A testimonial landed in my inbox last week. Warm subject line. First-name greeting. A short story about a problem solved, told in the cadence of a person, not a brand. The kind of email I would have forwarded internally a year ago as a model of voice.

I did not forward it. I read it twice and felt a small, useful suspicion. The cadence was too clean. The arc closed too neatly. The story carried the shape of intimacy without the texture of it.

My read is that this small moment, replicated across millions of inboxes, sermons, testimonials, reviews, and DMs, is what the customer trust problem actually looks like now. The genre used to do the work. The form of a sermon, the structure of a confession, the rhythm of a review, the warmth of a personal message: each of these used to carry a presumption of source. AI can produce all of them at scale, at quality, in any tone, by Tuesday. The presumption is gone.

The genre was the trust

The Economist recently observed that American sermons are starting to function as op-eds, borrowing the cadence of moral authority while operating in the logic of public argument. The same outlet ran a review of J.D. Vance's new book that takes him at his own scriptural word, quoting Matthew on judging a person by their fruits, and then asks which fruits the public record actually shows.

This is not a complaint about politicians or pastors. It is a clean look at a mechanism. A high-trust genre, in this case the sermon and the religious memoir, can be repurposed to do something other than what the form historically promised. The audience still hears the cadence and reaches for the old presumption. The speaker still gets the credit of the form.

What AI changes is not the existence of the maneuver. It is the cost of doing it well, at scale, in any register. A junior marketer with a frontier model can now produce confession-grade intimacy, expert-grade testimony, pastor-grade acclaim, and best-friend-grade warmth in an afternoon. The form is cheap. The presumption it carries is the actual product.

A short taxonomy of genres under pressure

A few patterns are worth naming, because once you see them you cannot stop seeing them in your own campaigns.

Moral authority. Sermons, manifestos, founder letters, mission pages. The form claims a higher register and asks for assent on those terms. AI can produce the cadence without the conviction, and increasingly does.

Testimony and expertise. Case studies, white papers, expert reviews, analyst commentary. The form claims earned knowledge. AI can produce the shape of expertise from a brief, complete with hedges and qualifications, with no underlying experience.

Acclaim and social proof. Reviews, testimonials, quote walls, press hits. The form claims aggregated outside judgment. AI can produce a thousand voices that read like a thousand customers and were authored by one prompt.

Intimacy and personal connection. DMs, personalized emails, support replies, romance. The form claims a person on the other end. The persistence of romance fraud, driven on one side by precarity and on the other by loneliness, is the canary here. A two-sided market for synthetic intimacy is already running at industrial scale, and the marketing version is the polite cousin.

These are not separate problems. They are the same problem in different costumes. Each is a presumption that AI has decoupled from its source.

The fruits test, as a design rule

The Vance review borrows a useful line: by their fruits ye shall know them. As Stoic operating discipline, that is just evidence-first assent in older clothing. Marcus Aurelius wrote that a rational nature assents to nothing false or uncertain and directs its movements to social acts. Epictetus opens the Enchiridion by sorting what is in our power from what is not. Both are saying the same thing the Matthew line says in a different idiom. Do not credit the form. Look at what it produces, and adjust your assent accordingly.

The interesting move for marketers is to make the fruits test a design rule rather than a private virtue. Ask, of any claim in any campaign: what measurable outcome would make this story collapse fast enough to protect the customer's decision?

That is a falsifiability test, written backwards. Not what proves the claim, which is easy to dress up, but what would disprove it, which is harder to fake. A testimonial that names a customer who will take a call. A case study with a number a prospect can verify against their own ledger. A founder letter whose claims can be checked against the product the prospect just used for ten minutes. A support agent whose escalation path is visible from message one.

The campaigns that will hold up are the ones built so that the customer can find the disproof if it exists. That is not a compliance posture. It is a trust architecture, and it is becoming a competitive feature.

What this makes possible

The opportunity here is not defensive. It is a new layer of customer experience design.

When genre cues stop carrying presumption, the brands that win are the ones that move verification inside the experience. Provenance becomes part of the creative, not a footnote under it. This is the deeper version of what I argued about the label becoming part of the ad: synthetic disclosure is the early case, but the principle extends. Every claim that used to ride on form needs a contestable proof path the customer can take if they want one.

It also reshapes what the adaptive storefront actually needs to do. Pre-sorted intent is only half the work. The other half is making the proof of the claim available at the moment of decision, in the same surface, without making the customer leave to verify. Trust verification becomes a UX problem before it is a content problem.

For agents, the same logic applies more sharply. A fluent agent that cannot point to the evidence behind its claim is, as a friend put it in a public LinkedIn thread on agentic marketing, a fluent liar with a calendar invite. The architecture that fixes this is not more persuasive output. It is context provenance and proof requirements wired into the agent before it ever opens its mouth.

The Juneteenth counterpoint

The same Economist newsletter that flagged the sermon shift ran a piece on Juneteenth, anchored in Nicole Taylor's cookbook and the idea that freedom shows up through ritual, food, and care. It is a useful counterpoint because it is the opposite mechanism.

A ritual you cook, eat, and share with people you know cannot be faked at scale without immediately failing the fruits test. The watermelon is either on the table or it is not. The family is either there or it is not. The meaning is contestable by lived experience, which is the same property a well-designed customer journey should have. The claim and the proof occupy the same room.

Marketers do not need to stage holidays. They do need to design moments where the claim and its proof occupy the same room. That is what owned meaning looks like in commercial form.

The question for the next review

The thing I would watch, in your next campaign review, is not the headline or the hook or the voice. Those are easier than ever to make excellent. They are also easier than ever to fake, by you and by your competitors and by anyone with a model and an afternoon.

Watch the proof layer. For each claim in the deck, ask one question: what outcome would falsify this story, and can the customer find that outcome quickly enough for the answer to matter?

If the answer is no, the campaign is still riding on genre. Genre is no longer enough. It has not been for a while. AI just made the gap impossible to ignore.

Sources and further reading