Telling a model what to avoid is often as useful as telling it what to do, and it’s a simple technique most people never think to use deliberately.
Negative prompting means explicitly stating what you don’t want alongside what you do want: “don’t use exclamation points,” “avoid corporate jargon,” “skip the introduction and get straight to the point.” A direct complement to positive instructions, not a replacement.
Models have default tendencies, certain phrasings or structures that show up repeatedly unless explicitly steered away from. If you’ve noticed a recurring pattern you don’t like, stating it directly as something to avoid is often more effective than describing what you want in more detail.
Avoiding recognizable AI writing patterns, certain transitional phrases that read as obviously generated. Steering away from a tone the model defaults to. Excluding specific content without specifying everything you do want in exhaustive detail.
If you keep getting the same unwanted pattern regardless of how you phrase the positive instruction, try stating it as something to explicitly avoid instead. See our collection of prompt examples for more techniques.
See Anthropic’s own documentation for more technical depth.
