AI-Powered Message Drafts

The AI-Powered Message Drafts feature allows live chat agents to generate AI-suggested responses with a single click. Using OpenAI’s Assistant API, it analyzes the full conversation context and produces a professional draft that the agent can review, edit, and send.

Prerequisites

Before you can use AI-powered message drafts, you need:

  1. An OpenAI integration configured in your organization (see Integrate Quriobot with OpenAI)
  2. An OpenAI Assistant created in your OpenAI platform
  3. A subscription plan that supports this feature

Step 1: Configure OpenAI Integration

If you haven’t already set up an OpenAI integration:

  1. Go to Organization SettingsIntegrations
  2. Add a new OpenAI integration with your API key
  3. Save the integration

For detailed instructions, see Integrate Quriobot with OpenAI.

Step 2: Create an OpenAI Assistant

You need an OpenAI Assistant that will generate the draft responses:

  1. Go to the OpenAI Assistants page
  2. Click Create to create a new assistant
  3. Give it a name (e.g., “Live Chat Draft Assistant”)
  4. Configure its instructions to match your support style and tone
  5. Select a model (e.g., GPT-4o)
  6. Save the assistant and note its ID

Step 3: Enable AI Drafts in Live Chat Settings

  1. Go to Organization SettingsLive Chat tab
  2. Scroll down to the AI Settings section
  3. Select your OpenAI integration from the “AI Integration” dropdown
  4. Select your OpenAI Assistant from the “AI Assistant” dropdown
  5. Enable the “Enable AI drafts” toggle
  6. Configure the Prompt Template (see below)
  7. Save your settings
AI Settings in Live Chat

Step 4: Configure the Prompt Template

The prompt template tells the AI how to generate draft responses. When AI drafts are enabled, a Prompt Template code editor appears where you can customize the instructions with markdown syntax highlighting.

The default template is:

You are a helpful customer support agent. Based on the conversation context, suggest a professional and empathetic response. If a current draft is provided below, review and improve it while keeping the original intent. Otherwise, create a new response. Current draft: {{currentDraft}} Additional instructions: {{agentInstructions}}

You can customize this to match your brand voice and support guidelines. For example:

  • Specify the tone (formal, friendly, casual)
  • Include specific product knowledge or FAQ references
  • Set language preferences
  • Define response format (short answers, detailed explanations, etc.)

Template Variables

The prompt template supports variables that are replaced at generation time:

  • {{currentDraft}} — the agent’s current draft text from the editor (used in Edit mode)
  • {{agentInstructions}} — instructions provided by the agent at generation time

You can insert these variables using the variable dropdown button next to the prompt template editor. A reset-to-default button also appears when the template has been modified.

Step 5: Using AI Drafts in Live Chat

Once configured, agents can use AI drafts while responding to live chats:

  1. Open a live chat conversation in the Control Room
  2. In the message composer toolbar, click the magic wand icon (✨)
  3. A compact popover appears with an optional instructions text field and two buttons:
    • Edit — AI reviews and improves your current draft text (only enabled when the editor has content)
    • Create — AI generates a fresh response from the conversation
  • Optionally type instructions to guide the AI (e.g. “be more formal”, “mention our refund policy”)
  • Click Edit or Create and wait for the AI to generate the draft
  • The draft appears in the message editor — review, edit, and send
AI Draft Popover with Edit and Create
Live Chat Message Composer

How It Works

When you click the magic wand button and choose Create or Edit:

  1. The full chat transcript is collected and converted to a readable format
  2. Template variables are substituted: {{currentDraft}} with your current editor text (Edit mode) and {{agentInstructions}} with any instructions you provided
  3. The transcript and processed prompt template are sent to the configured OpenAI Assistant
  4. The AI analyzes the conversation context and generates a suggested response
  5. The draft is inserted into your message editor, ready for review

The AI considers the entire conversation history, including both visitor messages and previous agent responses, to generate contextually appropriate suggestions.

Tips for Best Results

  • Customize your prompt template to match your support style — the more specific, the better the drafts
  • Always review and edit AI-generated drafts before sending — they are suggestions, not final responses
  • Train your OpenAI Assistant with specific instructions about your products, services, and common customer queries
  • Use file search in your OpenAI Assistant to give it access to your knowledge base documents for more accurate responses

Edit this page