

Your brand is no longer represented only by the pages you publish or the rankings you hold. It is represented by whatever information an AI system can retrieve, reconcile and confidently repeat.

Sometimes that information will come from you. Sometimes it will come from a legitimate competitor with a better answer. Sometimes it may come from somebody pretending to be you.

The brands most exposed will not necessarily be those with the weakest recognition. They will be the brands that assume recognition is enough.

It is not.

Brand authority still has to be maintained. Important facts still have to be crawlable. Claims still have to be corroborated. And the answers customers receive still have to be monitored.

Neural Network: A model inspired by the human brain made of interconnected processing nodes.

Learn More

One more thing: AI is only as good as it’s operator, and if you are reading this newsletter, you are better than most!

Till next time,

PS: Let me know what you think of this issue, or anything else here: [email protected]