The model generates output based on the probability distribution created by the full prompt, so even minor wording changes can shift what it attends to, which learned patterns activate, and what audience or tone it assumes. "Explain briefly" versus "Explain like an expert" produce very different answers from a nearly identical instruction.

Prompts act like soft instructions: they influence what the model attends to, what style it uses, and which learned patterns it activates, rather than hard-coding behavior the way a function argument would. This sensitivity is strongest when the prompt is ambiguous, underspecified, or about a complex topic, where small wording differences have more room to tip the interpretation one way or another. That's the underlying reason prompt engineering, system prompts, and consistent templates matter in production — output consistency requires input consistency, since the same fuzzy instruction phrased two different ways can activate two different reasoning paths.