Inserting Attributes in Prompt Instructions

Attributes are dynamic external values that AssistGPT can fetch from the system or receive as user inputs. These values are inserted in place of variables before generating a response, enhancing contextual information for greater accuracy.

To insert attributes in the instruction field:

On the Prompt Template page, from the General tab, navigate to Instruction.
Click the Insert Attribute button.

You can add a maximum of 10 attributes for a single prompt.
The Insert Attribute window opens. The options differ based on the selection in the previous fields. If you are creating this prompt for the Knowledge Console, then from the Insert Attribute window, click the dropdown button and select one of the following options:
1. Title: This attribute refers to the current article title. When you include this attribute in the instructions, AssistGPT considers the article's title and offers solutions accordingly. For instance, when you create a prompt to generate an article and use the title attribute in the instructions, AssistGPT will replace this attribute with the current article title that the author is working on in the Knowledge Console. It generates a response based on this title.
2. Content or Selected HTML: This attribute allows you to send the entire content in HTML format. If you select a specific portion of the content, only that selected HTML will be sent to the AssistGPT for output generation. This attribute helps preserve the existing structure of the content, such as tables, bullet points, etc., while passing it to the AssistGPT. For example, if you are creating a prompt to improve the existing content, it is helpful for AssistGPT to use the content in its existing format to enhance it further.
3. Content or Selected Text: This attribute is similar to the one above, but it sends the content in plain text format. That is, the format and structure of the content are not preserved when it is sent to AssistGPT for output generation. This attribute is useful when creating prompts that generate content without using existing formats.
4. User Input: When you select this attribute, users can provide a label for their input. If you include this attribute in the prompt instructions, the Knowledge Console prompts the author for text input when they select the prompt. Once they provide the input and click Generate, the AI will generate the response accordingly.
If you are creating this prompt for the Advisor Desktop for Emails, then from the Insert Attribute window, click the dropdown button and select one of the following options:
1. Reply Text: When you insert this attribute in the instruction, it uses the text in the reply pane area, the message draft, and the entire email thread and passes it to the AssistGPT to provide an accurate response. For example, while creating a prompt for Generating Email, using this attribute in the instructions provides the relevant context of the entire mail conversation that the AssistGPT can use to deliver a response.
2. Search Results: When you insert this attribute, click the dropdown button for the Search Attribute and select the value on which the search should be performed. Depending on the selected search attribute, a search is performed on the portal selected for the agent at the queue level. This attribute passes the top five search results to the AssistGPT to provide a relevant response. This attribute is helpful when creating prompts that require a response to customer queries, and it can leverage the knowledge base to provide an accurate and business-specific response to customer queries.
3. User Input: When you select this attribute, you can also provide a label for the user input. If you include this attribute in the prompt instructions, the Advisor Desktop prompts the author for text input when they click the prompt. Once they provide the input and click Generate, the AI will generate the response accordingly.
4. Classifications: When you insert this attribute in the instructions field, the classifications configured for the system are passed to the AssistGPT to comprehend the classifications' structure and classify the information in the search results. The classifications exist in a tree format and are processed by the AssistGPT in a flat format. For example, the parent folder of the classification tree consists of the following categories: Credit Card and Bank Account. The parent category Credit Card has the following sub-categories: Cancel, Fraud, and New. The other parent folder, Bank Accounts, has the following sub-categories: Savings and Checkings. When this attribute is passed to AssistGPT to generate a response, the classifications are sent in a flat format such as Credit Card, Credit Card - Cancel, Credit Card - Fraud, Credit Card - New, Bank Account, Bank Account - Savings, Bank Account - Checking. The prompt response in the Advisor Desktop for Emails is presented as comma-separated values.
If you are creating this prompt for the Advisor Desktop for Chats, then from the Insert Attribute window, click the dropdown button and select one of the following options:
1. Reply Text: When you insert this attribute in the instruction, it uses the text in the reply pane area of the chat and passes it to the AssistGPT to provide an accurate response. This attribute can be leveraged to pass relevant chat context to AssistGPT to provide a quick solution to the customer query.
2. Chat Conversation: When you insert this attribute in the instruction, it includes the chat transcript, customer persona, and agent messages as context for the AssistGPT to provide a relevant solution. For example, while creating a Generate Summary prompt, it is essential to provide the context of the chat conversation to AssistGPT, and inserting this attribute in the instructions provides the relevant reference to provide an accurate response.
3. Classifications: When you insert this attribute in the instructions field, the classifications configured for the system are passed to the AssistGPT so that it can comprehend the structure of classifications and classify the information in the search results. The classifications exist in a tree format and are processed by the AssistGPT in a flat format. For example, the parent folder of the classification tree consists of the following categories: Credit Card and Bank Account. The parent category Credit Card has the following sub-categories: Cancel, Fraud, and New. The other parent folder, Bank Accounts, has the following sub-categories: Savings and Checkings. When this attribute is passed to AssistGPT to generate a response, the classifications are sent in a flat format such as Credit Card, Credit Card - Cancel, Credit Card - Fraud, Credit Card - New, Bank Account, Bank Account - Savings, Bank Account - Checking. The prompt response in the Advisor Desktop for Chats is presented as comma-separated values.
4. Customer Screen Name: When you insert this attribute in the instruction field, it uses the customer name entered in the pre-chat form in the chat transcripts. This helps distinguish the customer text from the agent text. This attribute is commonly used where chat transcripts are used.
5. Search Results: When you insert this attribute, click the dropdown button for the Search Attribute and select the value on which the search should be performed. Depending on the selected search attribute, a search is performed on the portal selected for the agent at the queue level. This attribute passes the top five search results to the AssistGPT to provide a relevant response. This attribute is helpful when creating prompts that require a response to customer queries and can leverage the knowledge base to provide an accurate and business-specific response to the customer query.
6. User Input: When you select this attribute, you can also provide a label for the user input. If you include this attribute in the prompt instructions, the Advisor Desktop prompts the author for text input when they click the prompt. Once they provide the input and click Generate, the AI will generate the response accordingly.
Once you have selected the desired attribute, click the Done button.