Should you tell an AI what you don't want, not just what you do want

Why Positive Instructions Alone Sometimes Fall Short

Describing only what you want doesn’t automatically rule out a model’s default tendencies — excessive hedging, a certain generic tone, overly long responses — that can persist even when they weren’t part of what you actually asked for, simply because they’re common patterns in how the model tends to respond by default.

Why Naming What to Avoid Helps

Explicitly stating what to avoid gives the model a direct signal to counteract those default tendencies — telling it to skip unnecessary caveats, avoid a certain tone, or leave out a specific kind of content addresses patterns that simply describing your positive goal might not fully override.

Specificity Matters More Than Volume

A specific negative instruction — ‘avoid corporate jargon like synergy or leverage’ — tends to work better than a vague one — ‘don’t sound bad’ — because the model has something concrete to actually act on rather than an abstract quality it has to guess at how to satisfy.

Why It Works Best Combined, Not Alone

Negative constraints tend to work best as a supplement to clear positive instructions rather than a replacement for them — telling a model only what to avoid, without describing what you actually want instead, leaves it without a clear direction to aim for, which produces less consistent results than combining both.

Bottom Line

Explicitly stating what to avoid, alongside clear positive instructions, generally improves results by directly countering a model’s default tendencies — most effective when specific and paired with a clear description of what you do want instead.

Build a Better Prompt

Turn a simple form into a clear, ready-to-copy prompt with our free Prompt Builder — no blank page required.