Researchers identify recurring 'tells' in Opus 5.5's AI-generated writing

Analysts examining text generated by the AI model Opus 5.5 say the system has a number of recurring stylistic habits, or "tells," that can help distinguish its writing from human-authored text.

According to the analysis, the model's most notable tell is its frequent use of the word "dependable," which appears 23 times more often in its output than in comparable samples of human writing. The finding points to a broader pattern in which large language models can develop detectable word-choice preferences and phrasing habits over repeated use.

Why writing tells matter

As AI-generated text becomes more widespread across professional and casual writing, identifying these linguistic fingerprints has become a growing area of interest for researchers, educators and platforms trying to distinguish machine-generated content from human writing.

Such tells are not necessarily errors or flaws in the model's output; rather, they reflect the particular patterns a model has learned to favor during training, which can become statistically distinctive even when individual sentences read naturally on their own.

The identification of a specific, measurable marker, such as the elevated frequency of a single word, illustrates how researchers are approaching the challenge of AI detection quantitatively, comparing word and phrase frequency across large samples of model output and human writing rather than relying solely on subjective impressions of style.

No further details on the full scope of the research or additional tells identified in the model's writing were included in the available reporting.