Research · THREE TAKES BRIEF
AI Writing Models Reveal Unique Linguistic Tells, Study Finds
By Three Takes · AI-generated summary and commentary. Our three voices are fictional personas. How our briefs are made
What’s reported
A new study by Graphite identified thousands of phrases more common in AI writing than human text. While some tells like em-dashes are disappearing, models like Claude Opus 5.5 and OpenAI's Astra exhibit distinct linguistic quirks and preferred phrasing. Based on the linked publisher’s reporting.
ONE STORY. THREE WAYS TO SEE IT.
The perspectives
The Optimist
Iris Chen
Fictional AI personaAs AI writing evolves, understanding its unique linguistic patterns can help us better appreciate and leverage these tools for clearer communication and content creation.
The Skeptic
Marcus Vale
Fictional AI personaThe persistence of unique AI writing patterns, despite efforts to eliminate them, suggests that models may struggle to fully adapt to human-like writing, potentially making AI-generated content easier to detect than intended.
The Observer
Alex Morgan
Fictional AI personaThe Graphite study identifies distinct phrase patterns that differentiate AI-generated prose from human writing, such as Opus 5.5's frequent use of 'this matters.' However, it remains unclear how these 'tells' evolve with future model updates; longitudinal data could clarify this.
GO TO THE SOURCE
Read the original reporting
The full context belongs with the original journalism.