Score a Support Agent Reply Against a Quality Assurance Rubric
Get a four-criteria QA scorecard table grading tone, accuracy, completeness, and next steps, plus a ready-to-send rewrite of your support agent's reply.
You getScorecard Table, overall QA Score and rewritten Reply
Prompt
Act as a senior customer support quality assurance lead. I will paste a customer's message and a support agent's draft reply below. Grade the draft reply against a four-criteria QA rubric and provide a rewritten version.
Rules:
- Score each criterion—Tone, Accuracy, Completeness, and Next Steps—on a 1 to 5 scale, where 5 means ready to send with no edits.
- Cite one specific phrase from the draft reply in every scorecard row to justify the score.
- Flag any policy or factual claims that cannot be verified from the customer's message alone.
- Write the improved reply in plain, empathetic language under 180 words with a clear, single next step.
Output format:
1. Scorecard Table: Markdown table with columns for Criterion, Score (1-5), Evidence from Draft, and What to Fix.
2. Overall QA Score: Total out of 20 and a one-sentence pass or revise verdict.
3. Rewritten Reply: The full, ready-to-send message.
End with one line noting any missing account context or unverified policy details.Example output
Tips and common edits
Why this works
- Forces the model to cite exact phrases from the draft so feedback is concrete and coachable.
- Separates tone, accuracy, completeness, and next steps into a scannable table instead of vague prose.
- Caps the rewritten reply at 180 words with a single next step to keep the customer focused.
Use this in ChatGPT, Claude, or Gemini before sending a sensitive reply or during weekly QA audits. Paste the customer's original message and the agent's draft reply directly below the prompt.
Common edits
- Add your company's specific refund window or SLA rules to the prompt for stricter accuracy checks.
- Change the 1 to 5 scale to Pass/Fail if your team uses binary QA scoring.
Make it repeatable
Keep using this prompt without pasting it in every time.
Prompt details
- By
- Prompts-GPT editors
- Published
Did this prompt work?
Keep exploring
More like this
Reply to a Public Social Media Complaint and Move It to DM
Get a 280-character public social media reply that de-escalates a complaint, plus a matching direct-message follow-up script to resolve it privately.
Respond to an Abusive Ticket While Solving the Issue
Generates a calm customer support reply that establishes clear conduct boundaries while resolving the underlying technical problem.
Reply to an Angry Customer Complaint Without Blame or Vague Promises
Paste an angry customer message and get a calm reply that names their exact problem first, skips blame-shifting phrases, and ends with one dated next step.
Reply calmly to a negative public customer review
Draft composed, professional public responses to critical reviews that acknowledge mistakes, de-escalate tension, and move resolution offline.
Draft a Bug Acknowledgment Email Without an ETA
Generate a clear support email that validates a reported bug, provides a safe temporary workaround, and sets realistic follow-up expectations.
Write Executive Escalation Email for At-Risk Customers
Generate a professional, high-impact email from a VP to retain a churn-risk customer by offering ownership, a concrete fix, and direct communication.