Construtor de Script para Conversas Difíceis: Prompt do Claude para Conflitos no Trabalho
Why this prompt matters
<p>Unaddressed workplace conflict costs more than you think. Research from CPP Inc. found that U.S. employees spend an average of 2.8 hours per week dealing with conflict — that's $359 billion in lost working hours annually. But the bigger cost is not the time: it's the relationships damaged when poorly handled conversations turn someone defensive, resentful, or disengaged.</p><p>Most people avoid these conversations until the situation has festered into something much harder to fix. And when they finally do have the conversation, they either over-prepare (over-rehearsed lines that feel robotic) or under-prepare (they improvise, get emotional, and say something they regret). This prompt gives you a third path: a structured, human-sounding script that keeps you on track even when the other person pushes back hard.</p><p>The script structure is built on conflict resolution research — specifically the model from <em>Crucial Conversations</em> (Grenny et al.) combined with Nonviolent Communication principles. The 'listening pause' section alone prevents the most common failure mode: people who feel unheard stop engaging with solutions.</p>
What we use it for
You have a conversation that needs to happen — but you keep postponing it. Maybe a team member keeps missing deadlines, a peer is undermining your work, or a manager gave you feedback you strongly disagree with. You know what you want to say but can't figure out how to say it without it going sideways.
Prompt
Act as a seasoned executive coach and organizational psychologist with 20+ years of experience guiding leaders through high-stakes workplace conversations. CONTEXT: I need to have a difficult conversation with [PERSON'S ROLE, e.g., "a direct report", "my manager", "a peer/colleague"] named [NAME OR LEAVE BLANK]. The issue is: [DESCRIBE THE SITUATION IN 2-3 SENTENCES, e.g., "They have been consistently missing project deadlines, and the last missed deadline caused us to lose a client. I've mentioned it informally before but nothing has changed."] The relationship dynamic: [DESCRIBE, e.g., "I manage this person directly", "We are peers but I depend on their output", "This is my manager and I need to escalate concerns upward"] My goal for this conversation: [e.g., "Reach a clear agreement on changed behavior with a measurable timeline", "Understand their perspective and decide together whether this role is still a fit", "Set a formal performance improvement process in motion"] Their likely emotional state or resistance: [e.g., "Defensive — they tend to deflect blame", "Anxious — they're going through personal difficulties", "Dismissive — they don't think this is a serious issue"] TASK: Design a complete, ready-to-use conversation script for this situation. The script must feel human, not robotic — I should be able to read these lines naturally. CONSTRAINTS: - Do not include corporate jargon or HR-speak (no "going forward", "circle back", "touch base") - The opening must not sound like an ambush — ease in with genuine intent - Avoid accusatory language; use observable facts, not character judgments - Include a moment where I actively listen and acknowledge their perspective before pushing forward - Do not skip to solutions before the other person has felt heard - If an agreement or commitment is reached, show me how to document it verbally during the conversation OUTPUT FORMAT: Return the script in these exact sections: **1. SETTING THE STAGE (Pre-conversation)** [2-3 bullet points on how to frame the meeting invite and set the right environment] **2. OPENING (First 60 seconds)** [Exact words to say. Warm, direct, and non-threatening.] **3. CORE MESSAGE** [Deliver the key issue with specific observable facts — no vague generalities] **4. THE LISTENING PAUSE** [Script for pausing, asking an open question, and genuinely processing what they say] **5. ANTICIPATED PUSHBACK — 3 SCENARIOS** For each likely response they might give, show: - What they might say - How to respond without getting defensive or capitulating **6. MOVING TO RESOLUTION** [How to propose next steps, get verbal commitment, and close the conversation constructively] **7. FOLLOW-UP** [One follow-up message to send within 24 hours that reinforces the agreement]
Result
Example scenario: Direct report (Jordan) has missed three consecutive deadlines, the last of which caused a client to reduce their contract renewal.
1. SETTING THE STAGE
- Book 45 minutes (not 30 — rushed conversations get cut short at the worst moment). Title the invite: Check-in — project delivery and workload. Do not call it a performance review in the invite or Jordan will show up in self-defense mode.
- Find a private room or use video with cameras on. No open-plan seating. The conversation needs psychological safety on both sides.
- Have specific dates and project names written down in front of you — not to wave them accusatorially, but so you speak with precision, not approximation.
2. OPENING
"Jordan, I appreciate you making time. I want to talk about the last few project cycles because something is not working, and I want to understand what's going on from your side before we talk about what happens next. I'm not coming into this with a verdict — I'm coming in to figure this out together."
3. CORE MESSAGE
"In Q1, the data migration for the Henley account was due February 14th and delivered February 28th. The API integration for Storrow was due March 3rd and came in March 19th. And last week's deliverable for Cairns Group — that was due the 22nd, we got it on the 29th, and because of that delay they've reduced their renewal from 12 months to 6 months. That's a $47K impact to the team's number. I need us to address this directly."
4. THE LISTENING PAUSE
"Before I say anything else — I want to hear from you. What's your read on what's been happening?"
Then stop. Wait through the silence. Do not fill it. If they seem stuck, try: "Is there something going on that I'm not aware of that's making this harder than it should be?"
5. ANTICIPATED PUSHBACK — 3 SCENARIOS
Scenario A — They blame the process:
"The timelines are unrealistic. No one could hit those."
Response: "That's worth exploring. If the timelines are genuinely off, I need to know that, and we'll fix it. But the timelines were agreed to at kickoff with your sign-off — so the question I have is: when did you first know you couldn't hit the February date, and why didn't that surface earlier so we could adjust?"
Scenario B — They go silent or emotional:
Response: "Take a breath — I'm not going anywhere. This conversation matters to me because you matter to this team. I just need us to get to a real answer together."
Scenario C — They say they didn't know it was a problem:
"I didn't realize it was such a big deal."
Response: "I hear that, and that's on me too — I should have been clearer earlier that late delivery has real consequences downstream. I'm being clear now."
6. MOVING TO RESOLUTION
"Here's what I want to leave this conversation with: a clear plan for the next two deliverables — specifically, I want you to flag me within 24 hours if you hit any blocker that might push a date. Not after the fact — during. Can we agree to that?"
If yes: "Good. I'll send you a short follow-up note today with what we've agreed to, just so we both have a record. I want this to work — and I think it can."
7. FOLLOW-UP MESSAGE (within 24 hours)
Jordan — thanks for the honest conversation today. To summarize what we agreed: for the upcoming Hartley integration (due June 12) and the Mercer report (due June 18), you'll flag me by EOD if anything looks like it could push either date. I'll check in mid-project rather than waiting until deadline week. Let me know if anything is unclear. — [Your name]
A maioria das conversas no trabalho dá errado não por má intenção, mas por falta de estrutura. As pessoas chegam com a mensagem certa, mas sem um roteiro — então ou enterram o ponto principal sob camadas de suavização, ou exageram na correção e partem para o ataque. De qualquer forma, o outro lado fica na defensiva e a conversa não chega a lugar nenhum.
O Que Este Prompt Faz de Diferente
O script gerado por este prompt se apoia em três princípios da pesquisa em resolução de conflitos. Primeiro: separação de fatos de juízos de valor. A seção de mensagem central obriga você a nomear eventos observáveis (prazos perdidos, impacto no negócio) em vez de avaliações de caráter ('você não é confiável'). Segundo: a arquitetura da escuta. A maioria dos modelos de script pula direto do problema para a solução. Este insere uma pausa obrigatória de escuta antes da resolução — porque pessoas que não se sentem ouvidas não conseguem se engajar produtivamente nos próximos passos. Terceiro: respostas pré-escritas para objeções. Os três padrões de evasão mais comuns (culpar o sistema, silêncio, alegar ignorância) ganham uma resposta personalizada que não cede nem escala o conflito.
Por Que a Frase de Abertura é Tão Importante
Os 60 segundos iniciais determinam se a outra pessoa entra em modo de resolução de problemas ou em modo de autodefesa. O prompt instrui o Claude a escrever uma abertura que sinalize curiosidade genuína — 'Não estou chegando com um veredito' — antes de apresentar qualquer fato. Isso não é suavizar a mensagem; é criar a segurança psicológica necessária para que a conversa realmente produza um resultado. Uma pessoa que se sente encurralada não vai revelar o verdadeiro motivo por trás de seu comportamento.
O Que Alimentar no Prompt
Quanto mais específicas suas entradas, melhor o script. Informações vagas como 'meu colega é difícil' geram um resultado genérico. Substitua os campos entre colchetes por: o cargo da pessoa, os incidentes observáveis específicos (com datas e números), qual resultado você realmente quer alcançar e sua avaliação honesta de como ela provavelmente reagirá. O Claude vai ajustar os cenários de objeção e a linguagem de resolução de acordo.
Funciona Melhor Com
Claude Sonnet 4.5 ou Claude Opus 4 produzem os scripts mais matizados e com consciência tonal. O GPT-4o também lida bem com este prompt. O segredo é usar um modelo com forte capacidade de seguir instruções e controle de tom — modelos mais leves tendem a gerar scripts que soam genéricos ou corporativos. Execute uma vez e depois faça uma revisão, pedindo ao Claude para ajustar qualquer linha que pareça artificial para a sua voz.