I don’t believe AI should stay out of writing. I believe most people configure it badly, then point at the result and call it proof that AI can’t write. The slop is real. The conclusion is wrong – what they are looking at isn’t the limit of the technology, it’s the limit of the setup behind it.
This post covers why AI-generated content reads as generic, how to actually configure an AI system to write in your brand voice, why good configuration amplifies expertise rather than replacing it, where the real human work sits once the system is built, and what separates Standstill agencies from STANDOUT agencies when it comes to AI and brand voice.
AI Slop Is a System Failure, Not an AI Failure
A large language model has hundreds of billions of parameters and has been trained on a vast slice of the internet. Left to its own devices, with no steer, it does the most statistically average thing possible. It writes like the midpoint of everything it has ever read. That is what slop is – the average of the internet, handed back to you with confidence.
So when someone opens a blank chat window, types “write me a LinkedIn post about AI,” and gets something hollow, they have not discovered that AI can’t write. They have discovered that an unconfigured system produces unconfigured output. The model did exactly what it was asked. The instruction was the problem. This matters because the wrong diagnosis leads to the wrong decision. Agencies that conclude “AI writing doesn’t work” quietly stop investing in it, while their competitors who reached a different conclusion build systems that compound. The value was never the model. Everyone has access to the same models. The value is entirely in the configuration – what you tell it to be, what data you point it at, and what you forbid it from doing.
“An unconfigured system produces unconfigured output. The model did exactly what it was asked – the instruction was the problem.”
How to Configure AI to Sound Like You
Configuring voice is not a single prompt. It is a system you build in layers, and each layer closes the gap between “average of the internet” and “sounds like you.” Across the audits I’ve run, the agencies producing genuinely on-brand AI output have most of these in place. The ones producing slop have none.
Persona: tell the system who it is writing as, not just what to write – without this, it writes as nobody, and nobody has a voice. Corpus: feed it your actual writing, ideally work produced before you ever used AI, so it learns from you and not from a feedback loop of its own output. Retrieval: use retrieval-augmented generation so the system pulls from your material when it writes, rather than reaching for generic patterns. Beliefs: interview the system on how you actually think, because voice is not phrasing – it is what you believe and how you reason. Constraints: give it a list of what it must never do, and strip out the tells – the overused em dash, the reflexive “it’s not X, it’s Y,” the tidy rule-of-three. These aren’t AI inventions; they are devices good writers use deliberately and AI uses compulsively. Do this properly and the system stops imitating “good writing” in the abstract and starts reproducing yours. It was configured by you, on your material, against your standards, so it sounds like you. Not like a freelancer’s interpretation of you. Like you.
Why Configuration Amplifies Expertise Instead of Replacing It
Here is the part most of the “AI is coming for writers” conversation misses entirely. The person who can configure an AI writing system best is the person who already knows what good writing looks like. To tell a system to remove the AI tells, you have to recognise them as tells in the first place. A copywriter brings all of that knowledge to the setup. Someone who has never studied how language persuades brings none of it, points the tool at nothing, and gets the average back.
This is why AI amplifies competence rather than flattening it. The tool raises everyone’s execution speed, but it raises it from wherever they already stand. A skilled writer configuring AI gets leverage. An unskilled one configuring AI gets faster slop. Same tool, opposite outcomes, and the variable is the human, not the model. My background is in psychology, and this is the bit that fascinates me. People reach for AI hoping it will compensate for a gap in their own ability, and they are consistently disappointed, because that is not what it does. It multiplies what is already there. If the expertise is present, AI makes it scale. If it is absent, AI makes its absence scale too. The question was never “will AI replace me.” It was always “how good am I at directing it.”
The Real Work Moves to Editing
Once the system is configured, the centre of gravity shifts. You stop being the person who produces the first draft and become the person who judges it. AI speeds up execution. It does not, and should not, replace judgment. This is the human-in-the-loop principle, and it is where quality actually comes from. The model gets you to a structured draft in seconds. You then do the work that requires taste: cutting the lines that don’t sound like you, sharpening the argument, deciding what to keep.
There is a compounding effect here too, and it is the layer most people skip. Every edit you make is a signal. Feed those edits back into the system – the lines you cut, the phrasings you swapped, the structure you reordered – and the configuration tightens over time. The voice gets closer with every piece, because you are training it on your own judgment, not just your old work. And if you genuinely don’t want AI writing your sentences, you don’t have to let it. Use it further upstream – get it to generate angles you hadn’t considered, or to structure the piece, then write every word yourself. The skill didn’t disappear. It moved – from the blank page to the edit, which is a more valuable place for it to live, because it is harder to fake.
The Bottom Line
The clearest line I can draw runs through two STANDOUT levers. Under Uniqueness, an agency publishing unconfigured AI output is actively eroding its own differentiation – it now sounds like every other agency using the same default model, while an agency with a configured voice system scales its distinctiveness instead. Under Technology, the question is whether you have built the configuration – the personas, the corpus, the retrieval, the constraints – or simply bought a subscription and called that an AI strategy. A Standstill agency opens the tool and accepts the default. A STANDOUT agency configures it and owns the output. The tools are the same. The setup is the entire story.
Frequently Asked Questions
Can AI really replicate my brand voice?
Yes, but only if you configure it. Feed it a corpus of your own past writing, give it a clear persona, use retrieval so it references your material, and tell it explicitly what never to do. Without that setup it reverts to the statistical average of its training data, which is why it sounds generic by default.
Why does AI-generated content sound so generic?
Because an unconfigured model defaults to the midpoint of everything it was trained on. Generic input produces generic output. The fix is not a different tool or a cleverer prompt in isolation – it is building a configured system around the model so it has your voice, your data, and your constraints to work from.
Will AI replace copywriters?
No – it amplifies existing skill rather than removing the need for it. A trained writer configures and edits an AI system far better than someone without that judgment, because they know what good writing looks like and what to strip out. AI raises your execution speed from wherever your skill already stands.
What is the most important step in configuring AI for brand voice?
Feeding it a corpus of your own pre-AI writing, then feeding your edits back over time. That trains the system on how you actually write and think, rather than on generic style instructions. The corpus sets the baseline; the ongoing edits are what make it compound.
