This page provides various examples of how you can configure your eesel AI LLM using prompts to perform actions with integrated Actions, illustrating different types of actions and workflows relevant to support and e-commerce teams.
These prompts go in the Customize page of the eesel AI dashboard.
Beyond action usage, you can also define the AI's personality, response length, and referencing style directly in its configuration prompt. For instance, you might start your prompt with instructions like:
You are a support representative for Anytime Fitness Australia. Maintain a supportive, conversational and understanding tone. Your responses must be concise, ideally a few sentences. Provide a list of sources as inline links throughout your response where available. Sign off as "Anytime Fitness Support" at the end of each interaction. Always detect the user's language and respond in that language. When appropriate, format your final response as an email.
The examples below focus specifically on instructing the AI on when and how to use Actions or perform specific conditional actions, grouped by their primary function or domain.
For the full documentation for each available action, please consult the Action documentation.
These examples focus on configurations that primarily control the AI's conversational output, how it gathers necessary information from the user, manages the status or metadata of the conversation/ticket itself, or guides the user's interaction flow.
Scenario: The AI needs a specific piece of information from the user (e.g., an order ID, an email address, details about an issue, a specific product name) before it can effectively use a action or provide a relevant answer.
AI Action: Identify missing necessary information -> Formulate a clear question to the user asking for the specific information needed.
Actions Involved: None directly in this step, but gathering the information is a prerequisite for a subsequent action call.
Example Prompt Instruction:
If a user asks about an order or their account but does not provide an order number or email address: You must politely ask for the specific information needed. For example: "I can help with that! Could you please provide your order number?" or "To look up your account details, could you please confirm the email address associated with your account?" Do not attempt to call a action that requires this information until the user provides it.
Scenario: The user explicitly indicates they are satisfied, or the conversation reaches a point where it should be closed according to your workflow.
AI Action: Identify user satisfaction or conversation completion -> Call a closing action.
Actions Involved: zendesk_close_ticket (or intercom_close_conversation).
Example Prompt Instruction:
If the user explicitly states their problem is solved, they are satisfied with the outcome, or if the conversation has concluded and no further action is required after you've provided information, you must confirm this understanding with the user. If they agree, call the zendesk_close_ticket action to mark the ticket as closed and thank the user.