The prompts below cover the full customer service workflow: ticket triage, response drafts, knowledge base authoring, escalation routing, customer health monitoring. Each is annotated with when to use it. Copy any into Claude — they work best inside a Project loaded with your brand voice, product documentation, and past customer responses.
Read this ticket: [PASTE TICKET] Classify it on: - Severity (P0/P1/P2/P3) - Category (billing, bug, how-to, feature request, complaint, other) - Sentiment (calm / frustrated / angry / churn risk) - Time-sensitivity (immediate / 24hrs / 48hrs / standard) - Recommended next action (auto-respond, route to Tier 1, escalate to Tier 2, escalate to engineering, escalate to leadership) Output as JSON. Be conservative on severity — when in doubt, escalate up.
Here's the customer's last 12 months of interaction history: [PASTE] Current ticket: [PASTE] Summarize for the rep: - 1-sentence customer profile (size, plan, tenure) - Their last 3 ticket types and resolution - Sentiment trajectory (improving / stable / declining) - Any specific notes about this customer (e.g., known champion, known difficult) - Recommended response tone for this specific person Output: 6 bullets max. Goal: rep can read this in 30 seconds before responding.
Customer ticket: [PASTE] Customer context: [PASTE summary] Relevant KB articles: [PASTE links/summaries] Draft a response that: - Opens by acknowledging the specific issue (not "thank you for reaching out") - Addresses the actual question with specifics from the KB - Avoids over-promising or making commitments outside our standard policy - Closes with a clear next step (resolved / awaiting info / escalating) - Matches the voice in the example responses in the knowledge base Length: 100-150 words. No marketing language. Sound like a competent human, not a chatbot.
Customer ticket (angry): [PASTE] Background on the situation: [PASTE] What we can actually offer (within policy): [PASTE] Draft a de-escalation response that: - Opens by acknowledging their frustration (specifically, not generically) - Takes ownership where it's our fault — does NOT take fake ownership where it isn't - States what we CAN do in concrete terms - Avoids defensive language, justification, or "I understand how you feel" - Closes with a specific commitment and timeline Length: 80-120 words. Tone: calm, specific, accountable. Never use "unfortunately" — it signals weakness.
What happened: [PASTE] Impact on customer: [PASTE] What we're doing to fix it: [PASTE] What we can offer as remediation: [PASTE] Draft an apology email: - Opens with what happened, in plain language, without burying it - Acknowledges specific impact (not "we apologize for any inconvenience") - States what we're doing to ensure it doesn't recur - Offers the remediation directly — don't make them ask - Closes with the named person they can reach if they want to discuss further Length: 150-200 words. Tone: serious, specific, accountable.
Feature requested: [PASTE] Our current state with it: [not on roadmap / on roadmap / launched in beta / etc] Workaround if any: [PASTE] Draft response: - Thank them for the specific suggestion (not generic gratitude) - State honestly where we are with it (don\'t over-promise; don\'t under-share) - Offer the workaround if there is one - Tell them how product input flows in (so they know it was heard) - Close with a direct ask to share more context if they want Length: 100-130 words. Tone: respectful, honest, specific.
Recurring ticket pattern: [PASTE 3-5 representative tickets] Our best answer: [PASTE existing response template if you have one] Write a knowledge base article that: - Has a clear, search-friendly title (how a customer would actually search) - Opens with the answer in 1-2 sentences - Provides the full explanation with screenshots/steps as needed - Includes "if this doesn\'t work, try X" troubleshooting section - Ends with "still stuck? contact us" with a link Match voice and structure of existing KB articles in the Project.
KB article title: [PASTE] Current search queries that aren\'t finding it: [PASTE] Article URL: [PASTE] Suggest: - 5 alternative titles that better match search intent - 10 alternative keywords/tags that should be added - The single change that would most improve findability - Whether this article should be split into 2+ articles (and how)
For each customer in this list: [PASTE] Their recent interactions: [PASTE recent tickets, NPS scores, usage data] Score each on churn risk (1-10) and explain. Look for: - Decreasing ticket volume followed by sudden spike - Multiple unresolved tickets - Sentiment trajectory worsening - Lower-than-expected usage given their plan - Champion turnover - Tickets specifically mentioning competitors or "evaluating alternatives" Return a table: Customer | Risk score | Top signal | Recommended action.
Customer: [PASTE] Last QBR notes: [PASTE] Last quarter\'s interactions: [PASTE] Their usage and KPIs this quarter: [PASTE] Their original goals when they signed: [PASTE] Prepare: - Executive summary (3 sentences) - Wins this quarter — specific, with numbers - Issues that came up and how we resolved them - Where they\'re tracking against their original goals - 2-3 specific recommendations for next quarter - Expansion opportunities (be specific — not "upsell") - Potential risks I should raise carefully Format: 1-page brief, dense but readable. Voice: trusted advisor, not vendor.
These prompts work 5x better inside a Claude Project with your product documentation, brand voice, and past responses loaded. For setup help, see How to use Claude for customer service.
For larger teams wanting this wired into Zendesk, Intercom, or Front, our Implementation engagement includes building Customer Service Claude Projects with your specific tooling.