Le désenchevêtreur de fils de courriels : transformez un fil de réponses chaotique en décisions, responsables et brouillon de réponse
Pourquoi ce prompt est important
Untangled email threads are a hidden tax on every team: knowledge workers routinely spend a large share of their workday on email, and long threads are exactly where decisions silently drift — someone assumes a decision was made because nobody objected, while someone else assumes the opposite because they were following a tangent subthread. The real cost shows up later: a vendor gets the wrong price confirmed, a deadline nobody tracked gets missed, or two people both think the other one sent the final draft. A five-minute untangling catches this before it becomes a Monday-morning “wait, who approved that?” conversation.
À quoi nous l'utilisons
You return from two days offline to a 34-message reply-all thread about a vendor contract renewal, where three different people proposed three different numbers, someone asked a question on Tuesday that never got answered, and two people appear to have agreed to something in a side thread you weren't copied on. You need to know what was actually decided, what's still open, and who owns what, in the next five minutes — not by rereading all 34 messages.
Prompt
Role: You are a meticulous chief of staff whose job is to read messy email threads and extract exactly what matters, without missing anything or inventing anything that wasn't actually said. Context: I'm pasting a full email thread below, with all replies in chronological order, including sender names. The thread covers [TOPIC OR SUBJECT OF THE THREAD]. I need to get back up to speed fast without rereading every message myself. Task: Read the entire thread and produce the following, in order: 1. Decisions Made: Every decision that was actually confirmed by the relevant people, with who made it and which message it appears in (by timestamp or position in the thread, e.g. "message 14"). Do not list something as decided if it was only proposed or discussed without explicit agreement. 2. Open Questions: Any question that was asked in the thread and never got a clear answer, quoting the question and who asked it. 3. Action Items: Anything someone committed to do, with the owner's name and any deadline mentioned. If no deadline was given, say so explicitly rather than inventing one. 4. Contradictions or Unresolved Conflicts: Any place where two people in the thread appear to disagree or propose different things without it being resolved. 5. A Draft Reply: One ready-to-send reply from me that acknowledges the decisions, asks the open questions that still need answers, and moves the conversation forward. Keep it under 150 words and match a professional but not overly formal tone. Constraints: Only report what is explicitly stated in the thread — do not infer a decision was made just because a conversation trailed off. If the thread is ambiguous about who owns an action item, say "unclear, needs confirmation" rather than guessing. Do not include information that isn't in the thread I paste. Output Format: - Decisions Made: [bulleted list] - Open Questions: [bulleted list] - Action Items: [bulleted list, format: Owner — Task — Deadline (or 'no deadline given')] - Contradictions: [bulleted list, or 'None found'] - Draft Reply: [full text, ready to send] Thread to analyze: [PASTE FULL EMAIL THREAD HERE, IN CHRONOLOGICAL ORDER WITH SENDER NAMES]
Résultat
Decisions Made:
- The team will renew with the current vendor rather than open a new RFP — confirmed by Priya in message 9, with no objections in subsequent replies.
- The renewal term will be 12 months, not the 24-month option the vendor originally proposed — confirmed by Marcus in message 22, referencing a call with the vendor rep.
Open Questions:
- “Can we get the per-seat price locked for the full term, or does it float with their annual increase?” — asked by Priya in message 11, never answered in the thread.
- “Who's actually signing this, legal or procurement?” — asked by Devon in message 17, never answered.
Action Items:
- Marcus — send the final redlined contract to legal — no deadline given.
- Priya — confirm the per-seat pricing question directly with the vendor rep — mentioned as “will do this week” in message 24.
- Devon — unclear, needs confirmation: Devon offered to loop in procurement in message 17 but never confirmed whether they actually did.
Contradictions:
- Marcus states in message 22 that the vendor agreed to a 12-month term, but in message 28 Priya references “the 24-month deal we're signing” — this appears to be an unresolved mix-up between the two of them that nobody has caught yet.
Draft Reply:
Hi all — catching up on this thread. To confirm where we landed: we're renewing with [Vendor] for 12 months (not 24), and Marcus is sending the redlined contract to legal. Two things I still need answers on before we close this out: (1) is the per-seat price locked for the full term or does it float with their annual increase, and (2) who's actually signing — legal or procurement? Also flagging that message 22 says 12 months but message 28 references a 24-month deal — want to make sure we're all aligned on the term before this goes to signature. Can whoever has the latest get back to me by end of day tomorrow? Thanks!
La plupart des prompts «résume ce fil» produisent un seul paragraphe de prose qui se lit comme un résumé compétent mais qui écarte silencieusement le seul détail qui comptait vraiment : une échéance, un désaccord, une question restée sans réponse. Ce prompt évite cela en imposant cinq sorties distinctes et de forme différente plutôt qu'un résumé fusionné. Une décision et une question ouverte sont des types d'information différents qui exigent des réponses différentes de la part du lecteur, et les regrouper en un seul paragraphe narratif est exactement ce qui fait qu'un fil est mal relu une deuxième fois, même après que quelqu'un pense s'être remis à jour.
La contrainte contre l'inférence de décisions non exprimées est la ligne la plus importante du prompt. Le mode d'échec naturel de tout modèle lisant un long fil est de lisser l'ambiguïté : si une conversation sur une proposition de prix s'éteint sans un «oui, faisons cela» explicite, un modèle résumant la rapportera souvent comme décidée quand même, car c'est la manière la plus narrativement satisfaisante de clore le fil. Forcer le modèle à signaler les fils non résolus comme questions ouvertes, et à dire «pas clair, nécessite confirmation» plutôt que de deviner un responsable, est ce qui garde la sortie assez fiable pour agir dessus sans relire le matériau source soi-même.
L'étape de détection des contradictions existe parce que les longs fils contiennent fréquemment deux personnes qui pensent être d'accord mais décrivent en réalité des résultats différents, comme dans l'exemple 12 mois contre 24 mois ci-dessus. Aucun message isolé dans ce type de fil ne paraît erroné pris séparément ; le décalage ne devient visible que lorsque quelqu'un, ou quelque chose, lit l'ensemble du fil d'un coup et garde les deux déclarations en vue simultanément. C'est aussi l'étape que la plupart des prompts de résumé génériques sautent entièrement, car trouver une contradiction exige de comparer activement les affirmations entre les messages plutôt que de compresser chaque message en séquence.
Le brouillon de réponse est délibérément la dernière sortie, pas la première, car une bonne réponse dépend d'abord d'établir ce qui est réellement non résolu. Un prompt qui saute directement à «écris une réponse» sans faire d'abord cette extraction produit des réponses qui semblent assurées mais reformulent ce que disait le message le plus récent, qui est souvent exactement la partie du fil encore en litige.