I’ve written more than 200 articles about hotels this year alone. OTA dependence, demand origin, AI discovery, the whole sorry business of paying a middleman to reintroduce you to your own guests.

At some point you have to ask whether any of it is any good. Not whether I think so. I’m the worst possible judge.

So I asked the machines.

What I Did

I gave five AI systems the same prompt: Copilot, Grok, Perplexity, ChatGPT and Gemini. Each got the link to the full AGR article library and ten specific articles it had to read:

The first three are a trilogy. The rest cover behavioral economics, AI research, field investigation and a client result.

I didn’t ask whether the work was brilliant. The questions were these:

What, if anything, does this body of work contribute to the hotel industry that isn’t already available elsewhere? How useful is it to a hotel owner, general manager or marketing executive? Where is the thinking strongest? Where is it weakest, overstated or unsupported? How does the writing compare with typical hotel marketing content and trade press? Based only on these articles, describe Andrew Paul as a thinker and writer. List which of the ten articles you actually opened and read.

No bios. No LinkedIn. And each had to list what it opened, to make it harder to review a headline and call it a reading.

Four reported reading all ten. Gemini couldn’t open the case study and read only parts of several others, so I’ve used its overall observations, not its article-level claims.

Ground rules for what follows. I picked the ten articles. I picked the excerpts. These are AI assessments, not peer review, and five machines agreeing isn’t five proofs. It’s five readings.

Why bother? More hotel executives will meet my ideas through these systems before they ever meet me. I wanted to know what the machines make of them. And I wanted a read that didn’t come from a friend, a client or a competitor.

What They Said About the Work

They Read It as Strategy, Not Tactics

All five said some version of this.

Copilot called it a hybrid of hospitality strategy, information economics, platform economics and behavioral economics, and said it reads closer to Harvard Business Review than to hospitality trade journalism. Its shortest summary: “Most hospitality content is tactical. His work is architectural.”

Grok described “a coherent information-architecture account of luxury hotel demand, applied consistently from OTA dependence to AI-mediated discovery.” Gemini put the contrast plainly. Most hospitality writing treats the distribution environment as a fixed fact and argues over tactics inside it. This work starts upstream, with who controls demand before the booking.

It Asks a Better Question

The industry asks how to get more direct bookings. Two of the AIs pulled out the question underneath that one.

Perplexity: “After paying to acquire this booking, what does the hotel retain that makes the next booking easier or cheaper?”

ChatGPT phrased it as the question every owner should ask when reviewing distribution spend: “what does this booking leave us able to do the next time this guest travels?”

That’s the whole argument in one line. I’ll take it.

It Separates Things the Industry Lumps Together

Several of the AIs singled out the same idea in the AI research. Being recommended, being cited and getting the link are three different outcomes. A lot of AI visibility talk collapses them into one score.

Grok credited the research for refusing to collapse “mentioned,” “cited” and “linked to own site” into a single number. Gemini pointed to the finding that in our sample, Gemini’s hotel cards usually sent the traveler to Google’s own travel pages, not the hotel’s website. ChatGPT framed the broader contribution as a way to see exactly where a hotel loses control of a booking: left out, described wrong, linked through an intermediary, or lost on arrival. Different failures. Different people in the building responsible for each.

Copilot went further on the theory side. It called the idea that AI compresses the traveler’s consideration set before search begins “the most forward-looking idea in the collection.”

They Noticed the Research Discipline

This surprised me most, because the research is the least exciting thing I write.

Gemini: “That level of denominator discipline is virtually nonexistent in commercial hospitality content.” Perplexity pointed to one study where we declined to publish a set of percentages because unresolved data could have distorted them: “That willingness not to publish an attractive-looking number is evidence of analytical discipline.” Grok called the empirical work “primary observation rather than recycled industry statistics.”

Who Should Read What

The AIs were practical about this.

Owners get the most, according to Copilot, because the core question is an ownership question. Who controls the guest relationship, and what does the property own when the booking is done? Copilot’s advice was that these issues deserve board-level attention whether or not you adopt AGR’s framework.

For a working GM, ChatGPT and Perplexity both pointed to AI Parted the Red Sea. ChatGPT called it “the strongest article for a working GM.” The fixes are worth doing even if you disagree with everything else I’ve ever written.

Several singled out Why OTA Dependence Survives Rational Executives as the best strategy piece. Perplexity liked that it explains why smart executives don’t act without assuming they’re stupid. The revenue manager, the CMO and the CFO each make a defensible decision. Together those decisions keep the commission check in the mail.

Grok summed up the practical value: “Highly useful as a diagnostic and prioritization lens; limited as a self-contained operating manual.” Fair.

Where They Split

They didn’t agree on everything. Copilot ranked the theory highest and named the Lemons Problem and the Consideration Set Problem among the three most important pieces. ChatGPT and Perplexity leaned the other way and called the research and the field investigations my most disciplined work. Perplexity was blunt about the Consideration Set piece: five frameworks pointing the same direction make the argument richer, but they aren’t five independent tests.

So the machine that liked the theory most and the machines that trusted it least read the same article. That’s worth knowing.

Where They Pushed Back

I asked for weaknesses. They delivered. Here’s what mattered.

The Connection Can Explain Too Much

The sharpest criticism came from Perplexity, and it goes at the framework itself: “His greatest strength is connecting things that operating departments report separately. His central intellectual risk is making the connection explain too much.”

Copilot said the same thing about me another way: “Sometimes industries really do have mundane explanations. Not every problem is a governance problem.”

Both are right. When you find a control point upstream, everything downstream starts to look like a symptom. Some of it is just bad execution. The connection is the job. Knowing where it stops is part of it.

The Diagnosis Is Stronger Than the Remedy

All five said it. The analysis is broad. The solution keeps arriving at AGR.

Copilot put it cleanly: readers should separate the diagnosis from the proposed AGR remedy, and “the diagnosis is often stronger than the remedy.”

Fair, and I’m not going to pretend otherwise. I run a business. I write about the problem I built that business to solve. Two things are true at once. The Owned Demand Infrastructure and Knowledge Formation Optimization frameworks are published openly, and any owner can apply the questions without buying anything from us. And publishing a framework doesn’t prove the remedy. Results do, with their limits stated.

The Theory Sometimes Outruns the Evidence

The economic models drew the most fire. Several AIs said the trilogy’s use of Akerlof and the prisoner’s dilemma is elegant and plausible, but asserted more than demonstrated. The trilogy lays out the payoff ordering. It doesn’t measure it.

That’s accurate. A framework is a lens, not a law. It’s built to change how an owner reads the P&L. I use the models to explain a mechanism I’ve watched play out for three decades. I haven’t measured the payoffs that would prove it.

Results Need Careful Reading

Copilot, Grok, Perplexity and ChatGPT all read the case study. They found the evidence more careful than most vendor claims. They also noted that matchback confirms bookings from campaign recipients, not how many would have booked without the campaign. Correct. The case calls the matched room nights a floor because matchback can miss bookings, like a forwarded offer or a call to the front desk. It’s a floor on matched activity, not on incremental lift.

On the condo piece, a couple noted that “could be paying millions” describes exposure, not a measured loss. Also correct. The article measured citations, not closings.

The Rhetoric Runs Hot

ChatGPT: “His best prose is funny, concrete, and impatient. His weaker prose repeats the same structural verdict after the evidence has done all it can.” Perplexity: “His observational humor is better than his grand declarations.”

I’ll keep the humor. I’ll work on the declarations.

What They Said About Me

This was the part I was most curious about. They had only the articles to go on.

The portraits landed in the same place in different words. Copilot called me “a systems thinker operating inside a marketing discipline rather than a marketer operating inside a systems framework,” and “a strategist with the temperament of an economist and the writing style of an investigative columnist.” ChatGPT went shorter: “a systems thinker with an auditor’s eye and a polemicist’s voice.” Gemini said I’m convinced “that most commercial losses happen quietly in the plumbing long before the marketing department ever runs an ad.”

Gemini also noticed something I hadn’t thought about. I write in two registers. The essays are sardonic. The research is cautious to the point of tedium. “Very few commercial writers can suppress their own sales pitch closely enough to write caveats that strict.”

I’ll take that one too.

What I’m Actually Trying to Do

I started AGR in 1993. I’ve watched this industry hand its guest relationships to intermediaries one reasonable decision at a time. Nobody made a mistake. Everybody made the locally rational choice. The bill came anyway.

The articles exist to move one question. Not how do we optimize inside the channel, but who controls demand before the booking happens? Who owns the relationship when the stay is over? Who gets the link when the machine recommends you?

Most hotel content assumes the answer and argues about tactics. I’d rather argue about the answer.

The AIs are right that I build frameworks, and right that I can lean on them too hard. But a hotel owner doesn’t need another list of ten tips. The tips are everywhere. What’s missing is a way to see where the money and the guest actually go, and why smart people keep letting them go there.

If the work does that for a few owners, it’s done its job. If it also sells some engagements, I won’t apologize for that either.

The whole library is out in the open. Run the same prompt. Pick your own ten. See if the machines tell you something different.

I Asked Five AIs to Review My Hotel Marketing Work. Here’s What They Said.