A recent self-audit using Claude revealed that parenthetical asides - not em dashes - topped the list of writing habits that make prose sound AI-generated. The analysis, which examined roughly 11,700 words across 10 articles, ranked the most frequent offenders and offered a blunt verdict: the writing scored just 3 out of 10 on the "sounds like AI" meter, but the parenthetical problem was structural, not stylistic.
The experiment fed a simple prompt to Claude Sonnet 4.6: "Can you look at my writing samples in the directory and list all the ways in which I sound like AI, including how often I use classic AI tells in my writing? Give me a ranked list."
Claude identified 67 parenthetical asides across the sample - roughly one every 175 words. The AI explained why this pattern stood out. "It's the one pattern that's both frequent and structural. Long sentences can be a rushing problem, em dashes can be a style choice, but constantly tucking clarifications into parentheses instead of committing to the sentence reads like AI hedging its bets - saying the thing while also quietly footnoting the exceptions."
The full ranking of AI-like habits
Em dashes landed in second place with 78 instances - about one per 150 words. Claude noted that "AI uses em dashes as a crutch to splice clauses that should either be separate sentences or use a comma."
Sentence length came third. Roughly 21% of sentences exceeded 35 words, with an average of 25 words per sentence. The remaining red flags included filler hedges like "actually" (15 instances) and "rather" (17 instances), hedging words such as "may" and "might" (25 combined), and the classic impersonal opener - "It is/was/has/seems" constructions - which appeared 14 times. The generic intensifier "very" showed up 15 times.
What Claude got right about human writing
Despite the list of tics, Claude's overall assessment was reassuring. The vocabulary analysis came back clean on words that AI detectors and human readers "actually flag." First-person voice was strong, sentence starters varied and casual, and the writing committed to opinions "rather than hedging everything into mush."
Claude added: "The parentheticals and em dashes would read as stylistic quirks to most people, not AI tells. A detector might ding you for sentence length, but a human reader wouldn't think 'AI wrote this.' If anything, your writing reads more like a journalist who edits fast and doesn't always tighten - which is a very human problem."
Why this matters for writers
For anyone who writes professionally, this experiment points to a practical use of AI that goes beyond generation. Running your own writing through an AI for Writers audit - not to detect AI content, but to surface patterns you might not notice - can reveal the crutches that editors see and readers feel. The goal isn't to purge every em dash or parenthetical. It's to know which habits are choices and which are autopilot.
Your membership also unlocks:
