According to findings published by Pew Research in June 2026, a study analyzing approximately 490,000 web pages on Common Crawl between January 2021 and July 2026 using the Open Pangram tool revealed a dramatic surge in AI-generated content. The report showed that the proportion of AI text traces on .com domains grew nearly 10 times, from under 1% in 2021 to 9.35% in January 2026 alone—10 times higher than domains like .edu or .gov.
Pew Research also pointed out that the frequency of several hallmark characteristics of AI writing style on web pages increased significantly compared to 2021, specifically:
- The em dash (—): Doubled; this may be the clearest telltale sign, as humans rarely use this mark in articles and chat messages, typically sticking to the standard dash (-).
- The Oxford comma: Usage increased by more than 60%.
- Favorite AI words: Words such as delve, interplay, testament, underscore, pivotal, harness, and others more than doubled.
- Negative parallelism: Phrasings such as “not only X, but also Y” increased by 3 times.
However, we are not here to judge or reject AI-generated content; we are here to confront a reality: using AI at work to boost efficiency and productivity is virtually inevitable. Therefore, those of us at the dawn of the Digital Renaissance should not merely look at the numbers above with anxiety or disgust toward AI-generated material. Instead, we need a more open mindset and a proactive approach to using AI for content production.
As Steph Zinn, an editor at investment fund a16z crypto, wrote in an article, scrutinizing every tiny sign to guess whether a piece of writing comes from a machine is a flawed way of thinking. The right question should be: “Does this piece explain clearly, can the reader trust it, and is someone actually in the driver’s seat making the decisions?” Looking at the issue from the editorial desk, what people label “signs of AI writing” are actually an existing catalog of bad writing habits: sentences that sound momentous without saying anything, overly tidy structures, and a tone so neutral it could be dropped into anyone’s piece without looking out of place.
For Zinn, rather than trying to dissect every detail to “expose” whether a text was produced by AI, we should focus on the fundamental purpose of the content—its quality, whether it conveys the author’s message to the reader, and more importantly: Is an actual human in control and making the decisions behind it?
Whether you use AI or not, the evaluation standards and editorial fixes remain the same; the outcome is still content quality and the message delivered. The traits commonly attributed to AI writing are simply the natural defects of first drafts, and humans have written that poorly just as often as machines.
Another piece by Alex Danco, also from a16z, sheds light on why people react so strongly to AI writing. He acknowledged his own piece was 10% AI-generated and drew on theories from Barthes and Foucault to discuss why the tech community—rather than AI skeptics—is the group most obsessed with unmasking AI prose.
First, Alex pointed out an example: Claude writes “Give the pruning logic real attention” whereas a human writer would opt for “Pay attention to the pruning logic.” Both are grammatically correct, but something about the former feels off. This stems from AI’s token prediction mechanism: while humans decide what they want to say before composing a sentence, the model must generate text token by token until the sentence ends, unable to turn back or revise on the fly. Consequently, it must choose words that keep the widest range of future options open. The “weirdness” of AI writing does not lie in spelling mistakes or cliches, but in its very token-generation mechanics—something that will take a long time to change.
Second, what people react so strongly to when reading AI prose might not be that it looks ugly or violates community standards, but that the work has no author. As the 20th-century French philosopher Foucault argued, authorship is a social function we require to categorize, trust, and assign accountability to a text. AI writing scrambles that function.
Yet paradoxically, the model seems to compensate for this void with intensely confident phrasing, an urge to insist everything is profoundly interconnected—as if it were roleplaying an actual author. This explains why AI writing can feel grating even when factually accurate: it merely pretends someone is standing behind it, while the reader senses that nobody is.
Personally, I believe the defining criterion for judging a piece of writing does not lie in whether it originated from human or machine, but in its ability to achieve its objective. Even so, this does not diminish the intrinsic value of writing bearing a strong personal imprint—a space preserving a distinct voice that no advanced AI model can replicate or reach. It is this uniqueness that enables readers to clearly recognize and connect with the real human behind the page.
For content creators, the sensible boundary lies in openly embracing AI as a capable assistant for editing, structuring, or stress-testing arguments. Conversely, relinquishing the core message and conviction entirely to the tool leaves even a technically flawless piece devoid of soul—the single most crucial element to persuade and retain readers.