Triggers+ChatGPT integrates with ChatGPT. In the configuration, you need to select the Open AI ChatGPT Action to make for this Workflow.
In this Article
Selecting the OpenAI ChatGPT Action
Categorize
Categorize Text From
Categories
Category Description (the Instructions to ChatGPT)
Adding More than One Category
Using Categorize to Direct Tickets to a Specific Group or Agent
Detect Thank You Only
Translate
Translate Direction
Translating Inbound Messages
Translation Direction Selected as Inbound - Translate Customer Messages as Internal Notes
To Language for Inbound Messages
Translating Outbound Messages
Translation Direction Selected as Outbound - Translate Internal Notes as Public Replies
To Language
Internal Note starts with
Translation Direction Selected as Custom - Choose the field to translate
From Field
To Language
Extract Value
Extract Value From
What to extract
Format of the value to extract (the Instructions to ChatGPT)
Custom Prompt
User Message
System Instructions
Selecting the OpenAI ChatGPT Action
For the OpenAI ChatGPT Action, select one of the following:
- Categorize
- Detect Thank You Only
- Translate
- Extract Value
- Custom Prompt
The Open AI ChatGPT Action you select determines the selections for the fields that follow in the configuration.
Categorize
The Categorize action extracts from a part of the ticket and uses this to classify a ticket based on a list of possible topics. You can use this to automatically tag tickets, for example, with contact reasons, or to detect sentiment or emotion.
Categorize Text From
When you select Categorize as the Open AI ChatGPT Action, you are prompted to select where to Categorize Text From. This determines what part of the ticket ChatGPT will check to find the text used in the categorization.
Select one of the pre-set ticket fields.
Categories
You then add the Categories to set up in this Workflow. You can add as many Categories as needed for your requirements.
Click the + button next to Categories to add a new Category.
This displays the fields to add the Category Name and the Category Description.
Category Name
Enter the Category Name.
If the Category will be added as a Custom Field by the Zendesk Action, no changes will be made to the formatting or spaces. For example, if the ChatGPT output of "Order not received" is added as a Custom Field, it will show as "Order not received".
Refer to Configuring the Workflows - Zendesk Action.
Category Description (the Instructions to ChatGPT)
The Category Description tells ChatGPT what to look for in the text. These are the ChatGPT prompts. You enter this using plain English and describe the text you want ChatGPT to find for that Category. Be as specific as you can and you can include as many rules as you need. Also give examples of the text.
If you need additional space, pull the text box down from the bottom right-hand corner.
In the following example, this gives a description of how a customer might be enquiring about where their order is.
Adding More than One Category
When you have added a Category, you can then add more Categories by clicking the + button next to Categories to add the details of another Category. When the Workflow is triggered, this will check for all the Categories that have been configured and apply any or all that are relevant for the text.
Using Categorize to Direct Tickets to a Specific Group or Agent
You can also use the Categorize action to direct tickets to the relevant Groups or Agents based on the Tags generated by ChatGPT. For example, you would use Categorize to add a Tag, and then that Tag can be used as a condition in a separate Trigger to assign the ticket to a given Group or Agent. The additional Trigger needs to be created separately in your Zendesk Admin Center and this is not created by the Triggers+ChatGPT app.
Detect Thank You Only
You can use the Detect Thank You Only action to handle tickets that get reopened, after the ticket has been solved, due to a customer sending a thank you note as a response.
With this Action, for a "Solved" ticket, if a customer sends a thank you note, without any additional information, ChatGPT returns the text "Thank You Only". This is only triggered for "Solved" tickets. This text is then used to set a Tag thank_you_only in the ticket which can be used in a Zendesk Trigger to reset the ticket to "Solved".
To set up the Detect Thank You Only:
- In the configuration, set the OpenAI ChatGPT Action as Detect Thank You Only.
- In the configuration, set the Zendesk Action, checkmark Add Tag to Ticket.
When this triggers, it adds the Tag thank_you_only to the Ticket. - Set up a Zendesk Trigger which uses the thank_you_only Tag to reset the status of the ticket to "Solved".
Translate
The Translate action provides automatic translation of ticket content. You can set up the Workflows so that:
- New inbound messages are automatically translated as they arrive.
- Your Support Agents write their outbound replies in your language which is then translated so the reply is sent in the customer's language.
Translate Direction
For the Translate Direction, select one of the following.
- Inbound - Translate Customer Messages as Internal Notes
- Outbound - Translate Internal Notes as Public Replies
- Custom - Choose the field to translate
The Translate Direction selected controls which fields are then shown in the Workflow configuration for Translate.
Inbound - Translate Customer Messages as Internal Notes and Outbound - Translate Internal Notes as Public Replies set up the configuration for the translation of either Inbound or Outbound message.
Custom - Choose the field to translate allows you to make your own settings.
Translating Inbound Messages
Translation Direction Selected as Inbound - Translate Customer Messages as Internal Notes
To set up the configuration automatically for the Inbound message Workflow, select Inbound - Translate Customer Messages as Internal Notes as the Translate Direction.
This ensure that the translation will be added only as an internal note, for your Agents to see, and will not go out to the customer.
To Language for Inbound Messages
For the To Language, select the language to translate to. This should be the language that your Support Agents work in.
Translating Outbound Messages
Translation Direction Selected as Outbound - Translate Internal Notes as Public Replies
To set up the configuration automatically for the Outbound message Workflow, select Outbound - Translate Internal Notes as Public Replies as the Translate Direction.
For an Outbound Message, the Agent will write the reply as an Internal Note and when the ticket is submitted, this is translated and sent as a Public Reply.
To Language
For the To Language, it is recommended to set this to User Language from Zendesk as the language to translate to. This translates any Internal Notes (added by the Support Agent) to the language of the User who created the ticket.
If required, you can change this to a specific language.
Internal Note starts with
The value in the Internal Note starts with field, determines the hashtag that is used to translate the Outbound text. The default for this is #translate. The value held in the Workflow is also used when the Agent enters a reply and this also needs to be added to the Trigger for the Workflow
- When the Support Agent enters their reply in the Internal Note in the ticket, this hashtag needs to be entered at the start of the text that will be translated. followed by the text to be translated.
- The hashtag needs to be added In the Zendesk Trigger for the Outbound Translations that activates the translation.
- For an example, of how the Agent enters the reply and for details of the changes needed to the Conditions in the Zendesk Trigger, refer to Example Workflow - Translate an Outbound Message.
If you change the default, ensure that this is changed wherever it is used and needed.
Translation Direction Selected as Custom - Choose the field to translate
To set up the configuration yourself for the Translate, select Custom - Choose the field to translate as the Translate Direction.
From Field
Select which field in the ticket you want this Workflow to translate.
To Language
For the To Language, select the language to translate to. This may depend on whether you are setting up the Custom Translate Workflow for Inbound or Outbound messages.
- For Inbound messages, select this as the language that most of your Support Agents work in.
- For Outbound messages, it is recommended to set the language to translate to as User Language from Zendesk. This translates any Internal Notes (added by the Support Agent) to the language of the User who created the ticket, that is, to the language of the customer who created the ticket. For Outbound messages, you can also set this as a specific language.
Extract Value
The Extract Value action extracts a value from a part of the ticket.
Extract Value From
When you select Extract Value, you are prompted to select where to Extract Value From. These are pre-defined and can be selected as one of the following:
- Last Customer Comment
- Last Agent Reply
- Last Reply from Any User
- Last Internal Note.
What to extract
Enter a name so that ChatGPT can recognize what it is. The name has to be descriptive (such as "Order Number") but this does not need to match any of the fields.
Format of the value to extract (the Instructions to ChatGPT)
You then tell ChatGPT what the value looks like, using plain English. These are the ChatGPT prompts. Describe the format of the value and be as specific as you can. You can have as many rules as you need. Also give an example of the value.
In the following example, this is describing what the Order Number looks like and an example is included.
If you need additional space, pull the text box down from the bottom right-hand corner.
Custom Prompt
The Custom Prompt action allows you to add your own prompt. This allows you complete control on the instructions to send to ChatGPT.
Both the User Message and System Instructions are used to create the prompt sent to the ChatGPT API.
Within the Custom Prompt, you use one or more of the following Placeholders to insert ticket data into the prompt. These Placeholder codes tell ChatGPT where to look in the ticket.
{{ticket.last_customer_comment}}
{{ticket.last_agent_comment}}
{{ticket.all_customer_comments}}
{{ticket.all_agent_comments}}
{{ticket.all_comments}}
{{ticket.customer_name}}
{{ticket.last_internal_note}}
{{ticket.all_internal_notes}}
A common use for the placeholders for all comments or notes (ticket.all_customer_comments, ticket.all_agent_comments, ticket.all_comments and ticket.all_internal_notes) would be to summarize an entire conversation once a ticket was solved.
User Message
The User Message provides the instructions to ChatGPT. This provides requests or comments for the ChatGPT to respond to and the User Message is where the actual instruction for ChatGPT should be added.
In the User Message you can use one of the Placeholders to insert ticket data into the prompt.
System Instructions
The System Instructions are optional and these help to set the behavior of the assistant for the ChatGPT action. You can use this to modify the personality of the assistant or provide specific instructions about how it should behave. For example, you could add this as "Act as a professional help desk support agent without adding any unnecessary comments."