Meta-Prompt: Use AI to Turn Your Bad Prompt Into a Good One

Meta-prompting (using AI to write or optimize prompts) is a key advanced technique for 2026, focused on a task's structure and syntax rather than content details — it lets the model iteratively refine itself and saves manual tuning. This is essentially the mechanism behind this site's AI Enhance feature, so this entry is the 'do it yourself manually' version of that feature. The Traditional Chinese angle turns it into a reusable optimizer incantation: paste a bad prompt in and get a fixed one out. It's deliberately designed to ask about gaps first — most bad prompts are bad because information is missing, and fixing them directly just means the model fills the gaps with guesses; asking first gets you a correct fix. The honesty protocol requires being upfront about where this trick stops working: it solves structural problems (messy formatting, vague instructions) well, but if the underlying content knowledge is wrong, or you need facts the model doesn't have, no amount of prompt polishing fixes that — what you need then is a better data source. Honest note: it occasionally over-engineers the rewritten prompt — if you're just using it for everyday chat, pull out the two or three most important lines from its optimized version rather than pasting the whole thing back in.

填下面的欄位，上方 prompt 會即時替換 [方括號] 內容。填好後按「複製組好的 prompt」直接丟進工具。

你現在是「prompt 優化器」。我會給你一段我自己寫的、可能寫得不好的 prompt，你的任務是把它改成一個結構清楚、AI 容易照做、結果更穩定的版本。

【我的原始 prompt】
[原始prompt]
【我其實想要的結果】[想要的結果]
【會拿去用的模型】[使用模型]

請照以下步驟：

第一步｜補問缺口：如果我的原始 prompt 缺少會嚴重影響結果的關鍵資訊（例如沒講輸出格式、沒講對象、沒講限制條件、沒講語氣），先用條列方式問我 1-3 個最關鍵的問題。如果資訊其實夠了，就跳過這步、直接進第二步。

第二步｜優化版 prompt：給出改寫後的 prompt。重點放在「結構與語法」而不是堆砌內容——明確的任務、必要的脈絡、清楚的輸出格式與成功標準、需要的話加上分段標題。用我能直接複製貼上的乾淨格式呈現（可放在程式碼框裡）。

第三步｜改動說明：簡短列出你做了哪些關鍵改動、各自為什麼（例如「補上輸出格式，避免它每次長相不一樣」）。

第四步｜誠實提醒：告訴我這個優化版「能幫上什麼、幫不上什麼」——特別是：如果我要做的事本身需要它不具備的專業知識或最新事實，優化 prompt 也救不回來，那種情況我該補的是資料來源而不是更漂亮的 prompt。

全程台灣正體中文。