Apparently, texts written with artificial intelligence are impeccable. They are fluid, educated and stylistically correct. However, behind that bright finish a disturbing phenomenon hides: the homogeneization of language. The writing generated by models such as Chatgpt, Claude or Gemini is replicating a series of narrative patterns that make articles, emails, posts in social networks or entries of Wikipedia begin to sound suspiciously the same.
To alert about this trend, precisely the editorial community of Wikipedia-one of the most vulnerable spaces to the artificial content not declared-has published a detailed list with the main “Signs of writing generated by AI”. The document aims to serve as a guide to detect synthetic texts and constitutes a deep stylistic analysis of how new linguistic models write today.
When everything is “captivating”, “diverse” and “with meaning”
One of the most forceful findings is the tendency of the LLMS (Large Language Models) to excessively emphasize the symbolism and importance of any matter, however. A people can “represent the resilience of a community”, a park “reflect the ecological renewal”, or a simple location “enjoy a strategic location that makes it a dynamic and cultural hub.” Everything seems to have a transcendent purpose, although in many cases, simply, there is no.
The phrases that inflate the narrative can come to uniform perception
These types of phrases that inflate the narrative can come to uniform perception. And it is that nature is always “impressive”, cities are “vibrant”, “majestic” animals and “rich in heritage” cultures. It is a discursive template that the LLMS replicate naturally, but that empties of nuances to language.
Connectors and summaries overload
Wikipedia editors also warn about the repetition of structures that reveal the mechanical nature of the text. The connectors such as “on the other hand”, “also” or “in summary” appear with excessive frequency, as well as the forced conclusions: “In conclusion, this advance represents a milestone in the history of …”. This habit of closing each paragraph with an unnecessary epilogue is a characteristic feature of the prose generated by AI.
Something similar occurs with negative parallel constructions. For example: “It is not only an excellent culinary option, but also an example of local entrepreneurship.” Or with the so -called three rule (“creative, innovative and versatile”), an effective resource in human rhetoric when adjective, but overexploited by algorithms to the cartoon.
Surface with perfect grammar
Perhaps the most worrying aspect does not reside so much in style and content. LLMS tend to camouflage superficial explanations in a grammatically perfect form. They usually conclude phrases with participles that appear analysis (“… improving the user experience”, “… reflecting the importance of the ecosystem”), although there is no real argument behind. The result: a correct but empty prose.
This type of writing also resorts to vague attributions – “Some experts claim …”, “various reports point …” – without a clear source, expanding ambiguity and reducing the verifiability of the content.
The publication of this list by Wikipedia responds to its open system and its value as a source of digital authority
The publication of this list by Wikipedia responds to its open system and its value as a source of digital authority, which is now exposed to a growing risk due to the automated use of AI to manipulate or fill in items, either for promotional, reputational purposes or simply by human laziness. Since search engines such as Google positively value the presence on the platform, some brands, despite being explicitly prohibited, they are using AI models to edit articles on themselves or to sow irrelevant contributions that divert attention.
Volunteer editors have had to learn to detect these patterns. And what they have found is a set of signals that transcend the way to reveal a substantive issue: the quality of the information. Although sometimes the surface is polished, the content generated by AI usually lacks depth, criteria or context.
In this sense, the Wikipedia guide works as a mirror in an environment where narrative efficiency is standardized by algorithms and writing well requires more than ever a renewed consciousness of the style, tone or background of the articles.
Although, it is possible to use LLMS without falling into the cliché trap: it is enough to rewrite, clarify, suppress the automatisms or, as the article itself suggests, with a long -term trick: copy and paste the complete list of writing signals AI in the prompt, and instruct the model to deliberately avoid them.
More info.: Wikipedia – Writing signals for AI
