Best Practices

Improve your AI Agent’s reliability and usefulness by following these best practices. Clear instructions and well-organized data sources help ensure a better experience for your users.

Refine AI Agent Instructions

Instructions act as the "brain" of your AI agent, defining its persona, tone, and behavior. Precise instructions ensure your AI remains focused and helpful. These are configured in your Agent Persona settings, where you can define the agent prompt and restrictions.

Example: Support Agent Instructions

You can copy and adapt this example for your own agent. Update the bracketed information to fit your business.

For more details, visit the Agent Persona guide. You can also explore Prompt Templates for industry-specific examples and best practices.

Optimize Data Source Quality

The quality of your AI agent's responses is directly tied to the clarity of your data sources. While YourGPT uses advanced scraping technology, well-structured content always yields better results. Learn more about all available Training Sources and how to add them effectively.

Advanced Training for URLs

Use Advanced Training Options when a site needs custom crawling or cleanup. You can switch between Custom and With AI.

Advanced Training Options for URLs

With Advanced Training Options, you have precise control over how YourGPT gathers content from a website. Instead of simply selecting a list of actions, think through the flow you want the crawler to follow.

For example, you might want the system to interact with dynamic content by clicking buttons, scrolling to reveal more information, waiting for certain elements to load, or even running small scripts if the site is particularly complex. Whenever you add an action that interacts with a part of the page, specify its CSS selector (like an ID, a class, or an attribute) so the tool knows exactly where to act. If content doesn't appear right away, you can insert a waiting period (measured in milliseconds) after a step to ensure everything loads before moving forward. To confirm that your configuration works as you intend, use the "Test URL" feature and make sure the crawler’s path brings up all the content you need before you save your setup.

Choosing what information to extract is just as important as defining the actions. For clarity, leave the "Extract only main content" setting turned on—this tells YourGPT to focus on the essential text and avoid picking up unnecessary parts like navigation bars or ads. If there are specific sections you want to ensure are included, add their selectors in the "Include Elements" area. If there’s page clutter you’d rather exclude (such as pop-ups, cookie banners, or footers), list those selectors in the "Exclude Elements" area. When you’re working with long pages, consider adjusting the chunk size and the amount of overlap between chunks so YourGPT breaks information up logically, which helps the AI keep related context together. The "Optimised Chunking" option further refines this process for lengthy or detailed pages.

For sites that change over time, set a retraining period to keep your data current—this way, YourGPT will revisit and update the content automatically as the site evolves. If a website is fairly static, you can enable "Skip Retraining" so it won’t be revisited. Finally, for better organization and future analysis, add descriptive tags to your sources. Tags let you categorize and quickly filter URLs for analytics and management as your collection of training sources expands.

Readability Comparison

Compare these examples to understand the difference between poorly and well-structured training data:

Helping Your AI Agent Share Useful Links

Enabling your AI agent to recommend and share relevant links (like pricing pages, support forms, or documentation) requires that you explicitly include those URLs in your knowledge base or training data. This is crucial, especially if you're using Website Links: the AI can read the text of a page but does not inherently know the URL.

Pricing: https://yoursite.com/pricing
Support: https://yoursite.com/contact
Docs: https://yoursite.com/docs

Steps to Enable Effective Link Sharing

1. Pick the right training source type for navigation information:

   • Plain Text/FAQs: Simple lists like "Contact: [link]"
   • Documents: Full site maps or more complex collections of links

2. Write clear, structured content pairing each destination ("Pricing Page," "Contact," "Docs," etc.) with its exact URL.

3. Add or upload this source directly into your knowledge base from the Training Sources section.

4. Test your chatbot by asking for specific links (e.g., "What's your support page?") to confirm it provides correct, clickable URLs.

Tip: Maintain your navigation links in one easily updatable source—this keeps them current and helps your agent guide users efficiently.

Visual Responses with Markdown

You can enable your AI agent to send images by including specific instructions in your Agent Persona or providing image URLs in your training data using standard Markdown format: ![alt text](image-url). This works across all Chatbot Widget integrations.

"When a user asks about our team, show this image: ![Our Team](https://example.com/team-photo.jpg)"

Select the Right AI Model

YourGPT offers multiple AI models to match different needs. The model you pick impacts response quality, speed, cost, and support for features like Functions.

Model Types and Recommendations

• Quick Support (Fast FAQ/Basic Q&A): Use Gemini Flash or GPT-5.1-mini for speed and efficient customer service.

• Complex Queries (Deep Reasoning): Go with or Claude 4.5 Sonnet for technical support, analysis, and in-depth conversations.

• Balanced (Speed & Quality): Choose Claude Haiku 4.5 for a middle ground of quality and cost.

See the AI Models comparison page for detailed specs and guidance on which model best fits your use case.

Refine Responses with "Revise" and Smart Learning

Don't forget to monitor your conversation logs in the Activity tab. If you see a response that could be improved, use the Revise button. This adds the correction directly to your Q&A training data, ensuring the AI learns from its mistakes.

Use Smart Learning

YourGPT's Smart Learning feature uses Reinforcement Learning from Human Feedback (RLHF) to continuously improve your chatbot. It:

• Collected feedback from users during conversations
• Identifies unresolved queries that need attention
• Generates FAQs from past conversations for review
• Learns from human-guided improvements

Enable this feature to let your AI agent improve automatically over time based on real user interactions.

Use Functions for Enhanced Capabilities

Functions extend your AI agent's capabilities beyond knowledge base responses. They enable your agent to autonoumously perform dynamic, real-time operations.

Getting Started with Functions

Use Automated Responses for Common Queries

Automated Responses let you instantly address frequently asked questions without consuming AI credits, delivering rapid responses and reducing operational costs.

Setting Up Automated Responses

To set up Automated Responses:

• Identify common queries by reviewing your Session Logs for questions that come up often.
• Go to Training → Automated Responses in your dashboard.
• Add keyword patterns and specify helpful responses for each frequently asked query.
• (Optional) Set escalation triggers for phrases such as "talk to human" or "connect agent" to involve live support when needed.
• Test your automated responses to confirm they're triggering as intended and delivering clear, instant answers.

Test and Iterate

Before deploying your chatbot, thoroughly test it using the Testing Guide. Share test links with your team and gather feedback. Use the Debug Lab to troubleshoot any issues.

Testing Workflow

• Get your test link from the Integration tab in your dashboard; see the Testing Guide for detailed steps.
• Test core scenarios: common questions from your knowledge base, edge cases and unusual queries, multi-turn conversations, and function calls (if enabled).
• Review Session Logs in the Debug Lab to understand how the AI is responding.
• Use Training Audit to verify which training sources are being used for specific queries and check match scores.
• Address issues using debugging guides: Incorrect responses, Incomplete answers, Refusals to answer.
• Iterate and improve by updating training data, refining instructions, or adjusting model settings based on test results.

Next Steps

Now that you've optimized your AI agent, consider:

If you need a deeper dive, start with the full Training Guide and the Agent Management guide to keep improving your agent after launch.