Everyone has a theory about why marketing suddenly sounds the same, and almost all of them blame the robots. The complaints have become their own genre. The telltale punctuation. The word "delve." The paragraph that opens by restating your question back to you like a waiter repeating your order. LinkedIn is full of people spotting AI writing the way birders spot warblers, and a whole industry of "humanizer" tools has sprung up to sand the fingerprints off.
I want to offer an uncomfortable correction from inside the industry: your marketing sounded like everyone else's before the models showed up. The robots did not create the sameness. They inherited it.
A few months ago I read dozens of marketing-agency homepages. The same results-driven language kept returning. Nobody's ChatGPT wrote those pages. Humans wrote them over years, each one looking sideways at the others while believing they were differentiating. Estate-planning firms cluster around "peace of mind." Wellness practices cluster around "root cause." SaaS companies run the same small set of ad templates on the same stock-photo faces.
This is not laziness. It is mimetic desire doing what it always does. People copy what they see working, or what looks like it is working, and markets converge. Eugene Schwartz described the mechanics back in 1966: the first brand to make a claim wins with it, then competitors copy it until the claim goes invisible, and the whole category ratchets up the sophistication ladder together. AI did not invent that ratchet. It just spins it faster, because now the copying happens at the speed of a prompt instead of the speed of a rebrand.
Here is the part that matters. Nearly every complaint about AI writing is a complaint about tells. The punctuation, the pet words, the rhythm. Tells are cosmetic, and cosmetic problems attract cosmetic solutions, which is why the humanizer tools exist. Feed them your generic paragraph and they will hand you back a generic paragraph that no longer smells like a model.
A humanized average is still an average. You have not fixed the writing. You have laundered it.
Your buyer does not bounce off your page because the punctuation looked suspicious. Your buyer bounces because the page says the same thing the last four pages said, and none of them said the thing she actually needed to hear. She cannot always articulate what that thing is. She just knows this page is not it, the way you know a stranger's small talk from a friend's question.
Sit with what actually goes into most marketing before a word gets written. A two-paragraph creative brief. An "avatar" that says women 35 to 55, household income above $100K, values family. Three competitor links. A voice note that says professional but approachable.
Hand that input to a junior copywriter and you get the category average, because the category average is the only thing that input contains. Hand it to a language model and you get the category average faster. The model is not failing. It is doing exactly what it was asked: producing the most probable marketing for a market described in the most generic possible terms. Generic in, generic out, at any speed, from any writer, carbon or silicon.
The output sounds like everyone else's because the knowledge underneath it is thin. That is the whole diagnosis. Everything else is symptom management.
James Treadway sells SAT and ACT prep flashcards. His marketing talked the way test-prep marketing talks, scores and credentials and the product, and it produced steady, unremarkable results. Then the research showed what his buyers actually were: parents choosing whichever company made them feel like a good parent. The fear underneath was never a low score. It was failing their kid.
We rebuilt one sales page around that knowledge while the ad budget stayed flat and the order form remained untouched. The order rate moved from about 6 percent to about 10 percent, and monthly revenue for the compared periods moved from about $13,000 to about $26,000. The rebuilt page was the known material change and leading explanation, not proof that every other influence had been ruled out. The full autopsy is here, including the cart-value caveat.
A financial education company we worked with took the same medicine on its ads. Same month, same offer, rebuilt words. Click-through went from 1.78 percent to 2.70 percent, and cost per lead fell from $21.53 to $17.06. In neither case did anyone buy new design tooling or a smarter AI stack. The knowledge changed. Everything downstream followed.
It is not more analytics. Analytics are the surface layer, the record of what buyers did with no record of why. And it is not another round of surveys, because buyers do not report desires they have not noticed they have. Nobody writes "I bought the flashcards because another parent I envy bought them first" on a feedback form.
Thick knowledge comes from reading buyers where they talk to each other with nobody selling to them. Reddit threads. Forum arguments. Reviews where the anger is specific. A few hundred primary sources, read by a person, until the pattern surfaces: who these buyers are copying, what rivalries they are caught in, where their desire is heading, and what they would need to believe before saying yes. That is the hidden layer, and it is findable in any market where humans buy things.
Once you have it, something strange happens to the tooling debate: it stops mattering much. Write by hand, write with a model, either way the words have evidence beneath them. The work stops beginning with an empty page and starts with a source file.
So no, I am not going to tell you to fire your AI. I use software every day and I am not embarrassed about it. I am telling you that the slop discourse has the causality backwards. The tells are downstream of the sameness, and the sameness is downstream of the thin brief. Buy a humanizer and you have treated the punctuation. The page still does not know who it is talking to.
Fix the knowledge and the voice follows, because specificity is the one thing that cannot be faked from an average. A page that says the true, particular thing about your buyer does not sound like everyone else's page. It cannot. Nobody else knows it.
The free Mimetic Intelligence™ Snapshot is a human-built report from source-linked public buyer evidence in your market: one material pattern, the strongest contradiction, and one concrete move for this month. Delivery timing is confirmed after review.
Request a Snapshot