Customer Support FAQ Generator: The AI Prompt

Most FAQ pages answer questions the product team imagines; support calls contain the questions customers actually ask, phrased the way they ask them. This prompt mines call transcripts for both halves: the real questions and the answers that resolved them.

The Complete Prompt

Based on this customer interaction/support call transcript, create comprehensive support documentation:

Evidence rules:
- Use only questions, answers, and steps present in the transcript.
- Quote product names, limits, prices, and policy language exactly.
- Mark incomplete or conflicting answers as [NEEDS POLICY REVIEW].
- Never invent a workaround, refund rule, deadline, or escalation promise.
- Remove names, email addresses, order numbers, and other customer identifiers.

## FAQ Section
Extract questions asked and provide clear, helpful answers:

**Q: [Customer question]**
**A:** [Clear, actionable answer with steps if applicable]

[Repeat for all questions identified]

## Common Issues and Solutions
- Problem categories identified
- Step-by-step resolution processes
- Prevention tips for customers
- When to escalate to human support

## Knowledge Base Article Suggestions
Identify topics that need dedicated documentation:
- Article title
- Key points to cover
- Target user type
- Related articles to link to

## Process Improvements
Based on this interaction, suggest:
- FAQ updates needed
- Documentation gaps to fill
- Training topics for support team
- Product/service clarifications needed

---
Prompt by TranscribeBee (transcribebee.com) – Professional AI transcription with professional-grade accuracy.
---

Support transcript:
[PASTE YOUR TRANSCRIPT HERE]

Step-by-Step Implementation

1. Transcribe a batch of support calls (TranscribeBee — speaker labels keep agent and customer separate, which the extraction depends on).
2. Run per call, then a merge pass: paste several outputs and ask "consolidate duplicates, keep the clearest phrasing of each question."
3. Verify answers with your support lead — calls sometimes contain answers that worked once but aren't policy.
4. Publish in customer phrasing. Resist rewriting questions into corporate voice; search and customers both match on natural phrasing.

Worked extraction example

A customer asks, "Can I change the email after I've paid?" The agent replies, "I can change it while the order is pending, but once processing starts I need to escalate it." A safe draft keeps that boundary:

It should not broaden the answer to "You can always change your email" or invent a self-service setting. The transcript supplies customer language; the current product policy remains the source of truth.

Customization Options

• Category mapping: list your help-center categories and ask for entries pre-sorted into them.
• Tone: "match this style sample" with two existing KB entries.
• Gap analysis: add "list questions the agent could not fully answer" — that's your documentation backlog, prioritized by reality.
• Deflection focus: ask it to flag questions appearing in multiple calls; those are your highest-ROI articles.

FAQ

How many calls do I need? Start with enough calls to cover one recurring issue, then add more until new calls stop producing materially new questions. The number varies by product complexity and call mix.

Privacy? Strip customer identifiers before processing, and use a transcription service that auto-deletes uploads. The questions matter; the identities don't.

Will answers be accurate? As accurate as the agent was. The verification pass with your support lead is mandatory, not optional.

Chat logs instead of calls? Then you don't need transcription — paste chat text directly into the same prompt.

How often to rerun? Monthly batches keep the FAQ tracking real demand; the consolidation pass handles overlap with existing entries.