I Asked AI to Tear Apart My Book. It Found Something I Missed.

Seth Godin recently wrote about an interesting use of AI: find a safe critic.

Take something you’ve created—a book, an essay, a business plan, a resume—and ask an AI to attack it from a particular point of view. Ask for the one-star Amazon review. The disappointed CEO. The skeptical venture capitalist. The recruiter who doesn’t know what to do with you.

Or, in my case, the professor sitting across the table at a Ph.D. defense.

So I tried it on my book, B2B Marketing in the AI Era.

I gave AI several critics and essentially told them to have at it.

Some of the criticism was predictable.

The imaginary Amazon reviewer thought I was too confident that AI had already changed B2B buying. The CMO thought I spent too much time in the machinery of marketing and not enough time on corporate strategy. The CEO had perhaps the most CEO response possible: “Fine. What do I stop doing Monday morning?”

All fair.

The venture capitalist pushed the argument further. If AI makes production dramatically cheaper and agents can perform more marketing work, why assume the marketing organization of the future looks anything like the marketing organization of today?

That one deserves more thought.

But it was the imaginary Ph.D. committee that found the hole.

The professor went after one of the central ideas in the book: the Trust Stack.

The criticism was simple:

The Trust Stack is an interesting construct. But is it a framework, taxonomy, causal model or metaphor? If it is causal, test it.

That bothered me.

Which usually means there is something worth looking at.

The Problem Wasn’t the Trust Stack

I’ve spent enough years in B2B technology, data, AI and services to know that companies rarely lose complicated deals because the buyer hasn’t seen enough marketing.

They lose because somebody gets nervous.

The product might work, but can we implement it?

The business case looks good, but can we prove it?

They say they’re enterprise-ready, but what happens when Security gets involved?

The demo was impressive, but has anyone like us actually deployed this?

The salesperson says six weeks. Who is willing to put that in writing?

B2B buying is filled with these moments because B2B buying is about trust, risk and implementation reality.

That’s the reason I developed the idea of the Trust Stack.

The Trust Stack is the buyer-facing evidence architecture that makes choosing you less risky. It includes proof of outcomes, security posture, implementation credibility, commercial evidence and the other things a champion needs to get a decision through the organization.

The idea still holds.

What the criticism exposed was that I had described the architecture without making the operating mechanism underneath it explicit enough.

That mechanism is Proof Ops.

AI Is Creating Infinite Claims

This distinction becomes much more important in the AI era.

AI has made producing claims ridiculously inexpensive.

Go to almost any B2B technology website and you’ll see them.

Faster time to value. Seamless integration. Enterprise-grade. AI-powered. Secure. Scalable. Transformative. Built for the enterprise.

Give a reasonably capable model a positioning statement, an ICP and some brand guidelines and it can produce hundreds of variations of those claims before your old agency could schedule the kickoff meeting.

That’s the production shift I talk about in the book.

Content is abundant.

Claims are abundant.

Persuasion is abundant.

Proof isn’t.

You can generate “customers achieve faster time to value” in a second.

You cannot generate the customer who actually achieved it.

That changes what marketing needs to operate.

The job isn’t simply to make more persuasive claims. It is increasingly to create a system in which important claims can be verified.

That’s Proof Ops.

Positioning Makes the Claim. Proof Ops Has to Prove It.

Suppose you’re selling an enterprise AI platform and your positioning is “the fastest path from enterprise data to production AI.” That’s a useful position because it makes a choice. You’re not claiming to be everything to everyone. You’re saying that speed from data to production is the thing you want to own.

Now you have to prove it.

If the median customer reaches production in 47 days, that’s proof. If three customers can document that they went from contract to production in less than two months, that’s proof. If your architecture eliminates two integration steps competitors require, that’s proof. If your services team has data from 80 implementations showing where projects normally get stuck and why your approach avoids those problems, that’s proof.

And if none of that evidence exists, that’s important too.

This is where Proof Ops comes in. I think of Proof Ops as the operating discipline for capturing, validating, structuring, maintaining, distributing and ultimately measuring the evidence behind the claims a company makes.

The relationship is straightforward. Positioning makes the claim. Proof Ops substantiates it. The Trust Stack organizes that proof around buyer risk. Distribution puts the evidence where humans and machines can discover it. Revenue outcomes tell you whether any of it mattered.

That is the system I hadn’t articulated clearly enough.

Most Companies Already Have the Proof

I’ve seen versions of this problem for years. Marketing says there aren’t enough customer stories while Sales insists there are plenty. Customer Success knows which customers are producing extraordinary results, and Professional Services usually knows why. Product has benchmarks nobody outside Product has seen. Security has answered essentially the same questions hundreds of times. Finance built a great ROI model for a large opportunity six months ago, but it lives in someone’s spreadsheet. A salesperson remembers a customer cutting a process from three days to four hours, although nobody is quite sure where the number came from.

Meanwhile, Marketing is producing another ebook.

The problem isn’t necessarily the absence of proof. The problem is that nobody operates proof as a system.

If we’re going to make a claim important enough to build our positioning around, there should be evidence behind it, and somebody should own that evidence. We should know where it came from, when it was verified, which customers it applies to and when it becomes stale. More importantly, we should know what buyer concern it is supposed to resolve and whether Sales can actually find it when that concern appears in an opportunity.

In the AI era, there is another requirement: the evidence has to be understandable outside the company. A buyer needs to be able to find it, forward it and defend it. Increasingly, an AI system also needs to be able to retrieve it, understand what it proves and connect it to the claim being evaluated.

That’s Proof Ops. It isn’t a folder full of case studies. It’s an operating discipline for evidence.

The Trust Stack Tells Us Where the Proof Has to Work

This distinction also makes the Trust Stack more useful.

A complicated B2B purchase isn’t one decision. It’s a series of people trying to eliminate different kinds of risk. The business sponsor is trying to determine whether the product will actually produce the promised outcome. IT wants to know whether it will fit into the existing environment. Security is looking for the thing that could create unacceptable exposure. Finance wants to know whether the economics survive scrutiny. Procurement wants to know whether the vendor and agreement are defensible.

And somewhere inside the organization is usually a champion asking the most human question of all: Am I going to regret putting my name on this?

The Trust Stack organizes evidence around those risks.

A quantified customer outcome can reduce performance risk. A documented implementation methodology can reduce execution risk. Security architecture, certifications and governance practices can reduce technical and organizational risk. A credible ROI model can reduce financial risk. References, reviews and third-party validation can reduce vendor risk and, importantly, the personal risk felt by the person advocating for the purchase.

This is where I had blurred two ideas together.

The Trust Stack doesn’t create the evidence. Proof Ops does.

The Trust Stack gives that evidence a job to do.

That’s more than a wording change. It turns the Trust Stack from a collection of things a company ought to have into part of an operating system. Now I can start with a positioning claim, identify what would prove it, determine which buyer risk that proof addresses, distribute it to the places where the decision is being made and measure whether buyer behavior changes.

AI Changes Who Encounters the Proof

There is another reason this matters now. The buyer may no longer be the first audience evaluating your evidence.

The machine may be.

A buyer researching a new category can ask an AI system which vendors are credible, which products work with companies like theirs, which platforms integrate with their existing environment, which vendors have credible security practices and what implementation risks they should expect.

That changes what it means to be discoverable.

For years, marketers concentrated on being found. SEO was largely a competition for visibility and clicks. In the AI era, visibility remains important, but the bar is getting higher. You need to be found, understood, trusted and, ultimately, recommended.

Being understood is not the same as being trusted.

An AI system may understand perfectly well that your company sells an enterprise AI platform. It may understand your ICP, your category and your positioning. But when someone asks which vendor has demonstrated the fastest implementation, what evidence is available to support your answer?

That’s the Proof Ops problem.

If your best customer result is buried in a PDF three clicks deep on your website, your implementation expertise exists mainly in the heads of Professional Services, your security posture appears only after an NDA and your strongest customer story is something one salesperson tells during a demo, then you don’t have a content shortage.

You have a proof distribution problem.

The evidence surrounding the company needs to become part of the information environment from which both people and machines form judgments. Customer stories matter. Documentation matters. Security pages matter. Technical material matters. Reviews matter. Analyst coverage matters. Independent references matter. What matters even more is that these things reinforce the same claims with evidence that is specific, current and verifiable.

That’s the connection between Proof Ops and GEO that I hadn’t made strongly enough in the original framework. Winning the AI shortlist isn’t simply about getting your company mentioned. It’s about making it possible for the machine to understand why you belong there and to find evidence supporting that conclusion.

Proof Has to Reach Revenue

This is where my imaginary professor’s criticism becomes particularly useful.

If I’m going to describe this as an operating model rather than an interesting marketing framework, I have to be willing to test it.

The underlying hypothesis is simple: better evidence should reduce perceived risk. Reduced risk should help buyers move. And if buyers move, eventually we should see the effect in revenue.

That doesn’t mean inventing a new vanity dashboard called “Proof Engagement.”

It means looking at the places where uncertainty slows actual opportunities.

If we improve the security evidence available before a deal reaches formal security review, does the review cycle get shorter? If we make implementation evidence available earlier, do fewer opportunities stall when the buyer starts asking what deployment will really require? If salespeople introduce quantified proof from comparable customers, do those opportunities progress differently? Over enough opportunities, do we see changes in SQO conversion, stage velocity, win rate or sales-cycle length?

Those are questions worth answering because they connect marketing activity to buyer behavior.

They can also tell us when we’re wrong.

Perhaps we’ve created twelve implementation assets and buyers still hesitate. Maybe the evidence isn’t credible. Maybe Sales introduces it too late. Maybe we’re proving something buyers don’t actually care about. Or perhaps the uncomfortable answer is that the positioning claim itself isn’t important enough to influence the decision.

That is what makes Proof Ops interesting to me. It isn’t simply a better way to produce case studies. Done properly, it becomes a feedback system between what we say, what we can prove, what buyers care about and what actually produces revenue.

The Critic Found the Missing Mechanism

I’ve spent enough time running companies and marketing organizations to know that useful criticism isn’t always the criticism you agree with. It’s the criticism that sends you back to the whiteboard.

That’s what happened here.

The AI critic didn’t convince me that the Trust Stack was wrong. It made me realize that I had described an architecture without making the operating mechanism underneath it explicit enough.

Proof Ops is that mechanism.

Positioning establishes what we want the market to believe. Proof Ops establishes whether we have earned the right to make that claim. The Trust Stack organizes the evidence around the risks that stand between interest and purchase. Distribution makes that evidence available wherever humans and machines are forming their judgments. Revenue tells us whether any of it changed behavior.

That feels much closer to where B2B marketing is going.

We’re moving from a world where producing persuasive information required meaningful time, money and specialized talent to one where almost anyone can manufacture an enormous amount of competent persuasion.

When persuasion becomes abundant, persuasion itself loses value.

Evidence doesn’t.

The scarce asset becomes credible, current, specific and verifiable proof that what you say is actually true.

The AI critic didn’t destroy the framework.

It showed me the piece I hadn’t finished building.