WitPrompt underlines weak spots in your prompt as you type on claude.ai, chatgpt.com, gemini.google.com and grok.com, checks them against each provider's official prompt-engineering guidance, and offers one-click fixes and AI rewrites.
Install it once. It lives inside the chat sites you already use. Vague asks, missing context, shouted emphasis and other issues get underlined in place, each one backed by Anthropic, OpenAI and Google's own docs, including model-specific advice.
Write something about1 our Q3 launch for the team. Make it engaging2 and IMPORTANT: do not make it long!!3
Unspecified gaps get filled with the model's guesses. Name the artifact, the audience and the length you want.
Engaging by what measure? State the criteria you will judge the result by, and the model can aim at them.
All-caps and exclamation points read as noise, not priority. Claude models in particular respond better to a plain constraint.
Hover an underline for the suggestion card. Accept edits just that phrase live in the prompt box, undoable with ⌘Z, like any edit you made yourself. Each card names the rule and links the doc it came from, so the lesson sticks.
One click sends your prompt to the WitPrompt service and returns a rewrite following the provider's guidance, shown as a diff before it touches anything. You see what changed and why, then decide.
The linter is free and runs in your browser. You pay for the rewriter, and only if you use it a lot.
Not for the free tier: the extension works anonymously out of the box. For Pro you sign in with Google (one click, no password to manage), buy once, and every browser you connect to the account is upgraded automatically. You never handle API keys.
Local checks run entirely in your browser; nothing is sent while you type. When you click Improve prompt, the prompt is sent to the WitPrompt service, passed to the AI model that generates the rewrite, and returned. WitPrompt does not store your prompt text; usage counters record only an anonymous token and date. Details in the privacy policy.
Sign in on the account page and upgrade. Then, in each browser's extension options, click Connect account. It opens a one-click approval page, and that browser inherits your plan instantly. Up to 3 browsers per account, manageable from the dashboard.
From each provider's official prompt-engineering documentation (Anthropic, OpenAI, Google), distilled into checkable rules with links back to the source docs, including model-specific notes, like how Claude models respond to shouted emphasis.
From your account page, open the billing portal and cancel. Pro stays active until the end of the period, then all your browsers drop back to the free tier automatically. Payments and refunds are handled by Lemon Squeezy.