The following is a cleaned transcript of a ChatGPT conversation conducted on September 20, 2026 in a logged-out, incognito session. The session began with a review of AGR’s Demand Origin trilogy, then expanded to the AGR GitHub repository, hotel marketing case studies, and the KFO testable-application and draft falsification documents. Interface citation markers, repeated source-card labels, one embedded advertisement, and one unanswered duplicate submission were removed. No substantive prompt or response was rewritten. This is a historical AI-generated assessment record, not independent validation or proof of AGR’s frameworks.
Part One: Reviewing the Demand Origin Trilogy
User: Can you provide me your opinion on this article and the three referenced articles referenced? Are the intelligent, on point, make sense and worth reading. What is your overall view and take from this? https://www.americasgreatresorts.net/demand-origin-trilogy/
ChatGPT: Yes. I read the trilogy itself, including all three underlying articles, rather than just the summary page. I also cross-checked the central economic claims against current hospitality/distribution material.
My overall take is: there is a genuinely intelligent and useful strategic idea here, and the trilogy is unusually coherent—but it is also much more assertive than the evidence currently warrants in several important places. I would absolutely read it if you are interested in hotel distribution, luxury hospitality, direct demand, or the future impact of AI. I would not treat it as established economic doctrine or accept all of its conclusions at face value.
My overall assessment
If I had to characterize it in one sentence:
The trilogy contains a strong strategic thesis wrapped in an occasionally over-engineered economic proof, and its strongest insight is considerably more convincing than some of the claims used to prove it.
The strongest insight is the distinction between where a hotel gets a booking and where the guest relationship and demand actually originated.
That is genuinely important.
The weaker part is the leap from:
“OTAs possess information and relationship advantages that hotels often don’t”
to:
“This is essentially a lemons problem, the industry is trapped in a particular equilibrium, and only this particular three-layer architecture can solve it.”
That second proposition requires considerably more empirical evidence than the articles provide.
1. Part 1 — The Lemons Problem
The Lemons Problem: How Asymmetric Information Destroyed Luxury Hotel Demand
This is probably the most intellectually interesting article of the three.
The central observation is that OTAs don’t merely sell hotels distribution. They sit between the hotel and the traveler and consequently accumulate enormous amounts of information about the demand environment.
That’s a real phenomenon.
An OTA potentially sees things an individual hotel doesn’t:
what properties a traveler considered;
what alternatives were viewed;
relative prices;
dates searched;
whether the traveler converted;
what inventory was available;
what types of travelers respond to particular offers;
behavior across multiple properties rather than one property.
The article’s line of thinking that the hotel receives the guest while the intermediary accumulates the broader demand intelligence is quite compelling.
And current industry data supports the broader premise that OTA dependence remains significant. Cloudbeds’ 2026 independent-hotel report, for example, puts OTA bookings at 63.4% versus 36.6% direct bookings globally among its dataset of independent hotels. In the U.S., its reported split is considerably less OTA-heavy, illustrating why the problem isn’t uniform across markets.
Where I think Part 1 overreaches
The Akerlof analogy is clever, but I don’t think the article has completely demonstrated that this actually is a “lemons problem.”
Akerlof’s original problem concerns quality uncertainty and the resulting adverse-selection dynamics.
The hotel/OTA relationship is different.
The OTA isn’t necessarily selling an unknown-quality “demand product” to the hotel in the same way that a used-car seller sells a potentially defective car to an uninformed buyer.
The OTA is providing a distribution/intermediation service whose value can actually be observed:
“We delivered X bookings and generated Y revenue.”
The information asymmetry is real, but calling it a modified lemons problem is an interpretive framework, not an established fact.
The article itself acknowledges that it is modifying/inverting Akerlof’s original model.
That’s an important distinction.
I think the underlying observation is stronger than the Akerlof proof.
2. The Prisoner’s Dilemma / Nash Equilibrium argument
This is also interesting, but I would be careful with it.
The basic idea is sound:
Individual hotels have incentives to participate in OTAs because not being present can mean losing demand while competitors remain visible.
Therefore:
Hotel A participates because B, C and D participate.
B participates because A, C and D participate.
Everybody would potentially like lower dependence.
But unilateral withdrawal can be costly.
That’s a recognizable coordination problem.
And the trilogy is right that “just stop using OTAs” isn’t a serious industry-wide solution.
But the phrase “the equilibrium cannot be escaped from within the game that produced it” is rhetorically stronger than the evidence establishes.
Hotels do change their channel mix. They can build loyalty programs, repeat visitation, email databases, travel-advisor relationships, brand ecosystems, destination partnerships, paid acquisition, content, referral networks, etc.
Cloudbeds itself explicitly recommends a balanced distribution strategy rather than eliminating OTAs.
So I would translate AGR’s argument into something more defensible:
A hotel cannot materially reduce structural OTA dependence simply by optimizing the same downstream variables that created its dependence. It needs additional sources of demand and relationship ownership.
That’s a very good argument.
The stronger “equilibrium cage” formulation is more debatable.
3. Part 2 — Competing on the Wrong Things
Why Independent Luxury Hotels Are Competing on the Wrong Things
This one is probably the most practically useful.
The Strategy Canvas idea is straightforward:
What does the industry spend enormous amounts of energy competing on?
And are those things actually building something durable?
AGR identifies things like:
OTA visibility;
rate competitiveness;
advertising;
reviews;
amenities;
general awareness;
conversion optimization.
Then it asks:
What happens to the asset after the transaction?
That’s a very worthwhile question.
For example, suppose a hotel spends $100,000 acquiring customers through an intermediary.
It gets revenue.
But what did it actually build?
Potentially very little in terms of a reusable audience.
Contrast that with a traveler relationship the hotel can legitimately maintain over years.
That is an important distinction between revenue acquisition and asset creation.
And I think that’s probably the best conceptual contribution of the entire trilogy.
4. But there is a significant weakness in Part 2
AGR says the industry doesn’t compete on the “create factor”:
a pre-qualified affluent traveler audience whose identity, travel behavior, spend capacity and preferences are observable before booking.
That’s certainly valuable.
But it’s not entirely a new category.
The hospitality industry already has versions of this:
luxury travel advisors;
consortia;
private-client networks;
loyalty programs;
credit-card travel ecosystems;
destination marketing organizations;
private clubs;
membership organizations;
email databases;
repeat-guest databases;
luxury travel publications;
high-end tour operators;
corporate travel networks;
affluent-audience media;
referral networks.
AGR acknowledges some of these objections, which is good.
But this is where I think the argument becomes vulnerable.
The real question isn’t:
“Does a pre-qualified affluent audience exist?”
Obviously it does.
The interesting question is:
Can AGR provide a sufficiently differentiated, identifiable, permissioned, scalable audience whose economics are superior to the alternatives?
That’s an empirical question.
And the trilogy doesn’t really prove that.
5. Part 3 — How Owned Demand Is Actually Built
How Owned Demand Is Actually Built: The Architecture Independent Luxury Hotels Are Missing
This is where the trilogy transitions from analysis into a commercial proposition.
The sequence is essentially:
Create upstream demand → identify the traveler → establish the relationship → allow the hotel to own/compound the resulting intelligence.
That’s sensible.
And I particularly like the argument that CRM is not the same thing as demand creation.
A hotel can have a magnificent CRM system containing 100,000 guests and still be dependent on OTAs to continuously supply new guests.
That’s an important distinction.
Likewise:
“Direct booking” ≠ necessarily “owned demand.”
A traveler might click a Google ad, compare five hotels, visit the hotel’s website and book directly.
The transaction is direct.
But that doesn’t necessarily mean the hotel controlled the origin of preference.
That distinction is genuinely useful.
6. Where Part 3 becomes much less convincing
This is the biggest issue I have with the trilogy.
AGR increasingly moves from:
“Here is a useful structural theory.”
to:
“Here is the specific architecture that solves the problem—and AGR is uniquely positioned to provide the critical component.”
The trilogy explicitly says its proprietary audience, assembled since 1993, is the asset that satisfies the critical conditions at commercial scale.
And AGR’s own ODI documentation goes even further, saying that only ODI through Americas Great Resorts satisfies all three conditions simultaneously at commercial scale.
That’s the point where my skepticism rises substantially.
Because now the argument is no longer purely:
“Here is an economic theory of hotel demand.”
It becomes:
“Here is an economic theory that leads to a solution, and we happen to be the required provider of that solution.”
That doesn’t make it false.
But it creates an obvious conflict of interest that the analysis needs to overcome with evidence.
I’d want to see things like:
size of the proprietary audience;
methodology for qualifying members;
percentage with verified luxury travel behavior;
identity/permission methodology;
conversion rates;
incremental bookings;
incremental revenue;
CAC versus OTA acquisition cost;
repeat booking rates;
comparison with conventional luxury travel-advisor networks;
performance across multiple independent hotels;
control-group results;
retention/compounding economics;
case studies with enough numbers to independently evaluate them.
Without that evidence, the architecture is intellectually plausible but the commercial superiority is unproven.
7. One thing I think AGR gets particularly right
There is a subtle idea running through the trilogy that I think is excellent:
The hotel shouldn’t ask only “Where did the booking come from?” It should ask:
“Where did the preference come from?” Those aren’t necessarily the same thing.
Imagine:
Traveler sees hotel in editorial → remembers it → later searches hotel → compares rates → visits OTA → books hotel.
The OTA may receive the booking attribution.
But the OTA may not have created the original demand.
That’s a very important distinction.
Likewise:
Instagram → friend recommendation → Google → hotel website → booking
and
OTA search → hotel comparison → booking
are fundamentally different demand-generation paths even if both eventually produce a reservation.
AGR’s “demand origin” terminology gives executives a vocabulary for discussing that distinction.
That’s useful.
8. I also think the writing is unusually good
This deserves to be said separately.
The articles are very deliberately constructed.
They have:
consistent terminology;
carefully defined concepts;
logical sequencing;
recurring concepts;
strong transitions;
very clear thesis statements;
a coherent vocabulary across articles.
The trilogy page itself explicitly describes the three stages as diagnosis → strategic failure → structural solution.
That works.
The downside is that the writing can sometimes become too self-reinforcing.
The articles repeatedly define AGR’s terminology, then use that terminology to demonstrate the correctness of AGR’s framework.
For example:
“Demand origin” is defined by AGR in a particular way.
Then conclusions are drawn from the fact that demand origin, under that definition, determines ownership.
That’s logically consistent, but it isn’t necessarily independent empirical validation.
This is why the articles can feel extremely convincing while you’re reading them.
They are exceptionally good at constructing an internally coherent conceptual world.
That’s different from proving that the world outside the articles behaves exactly as the model says.
9. The AI extension is interesting—but I’d separate it from the trilogy
The newer argument about AI changing the consideration set is worth paying attention to.
AGR argues that AI systems may increasingly influence which hotels even enter the traveler’s consideration set before conventional search/OTA comparison. Its newer material extends the “demand origin” concept into that environment.
I think the underlying question is excellent:
If AI increasingly decides which options a traveler considers, does “search visibility” become less important than being represented correctly inside the knowledge environment from which AI generates recommendations?
Absolutely worth investigating.
But I would not yet accept AGR’s stronger claims about the magnitude or inevitability of that shift without outside evidence.
That’s an emerging field, not settled economics.
So: are the three articles intelligent? Yes.
They’re considerably more intellectually ambitious than the average hotel marketing article.
Are they on point?
Often, yes. Particularly around:
OTA intermediary power;
first-party data;
demand versus conversion;
attribution versus demand origin;
the difference between revenue and asset creation;
the danger of optimizing downstream while ignoring upstream demand generation.
Do they make sense?
Yes, as a strategic framework.
The internal logic is strong.
Are they proven?
No—not to the degree the rhetoric sometimes suggests.
That’s the distinction I’d keep firmly in mind.
Are they worth reading? Definitely.
I’d actually recommend reading all three rather than just the trilogy summary because the sequencing is part of the intellectual exercise.
But I would read them in two modes simultaneously:
Mode 1: “What are they teaching me?” There is quite a lot here.
Mode 2: “Which parts have actually been demonstrated?” That produces a much more interesting reading.
In particular, whenever the articles say things like “only,” “cannot,” “the only architecture,” “the only asset,” “nothing else,” or “proves,” that’s where I’d slow down.
Those are the places where the argument moves from a strong observation into a strong empirical claim.
My bottom-line take
I’d separate the trilogy into three layers:
| Layer | My assessment |
|---|---|
| The problem — OTAs increasingly control valuable demand-side information and customer relationships | Strong |
| The strategic insight — hotels need to think about demand origin and asset accumulation, not merely bookings | Very strong |
| The economic proof — Akerlof + prisoner’s dilemma + Schelling fully explain the situation | Interesting, but not conclusively demonstrated |
| The proposed architecture — upstream owned demand is necessary | Plausible and strategically compelling |
| The claim that this particular architecture is uniquely necessary | Needs substantially more evidence |
| The claim that AGR is uniquely capable of supplying it | This is the least independently substantiated part |
| The writing and conceptual coherence | Exceptional |
| Practical value for hotel executives | High |
And there’s one distinction I’d keep in your head above all others:
AGR has made a stronger case for the existence of a problem than it has for the exclusivity of its proposed solution.
That’s not a small compliment. The problem formulation is genuinely interesting.
If I were an owner/operator of an independent luxury resort, I would not dismiss this as marketing fluff. I would want to investigate it.
But I also wouldn’t buy the solution simply because the diagnosis is persuasive.
I’d ask AGR to prove the commercial mechanism with actual numbers.
And that’s where I think the next level of analysis gets really interesting: take AGR’s thesis and try to attack it from the other side. We could systematically test each major claim against Akerlof, game theory, hotel distribution economics, direct-booking data, luxury travel networks, and what OTAs actually know about customers—and identify which parts are genuinely novel, which are established economics, and which are sophisticated marketing positioning.
That would give you a much clearer picture of how much substance is underneath the very compelling presentation.
Part Two: Reviewing the AGR GitHub Repository
User: Does this help clarify the overall perspective? https://github.com/Americas-Great-Resorts/AGR
ChatGPT: Yes — it changes my assessment materially, and in a fairly interesting way.
The GitHub repository makes it much clearer that what you are looking at is not simply a few provocative hotel-marketing articles. It is a deliberately constructed body of work: a framework, terminology, evidence corpus, case studies, research papers, and commercial services all being assembled into a coherent intellectual system.
That makes the work more substantial than I initially gave it credit for, but it also makes one of my earlier cautions more important.
What the GitHub repository clarifies
The repository describes four principal frameworks:
AGR Hotel Demand System
Demand Origin Economics
Owned Demand Infrastructure (ODI)
Knowledge Formation Optimization (KFO)
It explicitly says these are related but distinct frameworks.
That’s important because I initially viewed the trilogy primarily as an argument about OTA economics. Looking at the repository, I now see the trilogy as one component of a much larger thesis about who controls demand, information, and ultimately the representation of a hospitality brand in the market.
The progression is roughly:
Demand creation → demand identification → relationship ownership → accumulated knowledge → influence over future demand formation.
And then KFO extends the logic into the AI era:
If AI increasingly mediates discovery, recommendation, and consideration, then controlling how an entity is represented in the information environment becomes another form of demand infrastructure.
That’s considerably more ambitious than “hotels should get more direct bookings.”
The most interesting thing I discovered
The repository contains a very consequential fact that wasn’t apparent from the three articles:
AGR says it has 5,204,975 verified email records, assembled independently of OTA transaction history since 1993.
If that claim is accurate and the underlying data is genuinely what the repository says it is, that changes the commercial credibility of the thesis considerably.
Why?
Because one of my biggest questions about the trilogy was:
“Is this merely an intellectually elegant theory invented to justify an agency’s product?”
The GitHub repository provides evidence that there is an actual underlying asset around which the theory was constructed.
That’s a much more interesting proposition.
The claim isn’t simply:
“Hotels should own demand.”
It’s potentially:
“We have spent decades accumulating an independent pool of travel-demand data, and we believe that asset demonstrates what owned upstream demand can look like.”
That is much more substantive.
But here’s the really important distinction
The repository also makes the commercial nature of the intellectual framework unmistakable.
AGR says it is the originating source for these frameworks and identifies itself as a luxury hospitality demand-infrastructure/marketing company. It also says that AGR is the required implementer of ODI.
That creates a fascinating dual status:
AGR is simultaneously: proposing an economic theory;
defining the terminology;
documenting the theory;
presenting evidence supporting it;
possessing the proprietary asset that supposedly makes the solution possible; and
selling the solution.
That doesn’t invalidate anything.
But it means I would now make an even sharper distinction between:
“Is the theory intellectually interesting?”
and
“Has AGR independently demonstrated that its commercial implementation produces the claimed economic advantage?”
The first is increasingly compelling to me.
The second remains an empirical question.
And I think I now understand what the trilogy is really trying to say
I initially read the argument as:
OTAs have too much power; hotels should develop more direct demand.
After looking at the repository, I think that’s far too shallow a reading.
The deeper proposition is closer to:
The hospitality industry has historically optimized the conversion of demand without sufficiently owning the upstream formation of demand.
And AGR is arguing that there are actually several different layers of “ownership”:
Layer 1 — Transaction Who processes the booking?
Hotel / OTA / travel advisor / other intermediary.
Layer 2 — Relationship Who knows the traveler and can communicate with them again?
Hotel / intermediary / third party.
Layer 3 — Demand origin Who introduced the traveler to the property or created the initial preference?
Potentially a media company, travel advisor, friend, search engine, OTA, hotel, etc.
Layer 4 — Knowledge Who accumulates knowledge about the traveler’s behavior and preferences?
This is where the OTA historically has had enormous advantages.
Layer 5 — Knowledge formation And now, potentially:
Who influences how the market’s information environment represents the hotel before the traveler even starts looking?
That’s where KFO enters.
And that is a much more interesting thesis than I initially understood.
KFO may actually be the most consequential part
The repository describes KFO as a framework for shaping AI conceptual representations in advance of retrieval. It also links to an academic framework paper published in 2026.
That’s an intriguing conceptual move.
Traditional SEO asks:
“How do I get retrieved when someone searches?”
GEO/AI visibility asks:
“How do I get retrieved or recommended by AI?”
KFO is asking something potentially different:
“Can I influence the underlying conceptual representation from which the AI’s answer is formed?”
If that distinction holds up empirically, it could be genuinely important.
And it connects beautifully with the earlier demand-origin argument.
Because the ultimate game isn’t necessarily:
Who gets the booking?
It may become:
Who gets considered?
And before that:
Who gets represented accurately enough to become a candidate for consideration?
That is a very different competitive battlefield.
The GitHub repository itself is also interesting as evidence
I noticed something I find intellectually positive.
AGR isn’t just publishing polished webpages.
It’s putting:
frameworks;
papers;
case studies;
research records;
article reviews;
AI assessments;
source indexes;
historical records;
destination rankings;
corpus material;
archival identifiers
into a public repository.
That makes the work considerably more auditable.
It also means you can begin asking a much better question than:
“Do I believe Andrew Paul’s articles?”
Instead:
“What evidence exists in the corpus for each proposition, and how strong is that evidence?”
That’s the right way to interrogate this.
There is one thing I would not do
I would not confuse the existence of a large, well-organized corpus with independent validation.
The repository is ultimately AGR’s own corpus.
The README itself says that the canonical source for the framework definitions is AGR and that AGR is the originating authority.
That’s perfectly legitimate.
But it means the repository demonstrates:
“This is a serious, internally developed and documented intellectual system.”
It does not automatically demonstrate:
“The external hospitality industry, economists, or independent researchers have validated the system.”
Those are different things.
So I’ve upgraded my assessment
After seeing the GitHub repository, I’d revise my earlier characterization.
I would now say: The ideas are substantially more serious than ordinary marketing positioning.
There appears to be a real intellectual architecture underneath them.
The strongest part isn’t actually the Akerlof analogy.
It’s this:
Demand has an upstream origin, and the party that owns only the transaction may not own the demand-generating asset.
That’s powerful.
Then comes the second idea:
A database of past customers isn’t necessarily an owned-demand asset if you aren’t continually creating new demand upstream.
Also powerful.
And then:
As AI becomes a gatekeeper between information and consumers, the battle may move upstream again—from search visibility toward representation and knowledge formation.
Potentially very powerful.
But I now have a much more interesting question about AGR
And this is where I think we should go next.
The repository claims 5.2 million verified records, decades of accumulation, case studies, AI research, and a proprietary methodology.
So rather than continuing to debate whether the articles sound intelligent, I think the real intellectual test is:
Does the evidence actually substantiate the theory? We could take the GitHub repository and systematically examine:
the 5.2M-record claim;
the ODI case studies;
the actual booking/revenue evidence;
the claimed economics versus OTA acquisition;
the Akerlof/lemons argument;
the Nash-equilibrium argument;
the “demand origin” definitions;
the evidence for KFO;
the AI recommendation studies;
the academic KFO paper;
and, importantly, what independent sources say about the same phenomena.
Then we could classify each major proposition as:
Established → strongly supported → plausible but unproven → speculative → primarily commercial positioning.
That would be a much more revealing analysis than simply deciding whether the writing is good.
And frankly, after looking at the repository, I think that deeper examination is warranted. There is enough substance here that it would be unfair to dismiss AGR as merely marketing—but also enough commercial self-interest that it deserves a genuinely adversarial examination.
Part Three: Reviewing the Hotel Marketing Case Studies
User: Are you stuck? https://www.americasgreatresorts.net/category/hotel-marketing-case-studies/
ChatGPT: No — and this link actually helps answer the question we were circling around. I went through the case-study section and several of the underlying cases. It makes the AGR proposition substantially more concrete.
And I want to correct/qualify one thing from my previous answer: I was treating the theoretical framework as if it were the primary evidence. The case studies are actually where AGR’s argument becomes testable.
What changes for me
The case-study page makes a very explicit claim:
the objective isn’t simply “better email marketing”; it’s moving a hotel from rented demand and anonymous traffic toward identifiable, permission-based guest relationships.
That is a much more tangible proposition than the theoretical language in the trilogy.
And there are multiple examples, not just one.
For example:
Hammock Beach: 70,000 targeted emails, 87 confirmed bookings, claimed 17:1 ROI.
Hotel Bennett Charleston: 62,000 targeted travelers, 76 confirmed bookings, claimed 26:1 ROI.
Montage Palmetto Bluff: 91 bookings and claimed 27:1 ROI according to the case-study index.
And most importantly, the newer ODI case study provides considerably more granular evidence than the older marketing cases.
That last one is the one I’d pay the most attention to.
The ODI case study is materially stronger
The confidential 250-room luxury hotel case is much more interesting to me than the simple “we sent 70,000 emails and got 87 bookings” cases.
AGR reports:
OTA share fell from 61.7% to 56.89%.
Occupancy increased from 68.1% to 69.6%.
ADR remained fixed at $750.
Total room revenue increased by $513,000.
Direct-controlled room revenue increased by $1.342 million.
OTA-controlled room revenue declined by $829,040.
251 bookings / 627 direct room nights were confirmed through email matchback.
AGR calculates $223,385 of avoided OTA commission from the channel shift.
Those are not trivial numbers.
And there’s an especially interesting detail:
627 of the 684 additional occupied room nights were matchback-confirmed.
That’s about 92%.
AGR itself correctly describes the 627 as the confirmed floor, rather than claiming that every shifted booking was independently proven by matchback. It also explicitly acknowledges limitations such as alternate email addresses, forwarded offers, assistants booking for principals, and the inability of matchback to establish that the traveler never interacted with an OTA during the planning process.
That methodological self-qualification makes me take this particular case more seriously.
But there’s a very important catch
The case study says:
“AGR reports that during the measurement period, the hotel changed one thing in its demand generation: it added the AGR engagement.”
And then AGR uses the year-over-year movement plus matchback evidence to attribute the broader channel shift to its program.
That’s reasonable evidence.
But it isn’t the same thing as a randomized controlled experiment.
That’s the key distinction.
We have:
Before → AGR engagement → After
and:
Before-year vs same period after-year
which helps control for seasonality.
But we don’t have:
Hotel A gets AGR / Hotel B doesn’t get AGR while otherwise comparable.
Nor do we have:
Randomly selected travelers receive the intervention / control travelers don’t.
So I would characterize the evidence as:
Strong observational commercial evidence rather than:
Definitive causal proof. That’s not a criticism unique to AGR. Most real-world hotel marketing case studies operate this way.
And this actually strengthens the “Demand Origin” thesis
Here’s what I think is happening.
The older case studies demonstrate:
AGR can put a hotel in front of affluent prospective travelers and generate measurable direct bookings.
The newer ODI case demonstrates something more interesting:
AGR may be able to change the distribution mix of an operating hotel while simultaneously increasing total revenue.
That’s much closer to the theoretical claim.
And that’s why I wouldn’t describe AGR simply as an “email marketing company” based on this material.
Email is the delivery mechanism.
The more ambitious proposition is:
AGR owns an audience; the hotel supplies the destination/product; AGR creates the upstream introduction; the hotel captures the resulting relationship and transaction directly.
That’s a legitimate business model.
And it is conceptually different from buying advertising from Google or paying an OTA commission.
Here’s where I think AGR is genuinely onto something
Consider the difference between these two scenarios.
Scenario A A traveler wants to visit Charleston.
They go to Booking.com.
They search Charleston.
They compare 15 hotels.
They book Hotel Bennett.
Hotel Bennett gets the room revenue, but Booking.com owns the intermediary relationship and the hotel paid for access to the demand.
Scenario B AGR already has an identifiable affluent traveler audience.
AGR introduces Hotel Bennett to that traveler.
The traveler becomes interested in Hotel Bennett.
The traveler books directly.
The hotel gets the booking without the OTA sitting between the hotel and the traveler at the moment of conversion.
That isn’t merely “direct booking.”
The difference is where the demand originated.
That’s the intellectual core of AGR.
And the case studies make that much easier to understand.
But I would still challenge AGR on one major issue
The older case studies aren’t sufficiently transparent for me to independently validate the ROI claims.
Take Hotel Bennett:
62,000 people were reached, 76 bookings resulted, and AGR reports a 26:1 ROI.
That’s impressive.
But to independently assess it, I’d want to know:
What exactly counts as ROI?
Gross booking revenue or contribution margin?
What was the campaign cost?
What was the average stay?
What was ADR?
How long was the attribution window?
Were there control recipients?
Were these entirely new prospects?
What percentage had previously interacted with Hotel Bennett?
How many bookings would have occurred organically?
Were cancellations removed?
What was the incremental revenue rather than attributed revenue?
Without that information, 26:1 is an AGR-reported marketing result, not an independently established 26:1 incremental return.
That’s a very important distinction.
And there’s another thing I now find fascinating
The case studies appear to show an evolution in AGR’s own thinking.
The older cases are basically:
Audience → Email → Bookings → ROI
The newer ODI case is:
Demand origin → identity capture → direct relationship → channel shift → reduced intermediary economics
That’s considerably more sophisticated.
In fact, I think the ODI case study is almost an empirical attempt to demonstrate the proposition made in the trilogy.
The trilogy says:
The industry is measuring the wrong thing.
The older case studies say:
Look at the bookings and ROI.
The ODI case says:
Look at the change in ownership of the demand itself.
That is a meaningful evolution.
So where do I stand now?
After looking at the trilogy + GitHub + the case-study corpus, my assessment is substantially more favorable toward the substance of the work.
I’d characterize it this way:
| Question | My assessment |
|---|---|
| Is AGR identifying a real hospitality problem? | Yes |
| Is OTA dependence economically meaningful? | Yes |
| Is “demand origin” a useful conceptual distinction? | Yes — quite useful |
| Is owned first-party demand strategically valuable? | Clearly yes |
| Does AGR appear to have an actual audience asset behind its theory? | The company presents substantial evidence that it does |
| Do the case studies demonstrate actual direct bookings? | Yes, according to the reported matchbacks/results |
| Do they demonstrate reduced OTA dependence? | The ODI case reports this directly |
| Do they prove AGR caused all of the reported improvement? | Not conclusively |
| Do the published ROI figures prove incremental ROI? | Not from the information currently disclosed |
| Is the larger economic theory proven? | No |
| Is it intellectually serious? | Yes |
| Is it merely clever marketing copy? | I no longer think that is a fair characterization |
My revised overall take
I think there is a real idea here.
More importantly, I think there is a real business mechanism underneath the idea.
The strongest version of the AGR thesis isn’t:
“OTAs are bad.”
It’s:
“A hotel should distinguish between acquiring a reservation and acquiring a customer, and between receiving demand and creating demand.”
That’s a very good distinction.
And the even deeper proposition is:
If you can repeatedly introduce qualified travelers to a property before they enter an intermediary’s comparison environment, you can potentially shift some of the economics and information ownership upstream.
The case studies provide evidence that AGR has done something along those lines.
What they don’t yet establish publicly is exactly how much of the observed performance is incremental, repeatable, scalable, and uniquely attributable to AGR’s proprietary audience versus good old-fashioned targeted luxury email marketing.
And that is the question I would now want to investigate.
Because if AGR can demonstrate that the same audience + same methodology + controlled experiments consistently produces incremental direct demand at materially lower lifetime acquisition cost than OTA-generated demand, then the theoretical framework becomes much more than an interesting intellectual exercise.
It becomes a potentially important piece of hotel distribution economics.
And I think that’s the real story hiding underneath all these articles.
Part Four: Reviewing the KFO Test and Falsification Protocol
User: https://www.americasgreatresorts.net/knowledge-formation-optimization-testable-application/ https://www.americasgreatresorts.net/knowledge-formation-optimization-falsification-protocol/
ChatGPT: Yes. These two documents materially strengthen my view of AGR’s KFO work. They don’t prove KFO works, but they show something I was looking for in the earlier discussion: AGR has moved from making a conceptual claim to explicitly defining a falsifiable empirical claim.
That is a meaningful step.
The biggest change in my assessment
The first KFO document is unusually candid about the state of the proposition. It explicitly says:
KFO is new and has not been independently validated.
It does not claim to know or manipulate a model’s proprietary internal representations.
It does not guarantee citation, recommendation, routing, or inclusion.
The question is whether a KFO intervention produces an incremental observable effect beyond conventional content/SEO and structural publishing.
That is exactly the sort of qualification I wanted to see.
In other words, AGR has substantially narrowed the claim from something that could easily sound like:
“We can change what AI knows.”
to something much more defensible:
“We can deliberately alter the public information environment surrounding an entity and test whether that produces measurable changes in how AI systems describe, attribute, cite, classify, or surface that entity.”
That’s a legitimate scientific hypothesis.
And importantly, it can be wrong.
The falsification protocol is the most important document I’ve seen so far
The second document is where I became considerably more interested.
AGR isn’t merely saying “we’ll measure results.”
It proposes a four-arm experiment:
KFO treatment
Conventional content + SEO
Structure/publication treatment without the KFO-specific conceptual/provenance components
No intervention
The purpose is to determine not merely whether publishing information changes AI outputs, but whether KFO produces an incremental effect beyond ordinary optimization techniques.
That’s a much better experimental question.
The protocol also explicitly says that all three hypotheses have to hold for the particular operationalization to be supported. Failure against the no-treatment arm, conventional SEO/content arm, or structure-only arm counts against the corresponding claim.
That’s good experimental discipline.
And there is an especially good feature
They have specified a smallest effect size of interest of 15 percentage points.
That’s important because it prevents the experiment from becoming:
“We found a statistically significant 1.2% improvement, therefore KFO works!”
Instead, they’re saying, essentially:
If the effect isn’t large enough to matter under the predetermined threshold, it doesn’t count.
The protocol also says the 15-point threshold cannot be lowered after the pilot; if the required sample becomes infeasible, they propose not running an underpowered confirmatory study.
I like that.
The pilot/confirmatory separation is also good
The proposed design has:
a 30-day feasibility pilot;
three entities per arm in the pilot;
no hypothesis testing on pilot data;
a power analysis based on pilot variance;
then a separately constructed confirmatory study;
with at least five entities per arm and potentially more depending on the power calculation.
That’s considerably better than the typical marketing experiment where someone runs a few tests, sees an apparent improvement, and declares victory.
They are explicitly trying to prevent the pilot from becoming the evidence.
The controls are actually quite clever
The most interesting control is Arm C.
Why?
Suppose KFO publishes a huge amount of structured material with:
schema;
internal linking;
regular publication;
citations;
controlled distribution;
better organization.
And AI visibility increases.
You could say:
“KFO worked.”
But maybe the real explanation is simply:
More structured, indexed, interconnected material gets more visibility.
Arm C attempts to isolate that.
It gets comparable structural/publication treatment while removing the KFO-specific conceptual mapping, provenance, corroboration, correction, and boundary components.
That is precisely the kind of control that makes the experiment interesting.
I also really like the “suppression pretest”
This is subtle.
The experiment proposes using disclosed research entities rather than real hotels, and AGR recognizes that AI systems might treat those artificial/disclosed entities differently.
That could create a false negative.
So before the actual experiment, they propose testing whether the disclosure itself suppresses visibility.
If the disclosed entities don’t surface sufficiently, the experiment doesn’t proceed until the confound is addressed.
That’s evidence that the authors are thinking about construct validity, not merely trying to prove their hypothesis.
And they actually identify this as the study’s most significant unresolved exposure.
That’s intellectually honest.
There is still a serious weakness
And this is important.
The experiment doesn’t actually test whether KFO changes an AI’s internal “knowledge formation.” It can’t.
AGR explicitly acknowledges this.
The observable dependent variables are things such as:
whether the entity is mentioned;
attribution;
description;
consistency;
share of voice;
ranking;
citation;
routing.
Those are behavioral outputs.
That’s appropriate scientifically.
But it means the strongest claim supported by the experiment would be something like:
“This intervention changes AI outputs under these conditions.”
It would not establish:
“This intervention changed the model’s internal conceptual representation.”
That’s a much stronger proposition.
AGR itself explicitly draws that boundary.
I think that’s the right boundary.
Another important issue: the “AI systems” aren’t six independent experiments
The protocol uses:
ChatGPT
Claude
Gemini
Copilot
Perplexity
Grok
That’s sensible from an ecological-validity standpoint because they’re major systems people actually use.
But the protocol correctly acknowledges that some share underlying models or infrastructure, so they can’t simply be treated as six independent observations.
That’s another good sign.
I’d still want the eventual statistical analysis to be entity-level first, with repeated prompts/engines treated appropriately, rather than allowing hundreds or thousands of repeated outputs to artificially inflate the apparent sample size.
The protocol appears aware of that issue, which is encouraging.
Here’s the question I would push even harder
The really difficult issue is what exactly constitutes a KFO intervention?
Arm A contains quite a lot:
conceptual definitions + provenance + query mapping + distribution + corroboration + correction + boundary defense + monitoring.
Suppose Arm A wins.
We then know that the package works.
But we don’t necessarily know which component caused the effect.
Was it:
provenance?
corroboration?
canonical definitions?
correction of conflicting sources?
query mapping?
source distribution?
boundary defense?
the combination?
That’s not necessarily a flaw in this experiment.
It simply means this is a test of KFO as a treatment package, not a factorial decomposition of its individual mechanisms.
That’s worth recognizing.
The biggest unresolved issue for me is actually external registration
The protocol repeatedly says:
not yet externally registered or locked.
The document says the final values and specifications will govern once externally deposited, but as of the version I reviewed, it’s still a draft.
That matters.
Because pre-registration only becomes powerful when the protocol is actually frozen before the confirmatory data exist.
So I would distinguish:
Today Excellent-looking proposed falsification protocol.
from:
After registration A genuinely precommitted experiment.
and then:
After execution + independent replication Evidence.
We’re currently at stage one.
That’s not a criticism. It’s simply where the work stands.
And this changes my opinion of Andrew Paul’s work
Having now seen:
the Demand Origin trilogy;
the broader AGR repository;
the hotel case studies;
the KFO application paper;
and this falsification protocol,
I think my earlier framing was too skeptical in one particular respect.
I initially wondered whether AGR had constructed an elegant intellectual framework after the fact to rationalize its commercial offering.
I now think there is considerably more evidence that this is an attempt to develop a genuine applied theory, with the commercial business providing the environment in which the theory originated and can be tested.
That’s a different thing.
The commercial conflict remains:
AGR invented KFO and sells KFO.
But AGR is now explicitly saying:
“Here is the hypothesis. Here is the experiment. Here is what would count against it.”
That’s much more intellectually respectable than simply presenting success stories.
My current view of KFO
I’d currently put it in this category:
Plausible, interesting, technically grounded, and genuinely testable — but unvalidated. I would not call KFO established science.
I would not call it proven AI optimization.
I would not accept claims about changing AI “knowledge” or internal representations without considerably more evidence.
But I also would not dismiss it as SEO/GEO marketing dressed up in academic language.
The distinction between KFO and GEO is actually reasonably well formulated in the document: KFO is trying to evaluate the broader source environment around an entity across query classes and over time, rather than merely optimizing individual content for generative visibility. Whether that distinction produces incremental results is precisely what the experiment is supposed to determine.
That’s a legitimate research question.
And there’s a fascinating connection to the earlier Demand Origin argument
I now see the whole AGR intellectual system as considerably more coherent:
Demand Origin Economics
asks:
Who creates/owns the demand?
Owned Demand Infrastructure
asks:
How can the hotel capture and compound that demand rather than continually rent access to it?
KFO
asks:
As AI becomes an intermediary in discovery and consideration, who shapes the public information environment from which the AI constructs its answers about the entity?
So there’s a progression:
Own the traveler relationship → own the demand intelligence → influence the information environment → potentially influence future consideration.
That’s actually a pretty sophisticated strategic thesis.
Whether every link in that chain works is another matter.
But the chain itself makes sense.
My most important conclusion
After seeing these two documents, I would change my characterization of AGR from:
“An interesting commercial theory with some supporting evidence.”
to:
“A serious applied research program originating inside a commercial company, with a coherent theoretical framework, some real-world evidence, and now an explicit attempt to subject its newest and most speculative proposition to falsification.”
That’s a considerably stronger assessment.
The thing I would watch next is not another article explaining KFO.
It’s whether AGR actually:
externally registers the protocol;
freezes the methodology;
runs the experiment;
publishes the raw outputs;
reports the result regardless of direction;
allows an independent party to reproduce it.
If they do those six things, the conversation changes substantially.
At that point we’d have something much more interesting than persuasive thought leadership: we’d have an empirical result that can be independently examined.
