Test & Train your AI Agent
Testing your bot is very simple. To get started, log in to ZipChat, go to Chat Settings, and click the Test Chat button in the top-right corner. This will open a testing interface where you can interact with the agent and simulate different conversations.
Your agent continuously learns over time from conversations and from your website content. When you create your ZipChat account, we automatically scan all the information available on your website. After that, we perform a full rescan every week to ensure the knowledge base remains up to date.
At any time, you can manually improve or modify your agent’s knowledge. To do this, go to the AI Training section, where you’ll find three main areas: Knowledge Base, Corrections, and Prompt & Tools.
Knowledge Base
In the Knowledge Base section, you can remove pages that were scanned from your website or add new ones. You can also upload product feeds, PDF files, sitemaps, and even product catalogs that ZipChat can process and learn from. If needed, you can simply upload plain text files. This allows you to add specific information that may not already be included in the automatic scan.
Corrections
In the Corrections section, you can review and manage specific responses you’ve asked ZipChat to modify. You can create a correction directly from either the Test Chat or from real customer conversations. Under each message, you’ll find a Create Correction button. By clicking it, you can update the Knowledge Base or adjust the agent’s behavior as you see fit.
Prompt & Tools
This section contains the core of ZipChat: the Core Prompt, which defines how the AI behaves in general scenarios.
In Additional Instructions, you can define how ZipChat should behave in specific situations — for example:
How to respond when a customer asks to speak with human support
How to handle order tracking requests
What qualification questions to ask when a customer requests product recommendations
You’ll also find important tools here, such as Escalate Conversation, which allows you to transfer a conversation to human support when the AI is unable to resolve it independently.
Once your agent has completed scanning your website, you’ve tested a few conversations, and you’ve adjusted the Core Prompt as needed, your AI agent is ready to go live.
Our recommendation is to launch as soon as possible so you can start testing real conversations with real customers. From there, you’ll be able to review ZipChat’s responses and fine-tune the agent according to your preferences.
Last updated