In an era where artificial intelligence tools are increasingly used to generate content, writers often worry that their own work might be mistaken for AI output. One technology journalist decided to confront this concern head-on by asking an AI assistant to analyze his writing for telltale signs of machine generation. The results were both surprising and instructive, revealing a nuanced picture of how human writing can inadvertently mimic AI patterns.

The Experiment

The journalist submitted ten of his most recent articles, totaling approximately 11,700 words, to Claude Sonnet 4.6. He instructed the AI to identify all ways his writing sounded like AI, including the frequency of classic AI tells, and to provide a ranked list. The goal was not merely to pass an AI detector but to understand the structural and stylistic habits that could trigger suspicion in human readers or automated systems.

AI detection has become a cottage industry, with numerous web-based tools claiming to distinguish human from machine writing. However, these detectors often rely on superficial markers like sentence length, word choice, and punctuation patterns. The journalist wanted a deeper, more qualitative analysis that only a sophisticated language model could provide. By using Claude as both the subject and the evaluator, he hoped to gain actionable insights into his own writing quirks.

Top AI Tell: Parenthetical Asides

The number-one red flag was the use of parenthetical asides. The analysis counted 67 instances across the 11,700-word sample, meaning roughly one parenthetical every 175 words. Claude noted that this pattern is 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 AI observed that human writers who do this much usually get edited out of it.

Parenthetical asides are a common tool for adding nuance or secondary information without breaking the flow of the main sentence. However, when overused, they can make writing feel cautious and overly explanatory, traits often associated with AI-generated content. The journalist admitted that this habit has been called out by editors before, but seeing it quantified at 67 instances was a wake-up call.

Em Dash Overuse

Coming in a close second, the analysis found 78 em dash instances, roughly one every 150 words. AI uses em dashes as a crutch to splice clauses that should either be separate sentences or use a comma. While many of the journalist's dashes were fine, the density was high. Em dashes are a favorite tool for adding dramatic pauses and emphasizing key points, but they can become a lazy shortcut for sentence structure.

The journalist confessed to a love of em dashes, and the analysis confirmed that this affection borders on overreliance. In fact, he had initially assumed em dashes would be his biggest AI tell. The experiment showed that while they are a significant factor, parentheticals were even more problematic. This highlights how writers can be blind to their own most conspicuous habits.

Other AI Red Flags

The analysis also identified several other patterns. Long sentences were common, with 21% of sentences exceeding 35 words and an average sentence length of 25 words. AI tends to write long, technically correct sentences that feel like they are covering too much at once. The journalist's writing mirrored this tendency, potentially raising flags for careful readers.

Filler words and hedging language were also noted. The word "actually" appeared 15 times, "rather" 17 times, and "very" 15 times. These are generic intensifiers or hedges that AI (and AI-influenced writing) leans on. At the sample word count, "actually" showed up once every 775 words, which is not extreme but contributes to an overall pattern.

Hedging words like "may" and "might" appeared 14 and 11 times respectively, totaling 25 hedges. While not alarming individually, they add up to a cautious tone that can feel less authoritative. Similarly, impersonal openers like "It is," "It was," "It has," and "It seems" appeared 14 times. Phrases such as "It's worth noting" or "It's clear that" are favorites of AI systems because they allow the model to avoid making a direct claim. The journalist used these constructions frequently enough to be noticeable.

Why Human Writing Still Shines

Despite all these red flags, Claude gave the journalist a low score of 3 out of 10 on the "sounds like AI" meter. The AI elaborated that the vocabulary analysis was the main positive factor. The journalist was completely clean on the words that AI detectors and human readers actually flag. First-person voice was strong, sentence starters were varied and casual ("but," "so," "don't"), and he committed to opinions rather than hedging everything into mush.

The parentheticals and em dashes would read as stylistic quirks to most people, not AI tells. A detector might ding for sentence length, but a human reader would not think "AI wrote this." Claude noted that the writing reads more like a journalist who edits fast and doesn't always tighten—which is a very human problem. This distinction is crucial: AI tells are not inherently bad; they become problematic only when they dominate the style.

The experiment also echoes a broader principle in writing: critique, don't rewrite. Using an AI to analyze your writing can be more productive than asking it to rewrite your drafts. The journalist noted that this approach gave him an honest evaluation without losing his voice. It's a technique that writers can apply to improve their work while retaining authenticity.

For those concerned about sounding like AI, the takeaway is nuanced. Overusing parentheticals, em dashes, long sentences, and filler words can contribute to an AI-like tone. However, a strong personal voice, varied sentence structures, and direct opinions outweigh these quirks. The best defense against sounding like a machine is to write like a human—embracing imperfections, committing to claims, and allowing your personality to shine through.

The journalist concluded that he would try to use em dashes more sparingly and reduce parenthetical asides. But he also recognized that these habits are not fatal flaws. They are simply editing opportunities. As AI continues to evolve, the line between human and machine writing will blur further. The key is not to avoid all AI-like patterns but to ensure that your unique voice remains unmistakably your own.

Source:PCWorld News