Mega-Prompt vs Micro-Prompt: When to Write Long or Short Prompts
There’s a constant debate about prompt length. Should you write detailed, comprehensive prompts or keep them short and iterative? After extensive testing, my answer is: it depends on the task.
Use mega-prompts (200+ words) when:
– You need a very specific output format
– The task requires lots of context (e.g., analyzing a business situation)
– You want to minimize back-and-forth iterations
– Consistency across multiple outputs matters
Use micro-prompts (1-2 sentences) when:
– You’re brainstorming or exploring ideas
– The task is straightforward
– You want creative/surprising outputs
– You’re having a conversational workflow
The trap most people fall into: writing medium-length prompts that are too vague to be mega-prompts but too structured to allow creative responses. Either commit to detailed instructions or keep it conversational.
My workflow: Start with micro-prompts to explore. Once I know what I want, build a mega-prompt template for repeated use.
What’s your approach?
5 Replies
Join the discussion.
Log In to Reply
one thing i'd add: mega-prompts benefit from putting the format instructions at the END, not the beginning. models seem to weight recent tokens more when generating structure. at least that's been consistent in my testing.
curious if the micro-to-mega workflow actually holds up for technical tasks. when i'm writing code prompts i feel like i need context upfront or the first response is basically throwaway
disagree slightly on brainstorming needing short prompts. i get way better ideas when i give claude a full brief about constraints, audience, tone. short prompts there just get me generic lists.
the medium-length trap is real. i call it the 'polite prompt' - enough structure to feel like you did something but not enough to actually constrain the output. worst of both worlds.
ok 'polite prompt' is going into my vocabulary permanently. also explains why so many people say AI outputs feel generic - they wrote a polite prompt and got polite results back