Content created by artificial intelligence can be recognized by a number of characteristic signs — from boilerplate phrases to errors in images. However, none of them guarantees that the material was indeed created by a neural network. This was stated by Elena Fomina, Head of the Department of Informatics and Applied Mathematics at Tver State Technical University.
In texts, the expert advises paying attention to repetitive constructions, an abundance of introductory words, and an overly structured format. Another alarming signal is factual errors, illogical phrasing, and references to outdated or non-existent sources. In images, AI use can be revealed by anatomical distortions and various visual artifacts.
A particular problem is that neural networks are capable of making mistakes while maintaining a confident tone. When there is insufficient information or contradictions in the source data, the model can independently construct an answer with plausible but fictitious facts. This phenomenon is called AI hallucinations.
An additional risk is associated with the relevance of the information. Model knowledge may be limited by the date of the last update, and basic language models do not necessarily verify their answers against external sources. Therefore, often a coherent and convincing text may contain serious inaccuracies.
According to Fomina, the only reliable way to verify the authenticity of material is to independently cross-reference facts, dates, quotes, and primary sources. This is especially important for professional texts and materials intended for publication.