More and more of the content we consume online is generated by artificial intelligence, and this applies not only to images and videos, but also to text documents. Tools like ChatGPT and Gemini are improving at a rapid pace, and in some cases understand if a text is generated by AI can be really complicated. However, by paying attention to some “clues” and using specific software (such as Detector IA or ZeroGPT) it is possible at least try to understand whether a content was conceived and written by a “real” user or whether it is the result of a cold algorithm.
The main signs and errors of an AI-generated text
The first thing that in our opinion should be done is to look for any AI errors in generating texts. Although AI-based algorithms have become good at generating convincing and seemingly exhaustive answers, they are still far from perfect and, in our opinion, are not yet fully capable of replacing human intelligence in creating high-quality content. Especially in texts of a certain length, it is possible to identify flaws and inaccuracies that can constitute valuable clues for those who want to understand the genesis of a text.
Basically, we could classify AI errors in writing texts into four broad categories:
- High repetitiveness and low consistency: AI tends to repeat phrases and concepts excessively and exaggeratedly, displaying a coherence that can be unnatural. This happens because generative AI algorithms try to maintain a coherent structure without having a clear understanding of the meaning of the content they are writing down. In contrast, a human writer introduces variations in the text to keep the reader’s attention, making the reading experience engaging and interesting.
- Contextual inaccuracies: AI-generated sentences, while usually grammatically flawless, can sometimes make little sense. This is because algorithms (as we have already said in the previous point) lack a true semantic and real-world understanding, which generates the so-called “AI hallucinations”, which clearly does not usually happen in text written by a human who has full knowledge of the real world and can thus avoid making logical or, indeed, contextual errors.
- Writing style absent: AI-generated texts often lack the warmth and nuance that distinguish human language. Even the best-performing chatbots often lack personality, and their content can sometimes come across as uniform and boring. This is because AI tends to follow predefined patterns without the creativity and expressiveness that characterize human writing, which, on the contrary, tends to have stylistic variations and vivid descriptions that increase the readability of the content.
- Lexical uniformity: AI can show limited lexical variety, using the same terms or phrases repeatedly, which is typically not the case in texts written by human writers, who are often stuffed with synonyms and related terms in an attempt to avoid monotony.
Software to recognize texts generated by AI
It must be said, however, that it is not always easy to understand whether a text is generated by AI or not. To try to further investigate the matter, it is possible to resort to some software designed to recognize texts generated by AI. Some examples are Detector AI by NoPlagio and ZeroGPT. Their operation is quite simple: you paste the “suspicious” text into the box of the chosen AI detector, start scanning the text by clicking on the appropriate button and wait for the response.
We would like to point out, however, that similar software is not infallible at all: from the tests we conducted for writing this article, we tried to have them examine content that we created without the use of artificial intelligence and the tools erroneously classified them as “generated by AI”. In other cases, however, they were not able to clearly identify the artificial origin of content that we created specifically with ChatGPT to carry out the test. The advice we give you, therefore, is to use similar Web applications only after having carefully examined a text looking for the errors we talked about in the previous chapter.
