Why are there no changes in responses after modifying AI employee configurations?

If you encounter issues such as the changes not taking effect, the AI's response not changing, or the response still following the original configuration after modifying the AI employee's persona, knowledge base, skills, plugins, or other configurations, you can troubleshoot using the following methods.

This article's link: "Why are there no changes in responses after modifying AI employee configurations?"

I. Confirm whether this modification needs to be published.

Different configurations take effect in different ways. After making changes, you can check the status in the upper right corner of the AI employee configuration page:

If the status shows "Draft" and the "Publish" button is clickable, it means there are unpublished changes, and you need to click "Publish".
If the status shows "Published" and the button on the right shows "Published" and is unclickable, it means there are no changes to be published and there is no need to publish again.

1. Configuration to be published

After modifying the following content, you need to republish the AI employee:

Basic configurations include character design, output language, and AI response settings.
Add, remove, or change the knowledge base for AI employees.
Add, remove, or adjust plugins and tools.
Modify the skill configurations that need to be published, such as "Learning Ad Information".

If not released, official access channels such as the website, Messenger, and WhatsApp will still use the configuration that was previously released.

2. Content that automatically takes effect after saving.

The following content will take effect immediately after saving; there is no need to redeploy the AI employee:

Edit documents, questions and answers, or materials that have been linked to the knowledge base.
Modify "Collect customer information".
Modify "Update Customer Tags".

If, after saving, the top right corner of the configuration page still displays "Published" and the button on the right is unclickable, it means that there is no need to republish.

II. Eliminating the influence of the original session context

Historical messages within the same session serve as context for the AI employee to understand the current problem.

Even if the changes have taken effect, when testing in the original conversation, the AI employee may still respond based on previous chat content, making it appear as if the configuration has not changed.

Please process according to the test location:

1. Test in the "Question and Answer Test"

Click "Clear Conversation" and then resend the question to avoid the previous test message affecting the current response.

Once it is confirmed that the configuration to be released has been published, a new test account or a new visitor identity can be used to initiate a consultation to avoid the original chat content affecting the judgment.

III. Using issues that can verify modified content

After modifying the configuration, you should verify it using questions that are directly related to the changes.

For example:

The tone of the reply has been modified, allowing users to send a frequently asked question to check if the AI employee's expression changes.
After modifying the knowledge base content, you can directly query for newly added or modified information.
The business process has been modified to allow sending a complete question that can trigger the process.
If you modify a skill, plugin, or tool, you need to provide information that meets the call conditions.

If the test issue is unrelated to this modification, it may not be possible to determine from the response whether the configuration has taken effect.

IV. Use the troubleshooting assistant to confirm the cause.

After completing the above checks, if the AI employee's response still does not reflect the modified content, you can use the "Troubleshooting Assistant" to view the processing of this response and confirm the reason for the unexpected response based on the troubleshooting results.

The troubleshooting assistant can be accessed through the "Q&A Test," the HelpKnow analysis page, or the SaleSmartly chat page; the operation methods differ depending on the page. For details, please refer to: "How to Troubleshoot When AI Employee Responses Do Not Meet Expectations?"