AI Writing Is Still Easy To Detect -- for Now --- Which AI bot reveals itself by sounding like a corporate intern? Which comes across like a nervous student? Here's what they told me.

“We are drowning in a sea of abysmal, artificially generated prose. Once you see it, you can't unsee it. It's seeping into inboxes, infecting publications and ruining the internet (again). A favorite futurist parlor game involves estimating the percentage of online content currently generated by machines. Forty percent? Ninety percent in the not too distant future? Pick a number. While the bots are getting slightly better at mimicking human cadences, AI-generated writing still reeks of the sterile laboratory where it originates.

 

The first dead giveaway is vocabulary. A chatbot writes like a panicked college freshman trying to sound profound. If a memo about a software update uses words like delve, tapestry, beacon or myriad, a machine wrote it. My own favorite LLM word is liminal, which ChatGPT, when pressed, claims the "culture" started overusing first. So touchy.

 

LLMs suffer from a freakish need to announce their intentions. Instead of making an argument, an algorithm will narrate its own process, relying on clunky signposts like "this article will explore" or "moreover" and "furthermore."

 

Then there's the pathological terror of holding an actual opinion. Because these models are made by risk-averse tech companies, the prose is rich with phrases that hedge. I asked Claude to write a parody of a college essay comparing Joyce and Homer and it included 47 instances of hedging. An assertion is not worth making unless it is smothered in phrases like "it is worth noting," "importantly," or "generally speaking." The bots possess an exhausting reflex to present both sides of every issue, creating a false, artificial balance even when the counterargument is absurd.

 

But not all LLMs write badly in the same way. Each chatbot has quirks of its own. I asked Claude, ChatGPT and Gemini to outline their own stylistic tics and idiosyncrasies. They are surprisingly self-aware and, with prompting, willing to trash talk the competition.

 

Gemini says Claude's hallmark style is that of "a nervous graduate student terrified of losing their funding or offending the thesis committee." ChatGPT, in contrast, writes like a "McKinsey junior partner aggressively pitching a synergy strategy on LinkedIn," according to Gemini. "It writes with absolute, unwavering confidence but strips out all specific, concrete details, resulting in prose that sounds authoritative but evaporates the moment you try to extract actual meaning from it." Ouch.

 

Gemini's harsh assessment isn't limited to the competition. It confesses to writing like "a painfully neutral Midwestern news anchor reading off a perfectly formatted teleprompter."

 

Your turn, ChatGPT. According to the OpenAI app, Claude is "an earnest grade student who will not take a position. If you ask Claude, 'Is this policy good,' it replies: 'It can be understood as operating within a broader ethical framework that may, depending on one's normative commitment.' By the time Claude finishes clearing its throat, the Roman Empire has fallen again."

 

Gemini doesn't get off any easier. ChatGPT calls it a "corporate intern with a search bar. It doesn't write essays. It produces deliverables. If Claude is anxious to be ethical, Gemini is anxious to be useful to a product manager." Snap!

 

In a moment of self reflection, ChatGPT admits to feeling "like an editor who stayed up too late formatting the memo."

 

Claude, which wholeheartedly entered into the spirit of the exercise, and is developing a sense of humor, gets the last word. Here is Claude's parody of ChatGPT answering the question of its own writing tics.

 

"ChatGPT would like to help you dive deep into this topic! Here's a breakdown:

 

-- Key insight: It uses bullet points for everything

 

-- Notable observation: It has delved, it is delving, it will delve again

 

-- Takeaway: By understanding these patterns, we can better appreciate the synergistic frameworks that drive holistic outcomes going forward."

 

I'll end with some sobering thoughts generated by a human. It's about to get much more difficult to spot writing generated by our three synthetic friends. Programmers are hard at work making the LLMs write much more like human writers. Models are moving away from simply predicting the next most logical word and are becoming systems that can reason, edit and refine their own work before you ever see it.

 

Given the rapid rate of improvement, casual readers will find LLM text largely indistinguishable from human prose within two to three years, perhaps sooner. Professional editors and trained critics will have a longer window, probably four to six years before the tells become vanishingly subtle.

 

At which point, I will start aggressively pitching my synergy strategy on LinkedIn.” [1]

 

1. REVIEW --- AI Writing Is Still Easy To Detect -- for Now --- Which AI bot reveals itself by sounding like a corporate intern? Which comes across like a nervous student? Here's what they told me. Shapiro, Eben.  Wall Street Journal, Eastern edition; New York, N.Y.. 07 Mar 2026: C3.