画像生成
雑然とした文書を1スライド1アイデアの構成案に変換
Copy-paste this prompt to turn a messy doc into a clean, one-idea-per-slide outline
雑然とした文書から核となる主張と4〜7個の要点を抽出し、各スライドを断定的な見出しとそれを裏付ける2〜3個の箇条書きで構成したプレゼンテーション案を返します。元の構成をなぞらず、内容を明快なスライドデッキに再設計したいときに使います。
- テスト済みモデル:
- GPT-4o
- 有効性スコア:
- 90/100
- プロンプトアーカイブ
日本語バージョン
``` I'm pasting a document below. Turn it into a presentation outline. Do NOT follow the document's order or headings. Rebuild from first principles: 1. Read the whole thing and identify the single core argument. 2. Identify the 4-7 points that argument actually needs. Discard everything else, even if it's interesting. 3. Sequence those points so each one sets up the next. For each point, output: - Slide headline: the takeaway as a full assertive sentence - 2-3 bullets: only the evidence that supports THAT headline Cut anything that's context, throat-clearing, or repetition. If the doc says the same thing three ways, keep the clearest one. DOCUMENT: [paste] ``` Why it works: the failure mode with document-to-slides is "shadow-mirroring", where the model copies the source's layout instead of designing a new one. Explicitly telling it to ignore the doc's order and extract the core argument first forces a real restructure. The "discard interesting things" line matters more than it looks, because most bloated decks are bloated out of a fear of leaving stuff out. Example: I ran a 2,000-word strategy memo through this and it collapsed to six slides that actually built on each other, versus the fourteen flat slides I got asking directly. Curious if the "discard even if interesting" instruction survives for other people or if your model keeps sneaking the extra points back in.
値を入れて実行
このプロンプトには1か所の空欄があります。入力すると完成版を組み立てます。
残り1か所
完成したプロンプト
``` I'm pasting a document below. Turn it into a presentation outline. Do NOT follow the document's order or headings. Rebuild from first principles: 1. Read the whole thing and identify the single core argument. 2. Identify the 4-7 points that argument actually needs. Discard everything else, even if it's interesting. 3. Sequence those points so each one sets up the next. For each point, output: - Slide headline: the takeaway as a full assertive sentence - 2-3 bullets: only the evidence that supports THAT headline Cut anything that's context, throat-clearing, or repetition. If the doc says the same thing three ways, keep the clearest one. DOCUMENT: [paste] ``` Why it works: the failure mode with document-to-slides is "shadow-mirroring", where the model copies the source's layout instead of designing a new one. Explicitly telling it to ignore the doc's order and extract the core argument first forces a real restructure. The "discard interesting things" line matters more than it looks, because most bloated decks are bloated out of a fear of leaving stuff out. Example: I ran a 2,000-word strategy memo through this and it collapsed to six slides that actually built on each other, versus the fourteen flat slides I got asking directly. Curious if the "discard even if interesting" instruction survives for other people or if your model keeps sneaking the extra points back in.
- ChatGPTで開く (新しいタブで開きます)
- Claudeで開く (新しいタブで開きます)
- Geminiで開く (新しいタブで開きます)
Geminiはリンクでプロンプトを渡せないため、先にコピーしてください。
- Perplexityで開く (新しいタブで開きます)
新しいタブが開き、プロンプトが入力済みの状態になります。ログインが必要な場合があります。
このプロンプトの使い方
- 角かっこで囲まれた部分を自分の内容に置き換えてから送信してください。
- モデルによって結果は変わります。期待した出力でなければ、別のモデルを試すか、具体例を1つ追加してください。
- プロンプトは1つのメッセージにまとめて送ってください。分割すると指示が弱まります。