Who Owns the Recommendation?

This is the second essay in an ongoing series exploring the Recommendation Economy—how recommendations are becoming the new organizing principle of the internet, and how brands, publishers, creators, and consumers will adapt. To start at the beginning, read The Mapmaker's Paradox.

How does one brand become the recommendation while another does not?

"We're driving to Moab this fall with our two Bernese Mountain Dogs. Can you recommend great hiking trails, restaurants with dog-friendly patios, and places to stay that genuinely welcome large dogs—not just small pets?"

Emily and Jake's two dogs weigh more than two hundred pounds combined. Every fall they load the SUV and head west for a long weekend. This year they're planning to explore Moab. They want to hike before sunrise, spend long afternoons among the red rock canyons, and end each evening somewhere their dogs are welcomed rather than merely tolerated.

Within seconds, they have an itinerary: trails to hike, patios where the dogs can stretch out beside the table, and a hotel that appears to fit what they're looking for.

For her query,  the LLM scans several sources, including BringFido, Discover Moab, and the hotel's own website. Each contributes something different to the answer. The niche blog ensures the response aligns with a dog-friendly trip, while the local tourism board's travel tips provide the experiences the couple is looking for. The hotel website, meanwhile, confirms the pet policy, completing a holistic recommendation. 

Recent research is beginning to put numbers around how those pieces contribute. In one 2026 randomized audit across twelve AI models, raising a hotel's guest rating from 3.9 to 4.7 stars increased its probability of being recommended by 31.6 percentage points. Higher review volume added 8.3 points. Price, eco-certification, and even where a hotel appeared in a list also changed the outcome.¹

No single signal determined the recommendation.

The confidence comes from the evidence accumulating around it.

What happens when the traveler becomes more specific is even more revealing.

Another 2026 hospitality study audited 1,357 Gemini citations across 156 hotel queries. Experiential searches drew 55.9 percent of their citations from non-OTA sources, compared with 30.8 percent for transactional searches.² The traveler wasn't simply getting a different answer. The nature of the question was changing the evidence used to construct it.

For Emily, that means each turn in the conversation can alter the evidence that matters. Two dogs weighing more than two hundred pounds make a hotel's pet policy unusually important. A question about hiking before sunrise makes location and trail access more relevant. As Google and OpenAI incorporate more remembered preferences, prior interactions, and personal context into responses, the number of meaningful combinations grows further.³⁻⁴

Whether it's a hotel that welcomes large breeds, a resort with a dedicated fryer for guests with celiac disease, or a beachfront hotel within walking distance of beginner surfing lessons, each request creates another context in which a brand can be recommended.

There may be tens of thousands of those contexts for a single destination or category, and potentially far more as recommendations become increasingly personal. The important shift isn't simply that AI can understand more specific questions. It's that those combinations of intent are becoming an organizing principle for discovery.

That creates a very different problem for brands.

A hotel might be consistently recommended for "dog-friendly Moab hotels" and almost invisible once the traveler adds "two dogs over 100 pounds." Perhaps its website never states its weight policy. Perhaps a specialist publisher has the policy wrong. Perhaps reviews contain the answer, but not with enough consistency to establish it. Or perhaps the absence is accurate: the hotel doesn't actually provide what that traveler needs.

Those are very different problems.

And knowing which one you're dealing with matters more than simply knowing that your hotel wasn't recommended.

At that scale, the recommendation itself is only the starting point. Brands need to understand the specific traveler intents that matter to their business, where they are and aren't being recommended, and the evidence surrounding those outcomes: which sources appear, what each contributes, where information agrees, where it conflicts, and where important evidence is missing.

Then they have to understand that across multiple recommendation systems.

A 2026 analysis of more than 161,000 prompts across ChatGPT, Gemini, Perplexity, and Google AI Overviews found remarkably little agreement among them. Among prompts in which all four systems returned citations, only 3.8 percent of cited sources were shared by all four. Roughly three-quarters of cited domains appeared in only one.⁵

There isn't one recommendation ecosystem to understand.

There are several, and they're changing at the same time.

That makes questions like whether Reddit or YouTube "matters" increasingly difficult to answer in isolation. A source can matter enormously for one kind of intent in one model and considerably less for another. What matters is understanding the relationship between the traveler, the recommendation, and the evidence the system appears to trust in that particular context.

Recent events show how unsettled even those relationships remain.

In August, Time began placing FAQ-style advertising from brands inside stripped-down versions of its articles designed for AI agents to ingest. If AI systems increasingly consume publisher content without sending a human reader to the page, the information presented to the machine itself suddenly becomes valuable real estate.⁶

Perplexity responded by blocking the ads from influencing its search index, describing the practice as deceptive and warning that publishers employing it could see their trust scores reduced.⁶

The significance of the episode isn't whether Time or Perplexity has the better argument. It's what the collision reveals.

Brands are experimenting with how to influence the answer. Publishers are experimenting with how to capture the value of the authority and knowledge they provide. AI companies are deciding which forms of influence they'll accept and which they'll suppress.

The rules are still being written.

Which is why the emerging challenge is considerably larger than optimizing content for an LLM.

Imagine trying to understand this across a hotel portfolio: tens of thousands of specific traveler intents; recommendations changing across ChatGPT, Gemini, Perplexity, Claude, and whatever comes next; evidence distributed among publishers, specialist sites, reviews, communities, and owned properties; source authority shifting by question and model; information becoming stale or contradictory; and the systems interpreting all of it continuing to evolve.

Simply measuring whether the brand appeared in an answer tells you very little.

The more useful questions are why it appeared, why it didn't, what evidence influenced the outcome, and what could credibly change it.

If the hotel genuinely welcomes giant dogs but no reliable source makes that clear, there is an information problem to solve. If respected publishers have never experienced or written about the property, there may be an earned-media opportunity. If the hotel's own information contradicts what guests report, there is a different problem entirely. And if travelers increasingly ask for something the property cannot deliver, no amount of optimization should make it the right recommendation.

The goal isn't to manufacture the evidence an AI system wants to see.

It's to understand the evidence universe around the brand well enough to know where the opportunity actually lies.

Doing that at scale begins to look less like another marketing channel and more like a new layer of intelligence across the business: continuously connecting traveler intent to recommendations, recommendations to evidence, evidence to sources, and those sources back to actions the organization can take.

No single function owns the recommendation.

Marketing and communications may influence what credible third parties say about a property. Digital controls much of what the brand says about itself. Revenue management influences the product, price, and availability a traveler can actually book. Operations determines whether the described experience is true. Guest experience creates much of the evidence that eventually appears in reviews and elsewhere across the web.

The recommendation system doesn't recognize those organizational boundaries.

It sees what they collectively leave behind.

Earning the recommendation, then, becomes a responsibility that crosses the business.

The drive home is quiet.

The dogs are asleep in the back seat, exhausted after three days of hiking. Emily laughs about the man they met on the trail whose tiny terrier seemed determined to keep pace with dogs ten times its size. Jake jokes that he'll never forget the look on the terrier's face when it reached the summit.

The conversation turns to next fall.

Maybe Glacier.

Maybe Acadia.

Somewhere they've never been.

Before long, another question will begin another journey.

Somewhere else, someone they'll never meet is already publishing a story, documenting a trail, reviewing a hotel, or sharing an experience that a recommendation system may one day weave into an answer.

That, perhaps, is the next story.