Every hotel has a plan for the thing that won’t happen and no plan for the thing that already did.
The plan for the hurricane. The plan for the norovirus. The plan for the fire alarm at 3 a.m. Binders, drills, laminated cards behind the front desk. All of it built on the same honest premise: something will break, and when it does, nobody should be inventing the response on the spot.
That premise is correct. It’s just been applied to the wrong list.
Because there is a failure happening right now, in production, at scale, and it does not appear on anyone’s contingency list. A machine is being asked which hotel to book. It answers. And what it says about your property is assembled from whatever was nearest to hand.
That’s the failure. Not silence. Not absence. Assembly from the nearest available material.
Failure Doesn’t Wait for Your Input
Here’s the part operators keep getting wrong. They imagine the failure mode is that the model doesn’t know about them, says so, and moves on.
Sometimes it does. More often it doesn’t. These systems are built to produce answers, and an answer will usually get produced whether or not the material behind it is good.
When a system is asked to describe your resort and your resort has not supplied a clear, consistent, machine-legible account of itself, the gap gets filled from whatever is available. The OTA listing written by someone who has never been to the property. A review aggregator compressing eleven years of sentiment into one adjective. A press release you issued in 2019 about a renovation that has since been renovated. And separately, a different failure worth naming as its own: the property four miles away that describes itself clearly enough to define the category you’re then placed inside.
Two things go wrong there. You get described badly. And you get positioned by somebody else’s clarity. They compound, but they aren’t the same problem, and fixing one doesn’t fix the other. That second failure is the one most properties never diagnose, because the retrieval-layer signals can all be correct while the property is still absent or misclassified.
Then it produces a fluent, confident, plausible paragraph. And the guest reads it as an answer, not a guess.
That is improvisation. You didn’t authorize it. You just didn’t prevent it.
The Gap Is the Whole Problem
Every system that generates an answer has to fill the space between the question and what it can establish. That space gets filled. The only variable is what fills it.
Plan, and you fill it. Don’t plan, and the market fills it with your competitor’s clarity, your OTA’s copy, and a general sense of your category that flattens you into the middle of it.
This is the entire operating logic of Knowledge Formation Optimization. Not “be seen.” Not “rank.” Fill the gap before something else does.
Which is why visibility, on its own, is the wrong finish line. It’s necessary. It just isn’t sufficient. You can be extremely visible and still be described wrong. You can appear in a dozen answers as the property that “is more casual than its neighbors” when you spent forty million dollars proving otherwise. Visibility says you were mentioned. It says nothing about what was said, and what was said is the entire transaction.
Information Friction Is What Decides It
So the question becomes: what determines whether the gap gets filled with your account of yourself or somebody’s approximation of it?
Friction. How hard your facts are to find, parse, and reconcile against everything else out there.
This is AGR’s operating model, not a claim about anyone’s source code. But it holds up against what we can observe. Information that is unambiguous, non-contradictory, plainly stated and corroborated across multiple independent surfaces is more likely to survive into a confident answer. Information that requires inference, that requires a system to convert your marketing prose into a usable statement of fact, competes badly against information that doesn’t.
Cost, in this sense, isn’t money. It’s how many steps to find the fact. How many conflicting versions to reconcile. How much guessing is required before your claim can be stated as a claim.
A property that says, in plain declarative language, across the sources and surfaces these systems can access, exactly what it is, meaning number of keys, category, location relationship, and what makes it structurally different, is cheap to use. A property that requires all of that to be inferred from adjectives is expensive. Not out of malice. Out of the ordinary mechanics of assembling an answer from incomplete material.
The plan isn’t to be visible. The plan is to be the cheapest correct source available at the moment of formation.
Read that as an operational spec, not a slogan. Cheapest: lowest friction to use. Correct: matching what’s true, so nothing else contradicts it. Available: present in the surfaces these systems can reach, not locked behind a booking engine or buried in a PDF. At the moment of formation: before the question is asked, because after is too late. The answer has already been given and the guest has already moved.
Consistency Is the Discipline That Produces It
The failure most properties are living inside isn’t absence. It’s contradiction.
The website says one thing. The OTA listing says another. The press kit says a third. The sister-property page implies a fourth. Three of the four are technically defensible. All four coexist. And a system trying to assemble a confident sentence about you now has to pick, or worse, average.
Averaging is where luxury goes to die. Every distinguishing claim you’ve made ends up regressed toward the category mean, because the category mean is the only thing all your sources agree on.
Planning here means something unglamorous. It means the same claim, worded the same way, everywhere it appears. It means retiring the 2019 renovation language. It means the number of keys being one number. It means deciding what you are and then not quietly hedging it in four places because four different people wrote four different pages in four different years. This is the work independent luxury properties are structurally worst positioned to do on their own, because there is no brand system enforcing the consistency for them.
None of that is creative work. It’s the corpus equivalent of the laminated card behind the front desk. It’s what you do so nobody has to improvise.
What the Plan Actually Prevents
It does not prevent the question. Guests will keep asking machines where to stay, and the volume of that is going one direction.
It does not prevent the answer. The system does not need your participation to produce one.
It prevents the invention. That’s the whole scope, and it’s enough. You are not trying to control the model. You are removing its need to guess.
The hotels that understand this stop asking whether AI will send them bookings and start asking a much more useful question: when the answer about us is assembled, what is it assembled from? Because that material already exists. Someone wrote it. The only open item is whether it was you.
We don’t plan to fail. Failure is not the contingency. It’s the baseline condition of any system that has to produce an answer from incomplete information.
We plan so failure doesn’t get to improvise.
