Skip to main content

FAQ Generator

FAQ Generator analyses your chat history using AI to find repeated customer questions and automatically draft answers based on how your team has responded in the past. The result is a curated set of suggested FAQ entries you can review and save directly to your knowledge base — without writing a single word from scratch.

Key Benefits

  • Discover your most common customer questions automatically
  • Generate draft answers grounded in your team's real responses
  • Build your knowledge base in minutes, not days
  • Control quality by reviewing every suggestion before it goes live
  • Run analysis as often as you like — weekly, monthly, or on demand
  • No writing required — AI drafts, you approve

How FAQ Generator Works

The generator runs a three-stage AI pipeline on your selected date range of chat history.

1

Configure

Choose how far back to look (past week, month, 2 months, or 3 months) and set the matching precision (Broad, Balanced, or Precise). Broader settings surface more suggestions; Precise surfaces only high-confidence patterns.

2

Analyse

Click Analyse Chat History. The AI scans conversations in the selected period, clusters similar customer questions using vector similarity, and pairs each cluster with the agent responses used most often.

3

Review

Suggestions appear as cards showing the customer question, the AI-drafted answer, and how many similar conversations were found. You approve the ones you like (saving them to your knowledge base) and dismiss the rest.

Temporary results: Suggestions are held for 24 hours. Anything not approved or dismissed within that window is automatically cleared. Re-run the analysis at any time to get a fresh set.

Step-by-Step Guide

1

Open FAQ Generator

  • Go to Automations in the sidebar
  • Expand AI Assistant
  • Click FAQ Generator
2

Choose a Time Range

Select how far back the AI should look when scanning your chats.

Past week

Best for teams with high daily volume. Surfaces very recent patterns.

Past month

Good balance of recency and pattern volume. Recommended starting point.

Past 2 months

Useful for lower-volume teams or spotting seasonal trends.

Past 3 months

Broadest view. Good for building an initial knowledge base from scratch.

3

Set the Matching Precision

Precision controls how strictly questions must match before they're grouped together.

BroadLower threshold — more suggestions, wider variety. Good for initial knowledge base building or teams wanting maximum coverage.
BalancedRecommended for most teams. Surfaces patterns with good confidence without being overly strict.
PreciseHigh-confidence matches only. Fewer suggestions, but each one represents a clearly repeated question.
4

Run the Analysis

  • Click the Analyse Chat History button
  • A progress banner shows the job status (Queued → Processing → Completed)
  • Analysis typically takes 1–5 minutes depending on history volume
  • You can navigate away and return — the job runs in the background
5

Review Suggestions

Each suggestion card shows:

  • A title summarising the question cluster
  • How many similar conversations were found
  • The representative customer question (in blue)
  • The AI-drafted answer based on your team's past responses (in green)

For each card you can:

Save to knowledge base

Opens a confirmation modal where you can edit the question and answer before saving. Saved entries become available to the AI auto-reply and agent assist features immediately.

Dismiss

Removes the suggestion. Dismissed suggestions do not reappear unless the pattern strengthens in future analyses.

Configuration Reference

Broad

  • Minimum cluster size: 2 conversations
  • Similarity threshold: 60%
  • Most suggestions generated
  • Best for building initial KB

Balanced

  • Minimum cluster size: 3 conversations
  • Similarity threshold: 75%
  • Recommended for most teams
  • Good precision-to-volume ratio

Precise

  • Minimum cluster size: 5 conversations
  • Similarity threshold: 85%
  • Fewest suggestions
  • Highest confidence patterns only

Best Practices

Do

  • Start with Balanced precision and 1 month of history
  • Run analysis monthly to keep your KB fresh
  • Edit AI-drafted answers before saving if they need polish
  • Dismiss low-quality suggestions rather than leaving them
  • Use Broad precision when starting from an empty KB
  • Check the cluster size — larger clusters mean stronger patterns

Don't

  • Save suggestions without reading the drafted answer first
  • Run analysis on less than 2 weeks of data — patterns need volume
  • Use Precise precision when your chat volume is low
  • Ignore the 24-hour expiry — review suggestions the same day
  • Expect perfect answers — AI drafts are starting points, not finals

Troubleshooting

No suggestions generated?

  • Try a longer time range — more history gives the AI more patterns to work with
  • Switch to Broad precision to lower the matching threshold
  • Check that you have at least a few weeks of chat history in the system
  • If the job failed, wait a minute and re-run the analysis

Analysis job stuck or taking very long?

  • Jobs can take up to 10 minutes for very large date ranges
  • If the banner shows Processing for more than 10 minutes, try dismissing and re-running
  • Check your internet connection — the status banner requires a live connection to update

Suggestions are low quality or off-topic?

  • Switch to a higher precision setting (Balanced or Precise)
  • Use a shorter date range to focus on recent, more consistent conversations
  • Dismiss irrelevant suggestions — this helps future analyses

Saved a suggestion but it doesn't appear in the knowledge base?

  • Navigate to Knowledge Hub and check the Conversation Examples section
  • Newly saved entries may take a few seconds to index
  • Ensure you completed the Save modal and clicked Confirm
← AI & AutomationKnowledge Hub →
FAQ Generator — Fiko Docs