"If you didn't bother writing, why should anyone bother reading?"
That line comes from developer Ibrahim Diallo, who recently published a sharp autopsy of his own AI-assisted blog posts. After rereading articles he'd run through an LLM, he found them "strange" and "bloated with words I would never use." The sentence structures didn't match how he actually writes. His own work felt alien to him.
The specific tells are worth cataloging. Phrases like "Here's the part I find disturbing" and "The irony is not lost on me" kept appearing across posts. These aren't Ibrahim's phrases. They're what he calls the "Mo Samuels" voice: the statistical average of all internet writing, compressed into a single synthetic personality that shows up in blog after blog.
You've Already Read This Voice Today
Once you notice the Mo Samuels pattern, you can't unsee it. That slightly overconfident, mildly conversational, relentlessly structured tone that dominates Medium, LinkedIn, and half the tech blogs you encounter. Every paragraph has a topic sentence. Every section ends with a neat takeaway. The writing is competent in the way a stock photo is attractive: technically fine, completely devoid of a point of view.
The problem isn't that AI writing is bad. It's that it's all the same kind of okay. When every blogger feeds their drafts through the same models, the output converges on a single voice. Individual style gets sanded down. Weird turns of phrase, the ones that actually make writing memorable, get "corrected" into blandness.
This matters for anyone using AI writing tools professionally. Content marketers running blog posts through ChatGPT or Claude for editing are making a tradeoff they might not realize: each pass through the model pulls the text closer to that statistical average and further from anything distinctive.
The Fix Is Obvious but Painful
Ibrahim's solution was to go back and rewrite the AI-touched articles by hand, restoring his actual voice. That takes time, which is exactly the resource AI writing tools promise to save. It's a real tension with no clean answer.
The practical middle ground most writers are landing on: use AI for research, outlines, and fact-checking, but write the actual sentences yourself. Or at minimum, rewrite aggressively after the AI pass instead of just accepting its output. The moment you hit "publish" on lightly-edited LLM output, you're adding to the pile of content that reads like it was written by the same person.
For tools like Grammarly, Copy.ai, and Jasper that market AI writing assistance, this is a product problem worth watching. The models that help the most with structure and grammar are the same ones that flatten voice. Finding a way to preserve individual style while still being useful would be a genuine differentiator. Right now, nobody has cracked it.