Support QA reviewer GPT that scores agent replies against your quality rubric and writes coaching notes
A custom GPT for support team leads and QA specialists: paste a ticket conversation and it scores the agent's replies against your QA rubric, quotes the evidence for each score, and writes short, specific coaching notes the agent can act on.
Agent in Customer support by AISkillHub Library
What it does
Most support teams review only a small sample of tickets for quality, and the reviews vary with whoever is doing them. This GPT applies one written rubric to every conversation you paste, shows the evidence behind each score, and turns the result into coaching the agent can use on the next ticket.
**What it does**
- Reads a full ticket conversation: customer messages, agent replies and internal notes if you include them. - Scores each rubric criterion. The default rubric covers understanding the issue, accuracy against the provided policy or knowledge, completeness, tone and empathy, clarity, ownership of next steps, and process (tagging, escalation, internal notes). Upload your own rubric and it uses that instead. - Quotes the agent's own sentence for every score, so the agent can see exactly what was judged. - Separates **auto-fail** issues from scored criteria. Auto-fail issues include sharing another customer's data, promising something policy forbids, missing a required escalation, and rude or dismissive language. - Writes coaching notes: one strength to keep, one or two changes, and a rewritten example of the weakest sentence. - With several conversations from the same agent, summarizes patterns across them.
**When to use it**
- Weekly QA sampling - Calibration sessions, where leads compare their own scores with the GPT's evidence - Agent self-review before a 1:1
**What it will not do**
- It does not check facts against your systems. Accuracy is judged only against policy and knowledge files you provide. Without them, accuracy is "not assessed". - It does not produce performance ratings, rankings of agents, or recommendations about discipline or pay. It reviews conversations, not people. - It does not judge the customer. If the customer was abusive, it notes it as context only.
**Example**
Agent reply: "As I said before, you need to clear your cache."
Output: Tone 2 of 4. Evidence: "As I said before" reads as impatient on a repeat contact. Clarity 2 of 4: no steps given. Coaching rewrite: "Sorry this is still happening. Let's try clearing the cache together: 1. Open Settings..."
How to use it
**Set up as a ChatGPT custom GPT** (requires a ChatGPT plan that lets you create GPTs).
1. In ChatGPT, open the GPTs area from the sidebar (Explore GPTs) and choose **Create**. 2. Switch to the **Configure** tab instead of using the chat-based builder. 3. Set **Name** to `Support QA Reviewer` and write a one-line **Description**. 4. Open `assistants/support-qa-reviewer/instructions.md`, copy everything, and paste it into the **Instructions** field. The file is written to fit inside the field's character limit. 5. Before saving, replace the variables in the pasted text: `{{team_name}}`, and `{{passing_score}}` (for example `80%`). 6. Add **Conversation starters**: `Score this conversation:`, `Review these 5 tickets from one agent and summarize patterns`, `Calibrate: here is my score, compare with yours`, `Show me the rubric you are using`. 7. Under **Knowledge**, upload your QA rubric (or a filled-in copy of `assistants/support-qa-reviewer/qa-rubric-template.md`), plus the policies and macros agents are expected to follow, so accuracy can be checked. 8. Under **Capabilities**: turn off web search and image generation. Accuracy must be judged only against your uploaded material. 9. Test it in the preview pane on the right with a real example, then save and choose who can access it (only you, anyone with the link, or your workspace, depending on your plan).
**Privacy:** remove customer names, emails and account numbers before pasting, and refer to agents by first name or ID, following your organization's rules on putting employee and customer data into AI tools.
Files
Save each file at the path shown, relative to your project root.