You ask ChatGPT a simple question and get back three paragraphs, a bullet list nobody requested, and a closing line that reads like it was lifted from a airline customer service script. Sound familiar? That default voice, polite, padded, cautious, is baked in. But it can be changed, and you do not need any technical skills to do it.
The first instinct most people have is to type something like "be more casual" or "stop being so formal." It works for one reply, maybe two. Then the model quietly slides back to its defaults. The reason is simple: words like "casual" or "direct" are too vague. The model has no idea what casual means to you specifically, so it guesses. It usually guesses wrong.
What actually works is a hard constraint paired with a concrete role. Instead of asking it to be less wordy, try something like: "Answer in the voice of a sharp editor reviewing a rough draft. No preamble, no summary of what I just said. Maximum three sentences unless I ask for more." That sentence cap is doing the heavy lifting. Models respond to numeric limits far more reliably than to adjectives.
Show, don't describe
The most underused trick is giving the model an actual example of the tone you want, not a description of it. If you want a slightly sarcastic tech-reviewer voice, don't write "be sarcastic." Write this instead: "Example of the tone I want: 'Sure, the new update fixes the bug from three versions ago. Only took them a year.' Now answer my question in that tone." The model pattern-matches to a real sample far more consistently than it does to a loose descriptor.
The same logic applies to negative instructions. Telling ChatGPT "don't sound like a corporate robot" barely moves the needle, because there is nothing for it to replace that habit with. The fix is to give the negative and the substitute together: "Don't use phrases like 'I understand your concern', open directly with the answer." Run that against the negative-only instruction and the difference shows up immediately, every time.
Lock the rules in once, not every session
Single-turn instructions fade across a long conversation. If you have access to ChatGPT's custom instructions field or a project-level prompt, put your rules there. Something like: "In all responses, skip introductory fluff, no bullet lists unless I ask, banned phrases include 'certainly', 'of course', 'great question', and 'I hope this helps'." Setting it once at the system level means you stop fighting the same drift every few messages.
When the tone does drift anyway, and it will, the fastest reset is dropping in a fresh example sentence rather than re-explaining the rules. One concrete sample recalibrates the model faster than a paragraph of instructions.
