Overview
To ensure that your organization provides an exceptional chat experience for your customers, we always recommend that you review generated and prewritten responses regularly. In addition, it's critical that your URL links, imagery, and branding remain up to date.
Determine the Response Type
In instances where you've found outdated or incorrect information you'd like to change, you will first need to determine the response type providing the incorrect information. You can do so using either of the two options:
Option 1
Review the AI Resolutions Log report under Tools -> Data. More information on the report can be found here: AI Resolutions Log
Option 2
Retype the question in chat and if there is more than one phrase like “let me check”, ‘’one moment please’’, or ‘’checking’’ before a response, that means it was generated. It is a prewritten response if there is only one of those phrases followed by an ellipsis (...).
Make Response Adjustments
Once you determined the response type, follow the steps below to determine what actions must be taken to update the response.
Prewritten Responses
If you're seeing the following within a prewritten response:
An image is outdated
The response content is incorrect
Links within buttons/quick replies need adjustment
Follow the instructions below to make updates to your content!
Edit Prewritten ResponsesOnce you have found the corresponding response, determine the following:
Are the trained input samples conflicting with another response name? This may prevent the system from firing the correct response. You can either condense similar prewritten responses into one response or generate responses with data for more accurate replies.
Check that you've trained the corresponding response name correctly! Add additional input samples and ensure they are formatted correctly. More information: Before Getting Started
Generated Responses
Search for the wrong info using keyword phrases in the Response Library as shown below:
Locate Responses/DataOnce you have found the incorrect data, determine if it's data scraped from your website or from a documented data response (the content type field will tell you either "Document" or "Web Import"). Follow the instructions below to make updates to your content!
Update/Delete Existing Documented Data
Note that if the information you scraped on your website is wrong, your scraped data will be incorrect as well! We recommend data that share similar keywords or topics be direct and well laid out so it's easier to read by the LLM. For example:
Bad Data: Price - $15 (price for what?)
Better Data: GA Price - $15, Seniors Price- $10, Parking Price $5 (clear what the price is and what you're paying for)
Search for keyword phrases in the Response Library as shown below:
Locate Responses/DataOnce you have found the corresponding data, determine the following:
If the data is imported from your website, is there enough information provided to generate responses based on end-user questions?
If you want to provide more data than what's provided on your website, we recommend adding documented data using the instructions here: Add New Documented Data
If you want more control of the messaging, we recommend adding a new prewritten response using the instructions here: Add a New Prewritten Response
Learn how to add or edit generative enrichments by clicking the link below:
Enrich Generated Responses"No Info" Responses
Search for keyword phrases in the Response Library as shown below:
Locate Responses/DataIf no data or responses are found, either:
Search for keyword phrases in the Response Library as shown below:
Locate Responses/DataOnce you have found the corresponding data, determine the following:
If the data is imported from your website or in documented data, is there enough information provided to generate responses based on end-user questions?
Confirm that your prewritten response is trained correctly! If you do not have input samples for prewritten responses that are similar to end-user questions, the system may not know to fire that specific response.
Last updated
Was this helpful?