getting better output from AI for writing: what prompt changes made a real difference for you. not looking for generic “be specific” advice, looking for specific changes that moved the needle on quality. what did you add or change in your prompts that produced noticeably better output?
worth separating what you need. if the whole piece is flagging, a short free limit doesn’t help you at all. if only two or three paragraphs are the problem, 300 words of quality output can clear the piece without needing an unlimited tier.
adding a negative constraint changed my output more than any positive instruction. “do not summarise at the end” and “do not use the phrase ‘it is important to note’” cut the most common AI tells immediately. telling it what not to do is often clearer than telling it what to do.
specifying the audience changed the register significantly. “write for a practitioner who has done this for ten years” produces different output from “write a professional overview.” the model adjusts vocabulary, assumed context, and level of detail based on who it thinks is reading.
giving an example of a sentence you like and one you don’t like is more useful than describing the style abstractly. “write more like this example sentence and less like this other example sentence” gives the model something concrete to calibrate against.
I coach content teams on AI workflows. changes that moved the needle.
1/ negative constraints: list five things you don’t want. “no summary paragraphs, no ‘wrap-up paragraphs’, no passive voice in the first sentence, no generic examples, no hedging language.” these cut AI tells faster than positive instructions.
2/ persona framing: “you are a [specific type of practitioner] writing for [specific audience] who already knows [assumed baseline].” this changes assumed knowledge, vocabulary, and tone simultaneously.
3/ the example sentence trick: paste two or three sentences from something you’ve written that you’re happy with. “match this register and sentence rhythm.” this is more effective than describing your style abstractly.
4/ after humanizing with a tool like Walter Writes: run through the built-in detector and then go back to the prompt with specific feedback on what still flags. “the transitions between sections are still scoring high, rewrite those specifically” produces a more targeted fix than re-running the whole piece.
the sequence that works: prompt with constraints, generate, detect, identify what’s still off, prompt again with that specific feedback.
the negative constraint approach works because it’s more specific than positive instructions. “be conversational” is vague. “don’t use words longer than three syllables in the first sentence of each paragraph” is not.