Why AI Cannot Write a Headline
It can assemble words. It cannot make a stranger stop.
I have been asked, with increasing frequency over the past two years, whether artificial intelligence will replace the copywriter. The question is usually posed by clients who are hoping the answer is yes, because copywriters are expensive and difficult and occasionally refuse to do what they are told. I understand the appeal. If a machine could produce serviceable copy at a fraction of the cost, with none of the artistic temperament, the economic argument would be irresistible. And so I have tried, with genuine openness, to evaluate what these tools can do. I have spent time with ChatGPT and with several of the specialised copywriting tools that have emerged in its wake. I have given them briefs. I have studied their output. And I have arrived at a conclusion that I suspect will satisfy no one: the tools are impressive, and they cannot write a headline.
I should explain what I mean by "write a headline," because the phrase conceals a complexity that is central to my argument. A headline, in the advertising sense, is not merely a sentence placed at the top of an advertisement. It is the single most consequential piece of writing in the entire campaign. It is the sentence that determines whether the other five hundred words will be read or ignored. It is the sentence on which eighty cents of every dollar is spent, as Ogilvy calculated. It is the sentence that must stop a stranger — a person who is not looking for your advertisement, not interested in your product, not inclined to give you a moment of their time — and compel them to read further. This is an act of persuasion so compressed, so demanding, so unforgiving of error, that it requires not just competence but something closer to intuition.
AI does not have intuition. This is not a criticism. It is a description. The large language models that power these tools work by pattern recognition — by identifying statistical relationships between words and phrases in vast corpuses of text. They are extraordinarily good at this. They can produce sentences that are grammatically correct, stylistically appropriate, and superficially indistinguishable from human writing. But they are producing sentences based on what has been written before. They are generating the average. And the average headline is, by definition, a headline that does not stop anyone, because it is a headline that sounds like every other headline.
The Test I Conducted
To test my hypothesis, I gave one of the leading AI tools a series of briefs drawn from campaigns I had worked on in my career. I gave it the same information I would have received: the product, the target audience, the key proposition, the tone of voice. I asked it to generate twenty headline options for each brief. The tool obliged, efficiently and without complaint, which is already a significant advantage over most copywriters I have worked with.
The results were revealing. The headlines were competent. They were clear. They communicated the proposition. They were, in many cases, better than the worst headlines I have seen produced by human writers. But they were not good. They were not surprising. They did not have the quality I described earlier — the quality of collision, of interruption, of making the reader's eye snag on the page. They were smooth where they needed to be rough. They were expected where they needed to be unexpected. They were, in a word, average.
I compared the AI output against the headlines that had actually run in those campaigns — headlines that I and my colleagues had spent days or weeks developing, testing, refining, and arguing about. The difference was not subtle. The human headlines had specificity that the AI headlines lacked. They had rhythm. They had a point of view. They had, in several cases, a quality of strangeness — of saying something in a way that no one had quite said it before — that made them memorable. The AI headlines had none of this. They had the shape of good headlines without the substance.
Why Specificity Matters
The root of the problem, I believe, is specificity. The great headlines in advertising history are specific. "At 60 miles an hour the loudest noise in this new Rolls-Royce comes from the electric clock." That is not a general statement about quietness. It is a specific, concrete, verifiable claim that paints a picture in the reader's mind. The specificity is what makes it believable. The specificity is what makes it surprising. And the specificity is what AI cannot generate, because specificity requires knowledge — not linguistic knowledge, but knowledge of the product, the market, the consumer, the culture, the moment.
When Ogilvy wrote that headline, he had spent weeks researching the Rolls-Royce. He had read the engineering reports. He had spoken to the designers. He had driven the car. The headline emerged from deep familiarity with the subject, and that familiarity cannot be simulated by a statistical model, no matter how large its training data. The model can tell you what a car headline typically sounds like. It cannot tell you what this particular car is actually like, because it has never driven it, never sat in it, never heard the silence that the headline describes.
I am aware that AI tools can be given product information. You can paste a specification sheet into the prompt. You can provide consumer insights and competitive analysis and brand guidelines. The tools will dutifully incorporate this information into their output. But incorporation is not understanding. The tool does not know which piece of information is the surprising one, the one that will stop a reader, because it does not know what readers find surprising. It knows what patterns of words are common. It does not know what moments of human attention feel like.
The Role of Experience
There is a dimension of headline writing that is almost never discussed, because it is difficult to articulate: the role of accumulated experience. When I write a headline, I am drawing on thirty years of watching people respond to advertisements. I have seen what works and what does not, not in the abstract but in the particular — this audience, this product, this moment, this medium. I have developed an instinct, honed over decades, for the kind of sentence that will stop a stranger. This instinct is not mystical. It is the product of repetition, of paying attention, of learning from failure. But it is mine, and it cannot be transferred to a machine, because it is not stored as data. It is stored as judgement.
The young AI enthusiasts will argue that the machine's judgement will improve with time and data. Perhaps it will. I am not in the business of predicting the future, and the people who are in that business have a poor track record. What I can say is that the current generation of tools, impressive as they are, cannot do the thing that matters most. They cannot write the headline that stops a stranger in their tracks. They can write the headline that fills the space. They can write the headline that satisfies the brief. They can write the headline that a tired creative director, running short on time and patience, might accept as good enough. But good enough is not good, and in advertising, the difference between good enough and good is the difference between invisibility and impact.
I do not fear AI. I do not resent it. I use it, occasionally, as a brainstorming tool — a way to generate raw material that I then shape and refine and, in most cases, discard. But I do not mistake it for a writer. A writer has a point of view. A writer has taste. A writer has the ability to surprise themselves, to produce a sentence they did not expect, to find the phrase that no one has found before. These are the qualities that separate the headline that works from the headline that merely exists.
AI can assemble words. It cannot write a headline. The difference is everything.