Training Suggestions

Overview

Your AI agent suggests its own training improvements. Instead of manually identifying gaps and writing new training content, the system surfaces opportunities for you to review and approve with a single click.

This creates a powerful feedback loop: unanswered questions become improvement opportunities, great responses get reinforced, and your agent gets smarter over time with minimal effort.

How It Works

Training Suggestions identifies improvement opportunities in two ways:

Proven Q/A Pairs

When users give a thumbs up to an AI agent response, we surface that question/answer pair as a suggestion. These are battle-tested responses that real users found helpful, making them easy wins for expanding your training data.

• Zero additional cost: We're just surfacing existing data
• Validated by real users: These responses actually helped someone
• Quick to review: One-click approval in most cases
• Unlimited: Available on Plus, Pro, and Enterprise plans with no monthly limit

AI-Drafted Answers

For questions the agent couldn't answer, our system drafts a suggested response by combining:

• Your agent's existing training data
• The AI's broader knowledge base
• Context from the conversation

This is different from the live agent, which is intentionally restricted to only answer from what it's been specifically trained on. For suggestions, we temporarily loosen those restrictions to give you a strong starting point that you can refine or just approve.

The Feedback Loop

Training Suggestions ties directly into the Knowledge Coverage stat you see in your dashboard. Here's how the loop works:

1. Problem identified: An agent can't answer a question, hurting the Knowledge Coverage percentage.
2. Solution generated: The system drafts a suggested answer or surfaces a proven response.
3. You review: Approve, edit, or dismiss the suggestion.
4. Improvement measured: Approved suggestions are added to your Training Sources as individual items and indexed immediately.
5. Coverage improves: Your Knowledge Coverage percentage increases as gaps are filled.

This creates a clear path from “here's a problem” → “here's the solution” → “here's the measurable improvement.”

Using Training Suggestions

To access and manage suggestions:

1. Go to AI Agents in your dashboard
2. Click on an agent, then select Training Suggestions from the menu
3. Review suggestions in two tabs: AI-Drafted Answers and Proven Q/A Pairs

For Each Suggestion

You have three options:

• Approve: Add the Q/A pair to your training data immediately. It's indexed right away and the agent can use it in future conversations.
• Edit & Approve: Modify the response before approving. Great for adding details or adjusting tone.
• Dismiss: Remove the suggestion if it's not relevant or useful.

Bulk Actions

For efficiency, you can select multiple suggestions and approve or dismiss them all at once. This is especially useful when you have many proven Q/A pairs to review.

Managing Approved Suggestions

Every suggestion you approve is saved as an individual item in your Training Sources. You can view and manage them from the AI Learning tab inside your agent's Training Sources page.

This means you can remove a specific approved suggestion at any time without affecting the rest of your training data. Each item shows the original question, its indexing status, and a word count so you know exactly what is contributing to your agent's knowledge.

Removing an Approved Suggestion

1. Go to your agent's Training Sources page
2. Click the AI Learning tab
3. Find the suggestion you want to remove and click the delete icon
4. Confirm the deletion. The item is removed from your training data and the agent's knowledge is updated on the next indexing cycle.

Automatic Cleanup

Pending and dismissed suggestions are automatically removed after a set period to keep your account tidy. Approved suggestions are not affected by this cleanup. They stay in your Training Sources until you delete them.

The cleanup window varies by plan. Higher plans retain pending and dismissed suggestions longer, giving you more time to review them before they are cleared.

Plan Limits

AI-drafted suggestions are limited by plan to manage costs:

The AI draft limit resets at the beginning of each billing cycle. Proven Q/A pairs are unlimited on Plus, Pro, and Enterprise plans since we're just surfacing existing data.

Requirements

To use Training Suggestions, you need:

• Active paid subscription: Plus, Pro, or Enterprise plan
• Minimum training data: At least 5 pages (~3,750 words) of training content. This ensures the AI has enough context to generate quality suggestions.

Best Practices

Get the most out of Training Suggestions:

Review Regularly

Set a weekly cadence to review suggestions. Even 10 minutes a week can significantly improve your agent's knowledge coverage over time.

Prioritize High-Confidence Drafts

Start with high-confidence AI drafts. They're usually ready to approve as-is and represent the quickest wins.

Edit for Accuracy

While AI drafts are good starting points, always verify accuracy for critical information like pricing, policies, or technical details. A quick edit ensures your training data stays reliable.

Watch Your Coverage Grow

After approving suggestions, check your Knowledge Coverage percentage in the dashboard. Seeing the number go up is great motivation to keep improving.