A luxury property will approve $40,000 for a macro travel creator to post about it once. Deliverables get contracted, usage rights get negotiated, reach gets reported. Almost nobody builds a revenue model first. The post runs, the engagement numbers get screenshotted, and the line item gets approved again next year.
That same property will ask an AI visibility vendor to prove a return before signing a $5,000 monthly engagement. Fair enough. So this page shows its arithmetic, which is more than most of your marketing budget can say for itself.
The short answer: for many independent luxury properties at real rate levels, yes, and the floor is lower than the pricing suggests. At a $5,000 monthly fee, on a property where 65 percent of business comes from travelers who did not already know your name, the floor is roughly $10.3M in annual room revenue. Below that, do not buy. The general formula and the stress case are in the model section.
But the number is the smaller half of this. Start with the test.
The ten-minute test
Open ChatGPT, Perplexity, and Google AI Mode. Ask each one for the best luxury hotels in your market. Ask twice more with different phrasing. Then ask each one directly what it knows about your property.
Write down two things. Which properties get named. And what the systems actually say about yours.
There are three outcomes, not two.
- Competitors named, you absent. An acquisition gap. The floor check below tells you whether the spend is proportionate.
- Named, and described the way you would describe yourself. A maintenance question. Protect the position. Do not buy an acquisition engagement.
- Named, and characterized in a way that talks the traveler out of you. The worst of the three, and the one nobody tests for.
The third outcome is worse than absence
These systems assemble a description of your property from whatever material is available. When your own corpus is thin, the loudest remaining source wins. Sometimes that is an OTA blurb written for a room type you stopped selling in 2019. Sometimes it is a competitor’s comparison page. Often enough it is a single review complaining about a wrinkled pillowcase, because a specific grievance is more quotable than forty generic five-star ratings.
Here is what that does. A traveler asks which luxury hotels in your market are worth considering. Your property gets named, followed by a qualifier: dated rooms, inconsistent service, guests report road noise. The traveler crosses you off and books elsewhere. They never see your website. Your photography, your rate strategy, your booking engine, your review responses, none of it gets a turn, because the decision happened upstream of all of it.
Both outcomes cost you, but not in the same way. An absent property loses consideration: the traveler does not know it exists in this channel, and can still reach it through a search, an agent, or a friend. A negatively characterized property gets active rejection, delivered on the traveler’s behalf by a system they trust more than an ad, and it does not get a second look. Absence costs you the chance to compete. A bad characterization costs you the competition itself.
And it reaches past discovery. A guest who found you through a referral, a repeat guest checking whether the property changed hands, a planner vetting you for a group: any of them may run the same query before committing. A bad characterization intercepts demand you already earned through channels that have nothing to do with AI.
This is not a marketing question and it does not belong in a return model. It is a defect. If your property management system reported the wrong room count to every OTA, nobody would ask for the ROI of correcting it. This is the same category of problem, sitting one layer further upstream, and it is worth fixing at any property size. A 40-room inn that fails every threshold on this page should still fix a bad characterization, because that repair defends demand it is already winning. It is scoped work with an end date, not a retainer.
Why this is a live question now
Phocuswright tracked US travelers using general search for trip research dropping from 51 percent in late 2024 to 36 percent by the second half of 2025, while generative AI platforms went from 6 percent to 15 percent over the same period. Its more recent work puts gen AI platform usage for trip research at 33 percent, roughly five times the 2024 level, and describes AI as nearing parity with general search rather than having passed it. Search is still ahead. The gap is closing quickly.
Separately, Cloudbeds’ 2026 State of Independent Hotels Report, compiled from 90 million bookings across 180 countries, puts OTA share of independent hotel bookings at 63.4 percent globally, up two points year over year. North America sits at 52.7 percent. EMEA reaches 76.5 percent, with some markets approaching 80 percent. Those are distribution figures from booking data, not traveler survey research. The two sources should not be blended.
Discovery is moving to a channel where you are either named or you are not, at the same time distribution costs are climbing. Those two trends make the same argument from opposite directions, and the shift itself has a precedent worth reading.
What the recommendations actually look like
Across 824 AI hotel recommendations captured in six US luxury markets on July 29, 2026, the distribution was not close to even. The full capture set and technical audit carry the detail behind these figures.
| Finding | Value |
|---|---|
| Total recommendations captured | 824 |
| Distinct properties named | 152 |
| Share taken by the top 25 properties | 53.4% |
| Properties recommended exactly once | 49 of 152 |
| Query sets where three systems named a different lead property | 70.0% |
| Audited properties publishing no lodging structured data | 46.7% |
| Properties with retrievable robots.txt issuing a full GPTBot disallow | 2 of 111 |
Twenty-five properties out of 152 took over half of everything. Forty-nine were named exactly once. That is what an unworked market looks like, and it is the reason the opportunity exists at all.
The three systems named a different lead property in 70 percent of comparable query sets. There is no single winner to displace. You are working three surfaces, and a property strong on one is often absent from another. What to do at each of them is a separate question with a longer answer.
Two properties out of 111 with retrievable robots.txt issued a full GPTBot disallow. Crawler blocking is a common vendor diagnosis and this sample does not support it as the general explanation. Nearly half published no lodging structured data, a real gap, but the audit did not explain who got recommended. We draw no causal conclusion from these captures. They record co-occurrence, and anyone telling you otherwise is selling past their evidence.
The floor check
This model answers one question: is the spend proportionate to the property, or absurd for its size. It does not forecast return, and no honest version of it could. Treat it the way you would treat addressable market on an investment memo.
200-room independent luxury property, $5,000 monthly fee.
| Input | Value | Type |
|---|---|---|
| 200 rooms, 70% occupancy, $650 ADR | $33.2M room revenue | Derived |
| Ancillary uplift on captured demand | 1.35x | Assumption |
| Total demand-linked revenue | $44.8M | Derived |
| Share that is discovery-dependent | 65% | Assumption |
| Share of discovery originating in AI | 5% | Assumption |
| Revenue touched by AI-assisted discovery | $1.46M | Derived |
| Target capture share in a five-property set | 20% | Assumption |
| Modeled monthly revenue at stake | $24,290 | Derived |
| Ratio to a $5,000 fee | 4.9x | Derived |
Three of those assumptions deserve a note. Ancillary revenue belongs in because a captured room night at a resort brings food and beverage, spa, and event spend with it. Discovery-dependent share belongs in because repeat guests and standing relationships were never in play, and leaving it out is the most common way these numbers get inflated. Capture share at 20 percent is an even split of a five-property set. That is the target state after the work succeeds, not a starting condition. Non-concentrated properties in our capture set currently run at 56 percent of an even rate, which is 11 percent.
Both numbers matter, so the threshold is published at both. At 20 percent the engagement has to work. At 11 percent it does not, and the property clears anyway on the rate it already gets. Any vendor showing you only the first number is showing you the easy one.
Require a 1.5x ratio before engaging, because the AI-share assumption is the one nobody in this industry has published a defensible figure for, including us. A 1.5x margin means the decision survives that being wrong by a third.
| Monthly fee | Floor at 20% capture (work succeeds) | ||
|---|---|---|---|
| 50% discovery | 65% discovery | 80% discovery | |
| $4,000 | $10.7M | $8.2M | $6.7M |
| $5,000 | $13.3M | $10.3M | $8.3M |
| $6,000 | $16.0M | $12.3M | $10.0M |
| $7,500 | $20.0M | $15.4M | $12.5M |
| Monthly fee | Floor at 11% capture (stress case) | ||
|---|---|---|---|
| 50% discovery | 65% discovery | 80% discovery | |
| $4,000 | $19.4M | $14.9M | $12.1M |
| $5,000 | $24.2M | $18.6M | $15.2M |
| $6,000 | $29.1M | $22.4M | $18.2M |
| $7,500 | $36.4M | $28.0M | $22.7M |
Qualifying room revenue equals monthly fee times 1,333, divided by your discovery-dependent share. In the stress case the multiplier is 2,424. A 100-room property at $550 ADR clears the first. A 200-room property at $650 clears both. Room count is not the test.
That gives three bands rather than a line. Below the 20 percent floor, do not buy at that fee. Between the two floors, the spend depends entirely on the engagement performing, so do not sign it without a measurement gate that lets you leave. Above the stress floor, the arithmetic holds even if the work only maintains the rate you already have, which is the only band where this is a comfortable decision.
Channel economics sit on top of this and are not counted in it. At the North America OTA benchmark of 52.7 percent and a 20 percent blended commission, the AI-attributable slice of commission spend on that same property is roughly $114,000 a year. At half conversion to direct, that alone nearly covers a $5,000 fee. The floor check ignores it, which is why the floor check is conservative.
What you already spend without a model
Now put the number in context, because a threshold in isolation is not a decision. Here is the comparison against the line item most luxury properties fund without hesitating.
| Mid-tier travel creator, one post | Macro travel creator, one post | AI visibility, one year | |
|---|---|---|---|
| Cash fee | $3,000 to $15,000 | $12,000 to $60,000 | $48,000 to $90,000 |
| Plus comped stay | 2 to 5 nights, $300 to $2,000 per night | 2 to 5 nights, plus travel | None |
| Luxury niche premium | 20% to 50% above lifestyle rates | Same | Not applicable |
| Attribution | Reach and engagement, booking attribution rarely isolated | Reach and engagement, booking attribution rarely isolated | Prompt-level capture, before and after |
| Asset after payment stops | Reusable content, if you licensed it | Reusable content, if you licensed it | Corpus, entity resolution, citations |
Three macro posts across a year runs $36,000 to $180,000 in cash, before comped room nights. That is not an argument against influencer marketing. Plenty of properties get real value from it, the reach is real, and licensed content has uses well beyond the original post.
It is an argument about the standard of proof. That spend gets approved on judgment and relationship. This one gets asked for a model. If you are going to demand arithmetic from a vendor, demand it from the whole budget or from none of it.
There is also a difference in kind, and it matters more than the price comparison. A campaign buys attention for a period. Some of it leaves an asset behind and some does not. What sits underneath this line item is different again: whether your property resolves correctly as an entity, and whether the account these systems give of it is true. Those are conditions of the asset. They are true or false about your property today, whether or not anyone is funding them, and they stay wrong until someone corrects them.
Where you rank in a recommendation set is not like that. Positions move, models update, and our own captures show three systems disagreeing with each other on the same day. Do not let anyone sell you ranking as though it were infrastructure. Accuracy is the durable part.
Nobody computes the ROI of a roof repair. They ask what it costs to have it wrong.
What a retainer buys that a one-time cleanup does not
A fair objection, and one we get: if this is a data defect, fix the data once and stop paying. For a good number of properties that is exactly right, and it is why the characterization repair above is scoped work with an end date rather than a subscription.
Here is the honest boundary. One-time work covers structured data, profile accuracy, name and entity consistency, and correcting whatever wrong account is currently in circulation. Do that, verify it, and stop. If your ten-minute test only turned up inaccuracy, you do not need a retainer and nobody should sell you one.
Ongoing work is for the other problem: not being in the set at all. Getting named requires source material that did not previously exist, published where these systems look, and corroborated somewhere other than your own domain. That is production, not repair, and it does not finish in a month. Models also refresh, competitors publish, and your position is measured against theirs rather than against a standard. The retainer pays for that production and for the measurement that tells you whether it worked.
If a vendor cannot draw that line for you, and cannot say which part of your engagement is finite repair and which is ongoing production, they have not thought about it or they do not want you to.
Do not buy this if any of these are true
- Your room revenue is below the threshold for your fee and mix. Ask for a smaller scope, or fix the characterization problem alone and wait on the rest.
- You cannot measure your baseline. Any vendor who will not measure before selling is selling on fear.
- Your basic digital footprint is broken. Stale Google Business Profile, no lodging structured data, your property name rendered three different ways across the web. That is developer work at developer rates. Do not pay retainer rates for it, and know the difference between a report and an audit before you buy either.
- Your direct channel cannot absorb the demand. If your booking engine converts badly, you are paying to raise awareness that an OTA monetizes.
- You are being sold a twelve-month minimum with no measurement gate. No.
How to tell a real engagement from a repackaged SEO retainer
The category has earned skepticism. Published reporting has documented contradictions between some AI visibility vendors’ claims about their own measurement methodology and publicly available information about their funding and operations. You do not need to resolve that to protect yourself.
It helps to know which of the three functions you are actually buying: measurement, retrieval optimization, or formation. Most of what is sold covers the first two.
Real engagement: measures a baseline before selling using disclosed prompts and a repeatable method. Shows you raw captures, not a proprietary score. Names the specific entity problems on your property. Quotes a fee before quoting a threshold, and asks about your repeat business before either. Commits to a measurement gate at 90 days with an exit if it fails. Tells you what it cannot do.
Repackaged SEO retainer: leads with a proprietary index number and will not explain the inputs. Diagnoses crawler blocking without checking your robots.txt. Presents a modeled opportunity as measured revenue. Applies AI discovery share to your total revenue rather than your discovery-dependent revenue. Sells “AI optimization” as a bolt-on to an existing SEO contract. Cannot tell you which prompts it monitors. Reports rankings instead of citations. Twelve-month minimum, no gate.
Seven questions to ask before you sign
- What is my current visibility baseline, and how did you measure it?
- Which specific prompts are you monitoring, and can I see the list?
- What do the raw model responses say about my property right now, is any of it wrong, and where is each claim sourced from?
- Show me your model for my property, with every assumption labeled and both your fee and my discovery-dependent share as inputs.
- Which of your numbers are measured and which are modeled?
- What is the 90-day measurement gate, and what happens if we miss it?
- What can you not fix?
A vendor who answers all seven in writing is worth evaluating. A vendor who deflects on three, four, or five has not done the work.
The decision
Run the ten-minute test. It sorts you into one of three positions, and the third one splits again on economics.
Described wrongly. Fix it now, at any property size, on any budget. This is not a marketing decision and no threshold applies to it. Scope it as a repair, get an end date, pay for it once. If a vendor answers a factual-accuracy problem with a monthly retainer, you are talking to the wrong vendor.
Named and described accurately. You have already won something most properties in your market have not. Find out what is holding it up before someone changes it for you, and do not buy an acquisition engagement to defend a position you already hold.
Absent. This is where the arithmetic decides, and it gives three answers rather than one. Take your quoted fee and your discovery-dependent share, and find both of your floors.
- Below the success floor. At $5,000 and a 65 percent mix, below $10.3M. Do not buy a retainer at that fee. Fix the foundation in-house, correct anything wrong in how you are described, and revisit when rate or scale moves. A vendor who tells you otherwise is not screening.
- Between the two floors. At $5,000 and 65 percent, between $10.3M and $18.6M. The spend only works if the engagement actually moves your capture rate, so treat it as a gated pilot rather than a program. Ninety days, disclosed prompts, a pre-agreed success criterion, and a clean exit if it misses. Do not sign anything you cannot leave.
- Above the stress floor. At $5,000 and 65 percent, above $18.6M. This is the only band where the arithmetic clears even if the work merely holds the rate you already get. Still take the 90-day gate, but here the decision does not depend on the optimistic assumption being right.
The thresholds on this page exist to keep anyone from selling you an engagement your property cannot carry. They are floors, not forecasts. None of this proves a retainer will pay back, and no honest version of it could. What it does is define the conditions under which paying for this becomes a rational thing to test, which is a higher standard than most of the budget clears.
How these systems describe your property is a condition of the asset, not a campaign. It is true or it is false today, whether or not anyone is funding it, and a growing share of your future guests will see whichever it is. Go find out which.
Americas Great Resorts has operated in luxury hospitality demand infrastructure since 1993 and publishes the AGR Luxury Hotel AI Visibility Index. We sell the services this page evaluates. The disqualifying conditions above are real and we apply them.
