Does my writing sound like AI? Claude has ideas
Abstract created by Good Solutions AI
In abstract:
- PCWorld explores how human writing can exhibit AI-like traits, utilizing Claude Sonnet 4.6 to investigate writing patterns for potential AI tells.
- The evaluation recognized parenthetical asides as the highest AI-like trait, adopted by em sprint overuse and lengthy sentences exceeding 35 phrases.
- Regardless of detecting these patterns, Claude rated the human writing as principally genuine with solely a 3/10 AI similarity rating, suggesting stylistic quirks shouldn’t be mistaken for synthetic intelligence.
What might be higher than an em sprint? They pepper my writing, including dramatic pauses and emphasizing my largest factors. I really like em dashes — so does AI.
The use — and overuse — of em dashes is, in fact, a traditional AI writing inform, together with such AI writing quirks because the “it’s simply this; it’s that” building (which itself will be mixed with an em sprint) and prolonged sentences that appear to stretch into infinity.
Like loads of different on-line writers, my articles are commonly accused of sounding like AI. Whereas web-based AI detectors have repeatedly assured me that my tales sounded human, I wished extra particulars — and most of all, I wished to find if my em-dash obsession was the writing crutch that drew essentially the most “you’re AI!” fireplace.
For the check, I turned to Claude, feeding 10 of my most up-to-date articles to Claude Sonnet 4.6 (utilizing the Claude Cowork) software and giving it a easy immediate:
Are you able to have a look at my writing samples within the listing and record all of the methods through which I sound like AI, together with how usually I exploit traditional “AI tells” in my writing? Give me a ranked record.
I braced myself for the dangerous information, anticipating that overused em dashes (as I beforehand talked about) can be my largest — if false! — AI inform.
Seems, not fairly. Right here’s what Claude ranked as my number-one “sound like AI” problem:
The parenthetical asides — 67 in ~11,700 phrases.
It’s the one sample that’s each frequent and structural. Lengthy sentences generally is a dashing drawback, em dashes generally is a type alternative, however always tucking clarifications into parentheses as a substitute of committing to the sentence reads like AI hedging its bets — saying the factor whereas additionally quietly footnoting the exceptions. Human writers who do it this a lot often get edited out of it.
Yep, it’s true, I’m a sucker for an excellent parenthetical, and it’s a behavior my editors have (fairly rightfully) referred to as me out for. (I do know, I’m doing it once more.)
However relaxation assured, em dashes got here in an in depth second:
Em/en sprint overuse: 78 situations — roughly one each 150 phrases. AI makes use of em dashes as a crutch to splice clauses that ought to both be separate sentences or use a comma. Lots of yours are nice, however the density is excessive.
And right here’s the remainder of the record:
- Lengthy sentences: 21% of your sentences exceed 35 phrases, and your common is 25 phrases. AI tends to jot down lengthy, technically-correct sentences that really feel like they’re protecting an excessive amount of without delay.
- “Truly” x15, “somewhat” x17 — each are filler hedges that AI (and AI-influenced writing) leans on. At your phrase depend, “truly” reveals up as soon as each 775 phrases.
- “Might” x14 + “may” x11 = 25 hedges — not alarming, however AI tends to hedge somewhat than commit. These add up.
- “It’s/was/has/appears…” constructions: 14 occasions — the traditional impersonal opener. Phrases like “It’s price noting,” “It’s a snap to,” “It’s clear that” — AI loves beginning sentences this fashion as a substitute of constructing a direct declare.
- “Very” x15 — generic intensifier that doesn’t add that means. Simple to chop.
Responsible as charged.
The excellent news, in response to Claude, is that my writing usually does sound human, rating simply 3 out of 10 on the “seems like AI” meter. Claude elaborated on my rating, tossing in a couple of em dashes of its personal for good measure:
- The vocabulary evaluation is the primary factor — you’re utterly clear on the phrases that AI detectors and human readers truly flag. First-person voice is robust, sentence starters are different and informal (“however,” “so,” “don’t”), and also you decide to opinions somewhat than hedging the whole lot into mush.
- The parentheticals and em dashes would learn as stylistic quirks to most individuals, not AI tells. A detector may ding you for sentence size, however a human reader wouldn’t assume “AI wrote this.”
- If something, your writing reads extra like a journalist who edits quick and doesn’t at all times tighten — which is a really human drawback.
Whereas I’m happy to listen to I principally don’t sound like AI, Claude’s final level is a good one – certainly, it echoes final week’s “critique, don’t rewrite” immediate, which makes an AI offer you an trustworthy analysis of a writing pattern.
In any occasion, I’ll attempt utilizing these em dashes extra sparingly sooner or later, and the identical goes double for parentheticals.
That’s not simply wishful pondering; it’s a promise.

