Is Copywriting Dead? The Part AI Can’t Touch
“Is copywriting dead?” is one of the most-searched questions in the field right now, and the people typing it are not being dramatic. A copywriter posted last week that her referrals dried up overnight and she can’t seem to land new clients. Another said the whole market is suffocating under AI output. A third put it flatter: even if you can do a better job, you can’t compete with instant and $200 a month.
I want to answer the question honestly, because the panic is pointing at something real. The thing that’s dying just isn’t the thing most people think it is.
Producing words is dying. Having something worth saying is not. Those used to be two different jobs that lived in the same paycheck, and AI just pried them apart.
Is Copywriting Dead, or Did Only the Easy Part Die?
The honest version of the fear sounds like this: a tool that costs less than a tank of gas a month can produce a clean, grammatical, on-format draft in eleven seconds, so why would anyone pay a human to do it slower?
That’s not a strawman. It’s true for one specific layer of the work. The output layer. The first-draft, fill-the-page, make-it-grammatical layer. That layer collapsed in price, and it isn’t coming back. Pretending otherwise is how careers stall.
Here’s what the fear misses. The output was never the scarce part. (It felt scarce because producing it took hours, and we price things by how long they take.) The scarce part was always the judgment underneath: knowing what the reader actually fears, which objection to answer first, what to leave out. AI raised the base level of the output to “competent.” It did nothing to the judgment. If anything it made judgment more visible, because now everyone has competent output, and the only thing separating two pages is whether one of them has a point of view.
One copywriter’s version of this got the most love of the week, and another reader said outright he was going to steal it: the words were always the cheap part. Having something worth saying was the job.
The “raised the base level” shift
One of the calmer voices in these threads framed it as a ladder. AI raised the base level at which paying a human is worthwhile. Below that line, the work is now free and instant. Above it, the work got more valuable, because the floor everyone stands on rose. The uncomfortable instruction inside that frame: be honest about where you are on the ladder, then climb.
This is also why “cheap tool versus expensive human” is the wrong fight. Price was never the thing sorting good work from bad. A $200-a-month tool and a blank ChatGPT window both produce the average. What sorts good work from bad is whether anyone brought something worth generating the words about.
The Mechanism: AI as a Force Multiplier, Not a Replacement
There’s a phrase that keeps showing up from the people who aren’t panicking: force multiplier. They describe AI working best as a force multiplier, not just a chatbot. That sounds like a slogan until you make it concrete, so here’s the concrete version.
A large language model is, mechanically, a machine for predicting the most statistically likely next word. That is its gift and its ceiling in the same sentence. Ask it to write your homepage and it hands you the most average possible homepage, the one that (as a marketer who reviewed fifty SaaS sites put it) could be swapped with any competitor and no one would notice. That isn’t a bug. You asked a machine for the average, and it gave you the average.
The force-multiplier move is to stop asking AI for the answer and start feeding it yours. The work splits into three jobs, and you keep the first one:
- You bring the point of view. The take, the specific reader, the objection you’ve decided to answer, the story only you have. This is the part that can’t be prompted out of thin air, and it’s the part the panicking copywriter still owns.
- AI handles the production. Drafting, restructuring, ten variations of a subject line, turning the voice memo you rambled into a tight outline. The eleven-second layer.
- You bring the judgment back. Cut what’s generic, restore what’s specifically you, kill the line that sounds like every other site.
The writers describing themselves bitterly as “AI Output Reviewers” are stuck doing jobs two and three for output that never had a job one. They handed the machine generic instructions, got the statistical average back, then spent more time auditing it than it saved. (I have done exactly this. It is miserable, and it is avoidable.) The fix happens before you open the chat: bring a point of view the average can’t generate, and the model has something to work with besides itself.
What This Looks Like When You Actually Do It
In practice, the writers staying valuable are doing one thing the threads keep circling back to: they stay close to how people actually talk, what they complain about, and what offers are circulating. They collect the raw material a model can’t invent, then use AI to shape it fast.
A working version of that loop:
- Keep a running file of the exact words your audience uses for their problem. Not your tidy marketing version. Their words. (Half my best hooks are lifted verbatim from a DM.)
- Bring that file, plus your actual take, to the model. Then ask it to draft.
- Read the draft for one thing first: where does this sound like anyone? Cut there. Replace with the specific.
The trap is the one the original poster fell into: treating AI as the thing that has the opinions. It doesn’t have opinions. It has averages. When you supply the opinion and let the model handle the volume, you get faster without getting generic. When you ask it to supply the opinion too, you get the suffocating sameness everyone is complaining about, just produced more efficiently.
This is the whole reason I build custom GPTs that hold a specific operator’s voice and point of view instead of starting from the blank, average-generating default. A GPT that already knows your reader, your take, and the way you actually phrase things still leaves job one to you. What it does is keep jobs two and three from sanding your voice off along the way. The point of view stays yours. The machine just stops flattening it. (If your AI content keeps coming out sounding like everyone else, that’s almost always a job-one problem wearing a prompt-engineering costume.)
So, is copywriting dead? The version that was only ever a words-per-hour machine, yes. The version that knows what’s worth saying and uses AI to say it at scale just got harder to replace.
(This is the writing-specific slice of a bigger question every service business is asking: will AI replace you, and how to use AI without sounding generic.)
