Chat Training

Train your agent on real customer conversations from Facebook, Instagram, and WhatsApp with automatic retraining

Chat Training#

Build your AI's knowledge from your Facebook, Instagram, and WhatsApp conversations. Your past chats are a goldmine of training data -- real questions, real answers, and your natural communication style.

Email inbox threads stay separate and are not used for Chat Training.

Why Import Conversations?#

Your past conversations contain:

• Real questions: What customers actually ask
• Real answers: How you actually respond
• Your voice: Your natural communication style and tone
• Edge cases: Unusual situations you have handled before

Supported Platforms#

Setting Up#

Step 1: Connect Platforms#

Before training, connect your messaging platforms:

1. Go to Settings > Chat Training
2. You will see the connection status for each platform
3. Click Connect for any unconnected platform
4. Complete the OAuth authorization

The interface shows connection status, connected page or account names, and action buttons for each platform.

Step 2: Preview Conversations#

Before training, preview what data is available:

1. Click the eye icon next to a connected platform
2. Browse conversations with participant names and message counts
3. Click any conversation to read the full thread
4. Business owner messages appear on the right, customer messages on the left

Training Your Agent#

Starting Training#

1. Click Train Agent
2. If you had a previous incomplete training, you will be offered to Resume or start fresh
3. If only one platform is connected, you will see a recommendation to connect others for better results
4. Configure your training options

Training Options#

Data Sources: Select which connected platforms to include.

Extraction Types (toggle each on or off):

• Q&A: Extract question-and-answer pairs from your conversations
• Style: Learn your communication style and tone
• Insights: Extract business insights and patterns

Additional extraction types may be available depending on your plan and connected integrations.

Training Mode:

• Append: Add new data alongside existing training data
• Find New: De-duplicate with existing training to avoid overlaps
• Overwrite: Full rebuild from scratch (replaces all previous training)

Content Enrichment: Photos, links, videos, audio, and documents can be included (toggleable per type).

During Training#

A progress display shows real-time status:

• Connecting then Fetching then Processing then Extracting then Saving then Completed
• Each step shows its status, message, and duration
• You can cancel training at any time
• Extracted counts update in real-time

After Training#

The results view shows:

• Total conversations trained (with per-platform breakdown)
• Extraction counts by type
• Integration recommendations based on your conversation patterns
• Last training timestamp

Reviewing Extracted Knowledge#

After training, the review area groups what was learned into four tabs:

The review area also surfaces pending counts and blocks approval of any Q&A that has unresolved conflicts.

Editing Suggested Alerts#

Alerts come with editable fields so you can tune how your agent recognizes the pattern:

• Trigger name -- a short label
• Customer phrase examples -- a handful of realistic ways customers phrase the request
• Signal profile -- the family this belongs to (e.g. delivery issue, payment issue), plus positive and negative examples
• Suggested response -- optional auto-reply text
• Auto-reply allowed -- toggle whether the agent may reply on its own

These controls let you refine the vocabulary and meaning of a trigger beyond simple keyword matching.

Social Feed Learning#

Your Facebook and Instagram public posts are handled separately from DM training through the Social Feed surface on the same page.

Each post is categorized as:

• Knowledge -- durable content worth adding to the knowledge base
• Temporary update -- time-sensitive content (promos, events, hours)
• Ignore -- not useful for training
• Needs review -- owner must confirm before routing

Social feed settings live under the feed surface and let owners configure:

• Enable / disable social feed learning
• Automation mode -- Off, Notify First, Auto Low-Risk, Auto Most
• Notification channel -- where approval requests are sent
• Sharing -- whether posts share links or images with customers
• Temporal windows -- how long temporary updates stay active before archiving

Social feed learning is configured separately from DM/chat-based training so promos and announcements never overwrite durable knowledge by accident.

Processing Time#

You will see real-time progress during training and can view results after completion.

Integration Recommendations#

After training, AlonChat analyzes your conversation patterns and suggests integrations that could improve your workflow. Recommendations are based on the types of requests your customers commonly make (scheduling, payments, file sharing, etc.).

Click any recommendation to set up that integration on the Actions page.

Auto-Training#

Enable automatic retraining so your agent stays up-to-date without manual intervention:

1. At the top of the Chat Training page, find the Auto-train toggle
2. When enabled, your agent automatically retrains after accumulating enough new conversations
3. The conversation threshold is configurable

When auto-training is active, the toggle shows a highlighted state. You will receive a notification when auto-training completes.

Privacy and Security#

Your conversation data is:

• Encrypted during upload and storage
• Private to your account only
• Not shared with any third parties
• Deletable at any time

Best Practices#

1. Connect multiple platforms -- combining Facebook, Instagram, and WhatsApp gives the best training results
2. Enable auto-training -- keep your agent learning without manual intervention
3. Select quality conversations -- skip spam or irrelevant chats
4. Include variety -- different topics and situations produce better training
5. Recent is better -- focus on conversations from the past year
6. Review extracted Q&As -- check quality before relying on auto-approved pairs
7. Resolve conflicts first -- unresolved conflicts block the matching Q&A from being approved

Related Pages#

• Business Vocabulary -- Teach the agent which words mean the same thing
• Voice -- Configure voice surfaces and language
• Learned Style -- How your agent learns your communication style
• Q&A Sources -- Manually create Q&A pairs
• Facebook Messenger -- Connect the page used for Chat Training
• Instagram DM -- Connect the account used for Chat Training
• WhatsApp Business -- Connect the number used for Chat Training