How do you write a prompt that changes the output for client work?

Tips & Best Practices
1

specific prompt changes that made a real difference in output quality for client work: not ‘be more specific’ as general advice, actual changes that moved the needle. i’m putting together a workshop and want the concrete ones.

2

adding a negative constraint list changed my output more than any positive instruction. ‘do not summarize at the end, do not use passive voice in the first sentence, do not use noting-as-filler phrases’ cuts the most common AI tells before they appear.

3

specifying the audience in concrete terms. not ‘professional audience’ but ‘a marketing director at a mid-size B2B software company who reads at speed and skims first.’ the model adjusts assumed knowledge, vocabulary, and detail level. same topic, noticeably different output.

4

pasting two or three sentences from your own writing that you’re happy with before asking for the draft. not describing your style, showing it. ‘write in a register consistent with these sentences’ produces more useful output than ‘write conversationally.’

5

for product content: ‘write this as a recommendation to a friend who asked why you bought it’ produces different material from ‘write a product description.’ the first frame creates voice. the second frame creates a spec.

6

synthesizing from the replies and my own testing for the workshop:

changes that move the needle:

  1. negative constraints, list 5-8 things you don’t want. these are more specific than positive instructions and cut AI defaults fast.

  2. concrete audience, not ‘professional’ but ‘a [role] at a [company type] who [specific situation].’

  3. style example sentences, paste 2-3 sentences you wrote that match the target register. more reliable than describing style abstractly.

  4. frame the task differently: ‘recommend’ instead of ‘describe’, ‘explain to a practitioner who’s made this mistake’ instead of ‘provide an overview.’

  5. after a Walter Writes pass on flagged sections, feed the specific detector feedback back into the next prompt. ‘the transitions between sections are still scoring high, rewrite those specifically.’ targeted iteration beats broad re-generation.