{
  "id": 3152018,
  "title": "Can you teach yourself to detect AI writing? Maybe",
  "url": "https://urgent.news/2026/08/25/can-you-teach-yourself-to-detect-ai-writing-maybe",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-25T00:21:51.000Z",
  "source": {
    "name": "The Conversation AU",
    "slug": "the-conversation-au",
    "url": "https://theconversation.com/can-you-teach-yourself-to-detect-ai-writing-maybe-289236"
  },
  "original_language": "en",
  "account": "As AI technology progresses, it is increasingly altering the way we write. This is not merely a new tool, but a genuine revolution in the structure of written language. The introduction to this article, crafted to mimic the style of AI-generated text, was not written by artificial intelligence. Instead, it aimed to emulate the prose created by large language models (LLMs).\n\nThis style of AI-generated writing is now prevalent in various forms, from low-quality websites seeking to attract advertising clicks to social media posts designed to boost engagement, even infiltrating book manuscripts. A notable case in point is crime novelist Jerry Falade who lost a book deal worth over $2 million after his literary agents could no longer accurately determine how his manuscript was produced.\n\nThe challenge lies in identifying AI-generated text. There is no single indicator, but rather a series of patterns that tend to appear in synthetic writing. These include specific words, sentence structures, and stylistic habits that are more common in AI-generated prose. For instance, the word \"quietly\" has seen a significant increase in usage since December 2021, suggesting its frequent use by LLMs rather than human writers. Similarly, words like \"genuinely\", \"delve\", \"meticulous\", \"underscore\", and \"boast\" have become overused in AI-generated writing, as well as in scientific literature since 2022.\n\nLLMs exhibit a distinctive sentence structure, often employing constructions like \"not X, but Y\" and the rule of three. They also favor the use of the em dash, a stylistic element that was once regarded with skepticism. While individual clues may be unreliable, their collective impact is significant. A report from The Atlantic highlights how prolonged AI usage can result in lengthy, polished prose that lacks the urgency of human-generated text. In contrast, human language is influenced by the constraints of time and effort, leading to concise and straightforward communication.\n\nThe dangers of attempting to detect AI-generated text are substantial. Current detection software is far from perfect and can perpetuate existing biases. A Stanford study revealed that seven popular AI detectors falsely identified 61% of essays written by non-native English speakers as AI-generated, with one tool flagging 97% of the time. However, the focus is shifting from detection to labeling or watermarking AI-generated content. This is particularly relevant in the context of the European Union, which is now mandating that all AI-generated content be labeled or watermarked.",
  "summary": "There’s no single giveaway. But there are patterns.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}