From Wikiprompt, the free prompt encyclopedia
Wikiprompt is a collaborative encyclopedia of AI prompts, featuring curated templates for various AI tools including ChatGPT, Midjourney, DALL-E, Claude, and more. It contains 3213 prompts by 225 community members.
All, Creative, Marketing, Business, Productivity, Coding, Education, Personal, Research, Other
All, GPT Image 2 (800), ChatGPT (298), Model-agnostic (247), Seedance 2.0 (242), Nano Banana Pro (226), Gemini Nano Banana (210), Gemini Nano Banana Pro (182), Grok Imagine (83), Claude (61), Gemini (44), Midjourney v7 (29), Nano Banana 2 (27), Veo 3 (25), Gemini Nano Banana 2 (24), GPT Image 1.5 (24), Midjourney (21), NotebookLM (20), Midjourney v8.1 (14), ChatGPT / Gemini (12), Kling AI 3.0 Turbo (12)
All, cinematic (198), photorealistic (115), editorial (106), portrait (96), commercial (39), fashion (38), minimalist (31), studio (31), luxury (30), product-photography (29), lifestyle (28), golden-hour (26), surreal (25), miniature (23), infographic (21), 3d-render (20), poster (20), candid (19), monochrome (19), macro (18)
All, Creative generation (2567), Template (321), One-shot (226), System prompt (50), Workflow (39), Persona (6), Agent (3)
115 prompts
A quick follow-up prompt that asks ChatGPT to find overlooked context and produce a stronger version of its last answer.
A follow-up prompt that gets ChatGPT to ask only the clarifying questions that would actually improve its answer.
A follow-up prompt that trims a wordy ChatGPT answer down to its essential meaning without losing content.
A follow-up prompt that rewrites a vague request into a concrete, actionable one with clear constraints and format.
A follow-up prompt that stops ChatGPT from validating your idea and forces it to argue against it honestly.
A follow-up prompt that converts an answer into a prioritized action plan split into now, later, and avoid.
A follow-up prompt that has ChatGPT rewrite your messy request into a better prompt before answering it.
A command code that makes the model cut the fluff and explain why something actually matters whenever a message ends in /SOWHAT.
A command code that makes the model argue the strongest possible case for the opposite of your position whenever a message ends in /STEELMAN.
A command code that makes ChatGPT or Claude point out the assumptions and gaps you have not stated out loud whenever a message ends in /HOLES.
A command code that makes the model drop the hedging and give the blunt version of its answer whenever a message ends in /PLAINLY.
A command code that makes the model state its honest confidence level, high, medium or low, and what would change it, whenever a message ends in /ODDS.
A command code you paste once at the start of a chat so that any message ending in /ATTACK gets argued against as hard as the model honestly can, instead of getting agreement.
A command code that makes the model predict the most likely way a plan goes wrong before you commit to it, whenever a message ends in /FAILHOW.
A task framing template for agentic tools like ChatGPT Work or Claude Cowork that pulls context from connected apps and only interrupts you when it truly needs your judgment.
Instruction for Claude's skill-creator to convert a working chat workflow into a reusable custom command that prompts for variables each run.
Example prompt for using a Model Context Protocol server so Claude reads real files or data directly instead of copy pasting between tabs.
A template for batching multiple related tasks, summarize, extract action items, draft email, into a single Claude prompt to save tokens and avoid reloading the context window.
A reusable set of global / custom instructions to paste once for a Claude agent (e.g. Cowork): it makes every session read your context files first, ask clarifying questions, show a plan before acting…
Has ChatGPT strip a plan down using the 80/20 principle, flagging what to remove, reduce, and keep, useful when a plan feels bloated, expensive, slow or too complicated.