Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)

The "dependable" data point is a fascinating artifact of how large language models are trained. When a model is reinforced to sound helpful, credible, and safe, it gravitates toward a narrow band of adjectives that signal reliability. Humans write "dependable" when it's earned; an AI writes it because it's statistically adjacent to trustworthiness. This is the inverse of the old "uncanny valley" — the text isn't too strange, it's too familiar, too eager to reassure. Opus 5.5 doesn't just want to answer your question; it wants to convince you the answer is solid, and that compulsion leaves a measurable trace. This matters beyond mere trivia or parlor tricks for detecting bots. As AI writing floods newsrooms, marketing departments, and code documentation, these tells become a form of unintentional watermark. The interesting part is that they are not cryptographic or deliberate — they are emergent properties of optimization. That means they will likely shift as models evolve, forcing detection tools into a perpetual game of whack-a-mole. The "dependable" spike of today could be the "leverage" spike of tomorrow. There's also a deeper editorial concern here. If AI output consistently leans on a compressed vocabulary of certainty, it flattens the texture of public discourse. Nuance, hesitation, and idiosyncrasy are human features that models are actively trained away from. The result is prose that reads like a confident press release even when discussing uncertain, evolving topics. For a publication like Tech Pulse, that's a red flag not just for authenticity, but for epistemic quality — a world of uniform, reassuring AI prose is a world where doubt is algorithmically scrubbed out. Source: https://techcrunch.com/2026/10/01/opus-5-5-loves-to-tell-you-this-matters-and-other-ai-writing-tells/