13,000 Writing Patterns Identified in AI Texts

Research from marketing company Graphite reveals that AI models frequently rely on certain words and sentence structures in their writing. In total, nearly 13,000 recurring writing patterns were identified that occur at least twice as frequently in AI texts compared to human articles. This suggests that familiar cues in AI texts seem to disappear, while other, less known habits take precedence.

For this analysis, the researchers compared the writing styles of various AI models, including the new Claude Opus 5.5. This was done by using 10,000 articles published before the introduction of ChatGPT as a reference. The researchers then had the AI models write new articles based on summaries of these articles. This allowed them to analyze texts on the same subject without the AI models literally taking over the original texts.

The results indicate that each AI model has specific preferences. The research showed that while there is more variation in writing patterns, there is no definitive checklist to identify all AI texts. For example, the phrasing "This is important because..." occurred remarkably often in AI texts without providing concrete explanations. This points to a trend where AI models prefer superficial statements over detailed explanations.

Moreover, there are significant differences in the way different models communicate. For instance, the Claude Opus 5.5 model uses certain words, such as "dependable", significantly more frequently than human writers. The findings suggest that writing patterns continue to evolve and that new models are developing further, making traditional markers like the use of long dashes in AI texts less reliable. This emphasizes the need for writers and editors to be aware of these patterns to enhance their text quality.

Read the full article from AI Insider.