Detect AI in the Text
The advancement of artificial intelligence (AI) has had a significant impact on various processes concerning text creation. It entered diverse fields being a tool for crafting ad copies in marketing, writing news reports in journalism, or assisting in writing research papers. Students, too, increasingly rely on AI for completing writing tasks — often without proper review. We, teachers, need to think how to organize the whole education process ensuring its academic integrity. Bathed on my observations, I have collected a list of detectable features that AI-generated texts have. Hope they will be useful for both writers and teachers.
Photo by Andres Siimon on Unsplash
By Sofia Stalskaya
Senior Lecturer at Department of Foreign Languages
HSE University in St. Petersburg
Features of AI-Generated Texts
1. Repetitiveness & Lack of Depth or Nuance
AI models often repeat phrases, ideas, or structures because they rely on patterns in their training data. This can lead to redundancy, where the same point is expressed multiple times in slightly different wording. Let us analyse the following example:
Incorporating a variety of fruits into your diet can significantly enhance your nutritional intake and contribute to a healthier lifestyle. Aim for a colorful assortment to maximize the range of nutrients and health benefits."
As you see, the only idea - consuming a variety of fruits can enhance your nutritional intake and improve your lifestyle - is already developed enough. The second sentence does not provide any additional information; therefore, it can be deleted.
2. Generic Tone and Absence of Original Thought
AI tends to use a neutral, formal tone because it is trained on a wide variety of texts that span many topics and styles. Without explicit instructions about the tone (e.g., casual, humorous, persuasive), the output may feel impersonal. For instance, AI tends to use “individuals” instead of “people,” which does not always suit the context.
3. Inconsistent Details
AI may occasionally mix up facts, dates, names, or other details because it generates content based on patterns rather than real-world knowledge. That especially happens in an attempt to gather some literature sources, just like in the example below:
Berenike, S., & Dancygier, B. (Eds.). (2011). Cognitive and discourse approaches to discourse analysis. Bloomsbury.”
Fact checking shows that the book mentioned in the output was never published. Moreover, the first author does not exist and might only sound similarly to the other one. Such inconsistencies occur because the model cannot verify information against real-world sources. Just like our students sometimes :)
4. Overly Perfect Grammar and Syntax
AI models rarely make grammatical errors, which can sometimes make the text seem unnatural. It is quite easy to catch strict grammatical rules, followed by the models, lacking any conversational quality.
5. Absence of Original Thought
| Unlock Your Future at HSE! Discover a World of Opportunities! At the Higher School of Economics, we don't just teach; we shape leaders. Join a vibrant community of thinkers, innovators, and changemakers from around the globe. | Join the Legacy of Excellence! Step into the iconic halls of Moscow State University, where tradition meets innovation. With a rich history dating back to 1755, MSU is Russia’s premier institution, preparing you for a world of possibilities. |
As you can see, both texts present the idea of opportunities and innovations without any significant nuances. Do you feel odd reading one and the same idea relating to such different places? I certainly do.
6. Predictable Structure
AI-generated text often follows predictable structures, such as rigid introductions, body paragraphs, and conclusions. Introductions typically begin with some broad statements before narrowing down to the main topic. As for body paragraphs, they strictly provide some point, then evidence, and only then the explanation, while conclusions summarize key points without introducing new insights. For some purposes it might be the best fit, but what if all the texts we are going to produce are so schematic and identical?
AI frequently uses bullet points, numbered lists, or headings — even when they are not necessary. These formats are pattern-based: such structuring can make the text a bit mechanical. It especially feels in texts where creativity and out-of-the-box approach are valued.
7. Specific Word Choice
Word-choice is also quite specific in AI-generated text. AI often reuses certain words or phrases within a text making sure they are not used separately but in a phrase pattern. At the same time AI uses quite advanced vocabulary which is usually very formal (see Table 1).
Although about 70% of academic vocabulary is formal by nature, AI is filling the texts with stiff phrases we would never be able to chain together naturally. The problem is AI lacks the “why” behind all the polished sentences and with overuse of such tools in writing we can lose the human touch that makes arguments really land. Overuse of too formulaic vocabulary might not fit the context best, so that any writing could become unnatural.
| adj. + noun | verb + smth | verb | adj. |
| notable works | aim to explore | align | comprehensive |
| equitable access | foster creativity | showcase | surpassing |
| significant boosts | play a significant role | encompass | (in)significant |
| powerful capabilities | tap into the power | empower | compelling |
| diverse perspectives | continue to evolve | enhance | robust |
| embrace your potential | impact | scalable | |
| evolve through | cater | ||
| encourage innovation | incorporate | ||
| delve into |
While AI creates grammatically correct and logically coherent text, its output still lacks individuality. People vary word choice based on context, audience, and purpose, while AI’s word choice does not provide emotional and creative expression, focusing mainly on functionality and clarity. Still, we are all personalities with our own individual voice, even in formal writing.
No doubt, AI-generated content shall be thoroughly evaluated by users and reviewed for repetitiveness or overly formal language and any inconsistencies. Leaning on AI tools should not cut off our voice but help in human expression and communication.
Tools Used:
-
Merlin.Ai. (2025). Artificial Intelligence Tool for Writing and Editing Texts. Retrieved February 8, 2025, from https://www.merlin.ai
-
Qwen.Chat. (2025). Artificial Intelligence Tool for Learning and Communication. Retrieved February 8, 2025, from https://chat.qwenlm.ai/