AI-Generated Copy and the Production of Simulacra
A large language model does not know what it is saying. This is not a moral judgment but a technical description. The model operates on statistical patterns: given a sequence of tokens, it predicts the most probable next token. It has no concept of truth, no experience of meaning, no relationship to the referent. It produces language that resembles meaningful language — that has the form and texture of communication — without any communicative intent. It is, in the most precise Baudrillardian sense, a machine for the production of simulacra.
I want to explore what happens when this machine is applied to marketing — when the generation of brand language is delegated to a system that has no understanding of brands, no experience of products, and no relationship to consumers. Because something happens. Something changes. And I am not sure we have fully reckoned with what it is.
The Copy Before Copy
To understand what AI-generated copy represents, we need to understand what human-generated copy was. A copywriter — a good one, the kind David Ogilvy or William Bernbach would have recognized — performed a specific intellectual operation. They took a product, understood it, understood the audience, and found the language that would bridge the gap between the two. The language was not a transparent window onto the product's qualities; it was a construction, a rhetorical artifact designed to produce a specific effect. But it was a construction made by a person who understood what they were constructing and why.
This understanding mattered because it produced friction. The copywriter who knew the product was mediocre had to work harder to make the language sing. The copywriter who knew the audience was skeptical had to find registers of persuasion that acknowledged the skepticism. The best advertising copy has always been marked by a tension between the commercial imperative (sell this thing) and the human complexity of the person doing the selling. The copy bore the trace of a consciousness — a conscience, in both the French and English senses of the word.
AI-generated copy has no such friction. The model does not know whether the product is good or bad, whether the claims are true or false, whether the audience will be persuaded or offended. It simply produces the most statistically probable arrangement of words for the given prompt. The result is language that is, in a technical sense, perfect: it follows all the conventions of marketing copy, deploys all the right emotional triggers, uses all the appropriate keywords. It is, in Barthes's terms, pure écriture — writing without a writer, language without a subject.
The Uncanny Valley of Language
There is something uncanny about AI-generated marketing copy, and the uncanniness is instructive. Freud defined the uncanny (das Unheimliche) as the experience of something that is simultaneously familiar and strange — something that looks right but feels wrong. AI copy produces this effect because it is, structurally, a simulation of human expression that lacks the thing that makes human expression human: intention, experience, the relationship between the word and the life that produced it.
Read a piece of AI-generated brand copy alongside a piece written by a skilled human copywriter, and the difference is often subtle but unmistakable. The AI copy is smoother, more consistent, more perfectly calibrated to best practices. It never makes an unexpected word choice. It never introduces a surprising metaphor. It never reveals a personality. It is, paradoxically, too good — too perfect in its adherence to convention, too seamless in its construction. It lacks the irregularities that signal human presence, the way a hand-thrown pot differs from a machine-made one.
But here is the disquieting question: does the consumer notice? Does the consumer care? If the function of marketing copy is to produce a specific effect — awareness, consideration, conversion — and AI-generated copy produces that effect with equal or greater efficiency, does the absence of human intentionality matter? Is there a meaningful difference between copy that is meant and copy that merely appears to be meant?
Baudrillard's Prophecy
Baudrillard described four stages of the image's relationship to reality. In the first stage, the image reflects a basic reality. In the second, it masks and perverts that reality. In the third, it masks the absence of a reality. In the fourth — the stage of the simulacrum — it has no relationship to reality whatsoever. It is pure simulation, referring to nothing outside itself.
AI-generated marketing copy exists in the fourth stage. It does not reflect the brand's actual qualities (stage one). It does not distort them (stage two). It does not conceal their absence (stage three). It simply generates a surface of brand language that has no relationship to any underlying reality — because there is no underlying reality for it to relate to. The model has never used the product. It has never met the founder. It does not know what the brand "really" is, because the question of what a brand "really" is presupposes a reality beneath the representation, and the model operates entirely at the level of representation.
This is not a critique of AI tools per se. It is a description of a structural condition. When brand language is produced by a system that operates exclusively on the surface of language — on patterns, frequencies, probabilities — the result is language that is all surface. And when this language is deployed at scale — across thousands of brands, millions of touchpoints, billions of impressions — the effect is a marketplace in which the semiotic environment is composed entirely of simulacra. Signs without referents. Meaning without origin. A desert of the real, populated by perfect, empty words.
The Human Remainder
What is lost when copy is generated rather than written? I think what is lost is the trace — Derrida's term for the residue of meaning that every sign carries from its previous uses, its prior contexts, its history of deployment. When a human copywriter writes "This changes everything," they bring to that phrase their entire experience of language: every time they have heard it used sincerely, every time they have heard it used ironically, every time they have deployed it themselves and felt it ring true or ring hollow. The phrase carries their life within it, even if that life is invisible in the final text.
When an AI generates "This changes everything," it brings nothing. The phrase is assembled from statistical patterns. It carries no trace, no history, no life. It is a perfectly formed shell with nothing inside. And yet it functions. It circulates. It produces effects. The consumer reads it and feels... something. Maybe not the same something they would feel if a person had written it with conviction. But something.
Perhaps this is the most Baudrillardian outcome of all: the discovery that meaning was never in the sign to begin with. That it was always a projection of the reader, a construction of the receiver. That the sign itself — whether produced by a human consciousness or a statistical model — was always empty, and the fullness we attributed to it was our own.
If that is true, then AI has not changed anything. It has merely revealed what was already the case: that marketing language was always a simulacrum, and the human writer was always, in a sense, a machine — a pattern-matching device that produced conventional language in response to commercial prompts.
But I don't quite believe that. Something in me resists it. I just can't say, with any certainty, what that something is.