Jasper, ChatGPT, and the Question of Craft
The tools have arrived. The question is what we do with them.
I have now spent the better part of a year using AI writing tools in various capacities, and I find myself in an uncomfortable position. I am neither the enthusiast nor the Luddite. I am the practitioner who can see what these tools do well, can see what they do poorly, and is troubled by the fact that most of the people using them cannot tell the difference. The tools are not the problem. The loss of standards that the tools enable — that is the problem.
Let me describe what I have observed. Jasper, ChatGPT, and their various competitors can produce copy that is fluent, grammatical, and tonally consistent. Given a brief, they will generate multiple options in seconds. Given feedback, they will revise without complaint. They will adopt any voice, mimic any style, address any audience. They will produce social media posts, email subject lines, product descriptions, and blog articles with a speed and consistency that no human writer can match. These are genuine capabilities, and I do not dismiss them.
What the tools cannot do — what I believe they are structurally incapable of doing — is exercise craft. And here I must be precise about what I mean by craft, because the word is used so loosely in our industry that it has nearly lost its meaning. Craft, as I understand it, is the application of skill and judgement to the solution of a specific problem. It is not fluency. It is not speed. It is not consistency. It is the ability to see the problem clearly, to consider the available solutions, and to choose the one that is not merely adequate but right. The craftsman does not produce the first acceptable option. The craftsman produces the best option, and knows the difference.
The Adequacy Trap
The danger of AI writing tools is not that they produce bad copy. It is that they produce adequate copy — copy that is good enough to use, good enough to publish, good enough to satisfy a client who is not paying close attention. This adequate copy fills the space. It says approximately the right thing in approximately the right way. It does not offend. It does not surprise. It does not fail conspicuously. And because it does not fail conspicuously, it is accepted, and the standard of what constitutes acceptable work drops imperceptibly but relentlessly.
I have seen this happen in agencies that have adopted AI tools for first drafts. The workflow is seductive: the AI generates a draft, the human refines it, the result goes to the client. In theory, this is efficient. In practice, something insidious occurs. The human refiner begins to accept the AI's framing of the problem. The structure the AI suggests becomes the structure of the final piece. The vocabulary the AI chooses becomes the vocabulary of the finished copy. The human is not writing. The human is editing, and editing AI output is not the same as writing from scratch, because the range of possibilities has already been narrowed by the machine's initial output.
This is what I call the adequacy trap. The tool produces something adequate, and adequacy becomes the ceiling rather than the floor. The writer who begins with a blank page must confront the full range of possibilities — the terrifying, exhilarating, infinite range of things they might say and ways they might say them. This confrontation is uncomfortable. It is slow. It sometimes produces anxiety and self-doubt. But it is also the condition in which originality occurs. The writer who begins with an AI draft has been spared that confrontation, and they have been spared the originality that comes with it.
What Hopkins Would Have Made of It
I have been thinking about what the great practitioners of our craft would have made of these tools. Claude Hopkins, I suspect, would have been cautiously interested. Hopkins was a pragmatist above all. He measured everything. He tested everything. He cared about results, not about how those results were achieved. If an AI tool could produce a headline that pulled more responses than a human headline, Hopkins would have used the AI tool without a moment's hesitation. But — and this is the crucial point — Hopkins would have tested the AI's output with the same rigour he applied to human output. He would not have assumed that the tool's fluency was evidence of its effectiveness. He would have measured, and I am fairly confident that the measurements would have confirmed what I have observed: AI copy is adequate, and adequate copy is not what wins the test.
Ogilvy, I think, would have been more sceptical. Ogilvy believed in the primacy of the idea, and ideas — real ideas, the kind that transform a brand — are not generated by pattern recognition. They are generated by deep understanding of the product and the consumer, combined with the creative leap that connects the two in a way that has never been done before. Ogilvy spent weeks researching before he wrote a single word. The research informed the idea, and the idea informed the copy. An AI tool that skips the research and jumps straight to the copy is not replicating Ogilvy's process. It is short-circuiting it.
The Fair Acknowledgement
I want to be fair, because fairness is a virtue I try to practise even when it is inconvenient. AI writing tools are useful. They are useful for generating raw material. They are useful for overcoming the paralysis of the blank page. They are useful for producing high volumes of low-stakes copy — product descriptions, meta tags, social media captions — where speed matters more than originality. They are useful, in other words, for the work that does not require craft.
The industry has always had a great deal of this work. Not every piece of copy is a headline. Not every brief demands brilliance. Some work is functional, and functional work can be produced functionally. If AI tools free human writers from the burden of functional work and allow them to spend more time on the work that actually demands their skills — the headline, the campaign idea, the long-form argument — then the tools will have served the craft rather than undermined it.
But this is not what I observe happening. What I observe is the opposite. The tools are being used not to free writers from low-stakes work but to replace writers on high-stakes work. Agencies are using AI to generate campaign concepts. Clients are using AI to produce brand messaging. Marketing departments are using AI to write the copy that will represent their company to the world. And the results are, predictably, adequate. Smooth, fluent, professional, adequate. The kind of work that fills the space without filling the mind.
The Standard Must Be Held
My concern is not that AI will replace copywriters. Some copywriters will be replaced, and some of those replacements will be justified, because some copywriters produce work that is no better than what the machine can generate. My concern is that the standard of the craft will be lowered — that adequate will become the new excellent, that fluency will be mistaken for skill, that the difficult, slow, painful work of producing genuinely good copy will be abandoned because a faster, cheaper alternative is available.
This is not a new pattern. Every technological advance in advertising has carried the same risk. Desktop publishing allowed anyone to create a layout, and the average quality of layouts declined. Stock photography allowed anyone to illustrate an advertisement, and the average quality of imagery declined. The technology is not at fault. The decline in standards is at fault, and standards are maintained by people, not by tools.
The question is not whether AI can write copy. It can, after a fashion. The question is whether we, the practitioners, will hold the standard of craft that distinguishes good copy from adequate copy, and adequate copy from the kind of fluent, forgettable, friction-free prose that the machines produce so effortlessly. The tools have arrived. What we do with them will determine whether the craft survives.
I intend to keep writing by hand, by which I mean by mind. You may make of that what you will.