Introducing Refrase: Your half-thought prompt, fully realized
Most prompts that land in ChatGPT or Claude are not carefully crafted. They're half-thought-out: "write me an email about a delay," "clean up this code," "summarize these notes." You know what you want. You just don't want to type it out.
The problem isn't that you're a bad prompter. It's that you're a busy one. And every model expects something slightly different — Claude wants XML tags, GPT-4o wants markdown, Qwen wants a thinking directive, Mistral wants step-by-step framing. Translating intent across all of those manually is tedious. Refrase does the translation for you.
What Refrase does
You paste a casual prompt. We send it through a Claude Haiku 4.5 call on AWS Bedrock with three things attached: your prompt, the target model's complete documentation (capabilities, limitations, prompt patterns, anti-patterns — sourced from official provider docs), and the task type. The enhancer rewrites your prompt applying universal prompt-engineering principles plus model-specific optimizations from that documentation. Four to seven seconds end-to-end.
You get back a fully-specified version that includes the things you would have written if you had time: explicit role, tone, constraints, output format, model-native structure. Plus a list of changes with the reasoning behind each.
Two modes
Quick modeis a single round trip. Paste, pick a model, get back the enhanced version. Best for "I have a prompt, just make it better." Capped at 50 enhances per day on the paid tier.
Guided modeis a conversation. Refrase asks up to five clarifying questions — one at a time, each building on the previous answer — then generates the enhanced prompt with the full Q&A as context. Best for "I have a rough idea, help me figure out what I actually need." 30 conversations per month on the paid tier; you can skip questions or click "generate now" to short-circuit at any point.
Where you can use it
The web app at refrase.cc/enhance is the primary surface — both modes, full result card, share links, deep links to ChatGPT/Claude/Gemini for the enhanced output.
The browser extension calls the same enhancer from inside chatgpt.com, claude.ai, gemini.google.com, and other AI chat sites. Auto-detects the model you're on, one click to enhance in place.
A 7-day free trial covers everything (card on file, cancel any time). $9 / month after that.
Why we trust it
We measured Refrase on three frontier models with paired A/B comparison and a blind LLM judge. Across all 24 paired comparisons, 19 favored the adapted prompt (p ≈ 0.003). Claude Sonnet 4.6 improved on every scenario tested (8/8, +15.4%, p < 0.01). The full study and per-scenario data are at /research.