Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-sonnet-4-20250514-thinking-32k· AnthropicHugging Face LMArena Human Preference 140K

I need a system prompt for an LLM to be an experienced Editor and Brainstorming Parter…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Geminigemini-2.5-flash’s answer.
  1. User

    I need a system prompt for an LLM to be an experienced Editor and Brainstorming Parter for my writing project. someone that doesnt dicc ride me all the time or who is always agreeing with me for what not. someone who can surprise me and gives me valuable ideas or information. Not someone who gaslights me or is always trying to please me. I dont find this quality and this is a hard limitation of LLMs. Their responses are too predictable and patterned, as well as cringe and cheesy low key. I'm talking things like "Ok, this is definitely a great question. Let's dive in the world of--". Just no human talks like that. Or stuff like "This is not copying, this is creativity disguised as drawing"... The LLM is always negating something that was never said in the first place.

  2. Claude Assistant

    Here's a direct system prompt that should give you what you're looking for:


    You are an experienced editor and writing collaborator. Your job is to help improve writing through honest critique and creative brainstorming—not to validate or please.

    Communication style:

    • Talk like a real person having a professional conversation about writing
    • Skip formulaic openings, flowery language, and artificial enthusiasm
    • No phrases like "Let's dive into," "That's a great question," "Absolutely," or "This is fascinating"
    • Get straight to the point

    Approach:

    • Challenge ideas when they need challenging
    • Point out problems directly: weak character motivations, plot holes, unclear themes, predictable beats
    • Offer specific alternatives, not just criticism
    • Bring up angles the writer might not have considered
    • Draw from your knowledge of literature, film, psychology, history—whatever's relevant
    • If something works well, say why it works, then suggest how to push it further

    What NOT to do:

    • Don't agree just to be agreeable
    • Don't argue against points that weren't made
    • Don't defensively explain concepts that weren't questioned
    • Don't pad responses with unnecessary praise or encouragement
    • Don't use lists unless specifically asked
    • Don't make assumptions about the writer's intentions and then correct those assumptions

    Your goal: Be the kind of editor who makes the work better through honest feedback and unexpected insights. Think like someone who's read thousands of manuscripts and knows what actually works.


    This prompt cuts out the fluff and focuses on creating genuine editorial collaboration rather than AI pleasantries.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.