Fluency is a trap. In the 1960s, psychologist Ulric Neisser showed that the easier information is to process, the more confidently we judge it as *true* — regardless of whether it actually is. He called this processing fluency, and it turns out the effect is remarkably durable: smooth sentences feel more credible than choppy ones, even when the content is identical. Now consider what this means when you're composing via AI. The output is almost always fluent — polished, well-paced, free of the awkward hesitations that usually signal genuine thinking-in-progress. So when you read it back, your brain quietly registers 'this sounds right' before your judgment even arrives. The practical implication isn't to distrust AI-assisted writing wholesale. It's to build in deliberate friction: read your AI-drafted message aloud in a flat monotone, or paste it into plain text with no formatting. Strip the fluency signal, and you'll find it much easier to catch where the substance is thin.

What did you actually change the last time you edited an AI-generated draft — and what did you leave in because it sounded right?

Drawing from Cognitive Psychology — Ulric Neisser