AI Actions transform your Social Intents chatbot from a simple question-and-answer tool into an automation engine that takes real action during conversations. Instead of only responding with text, your chatbot can book meetings, capture leads in your CRM, call external APIs, route conversations to specific teams, display interactive buttons, and close chats automatically - all without human intervention.
Below you will find what AI Actions are, how they work under the hood, the different types available, and how to get started building your first action. Whether you want to connect your chatbot to Calendly, push leads into HubSpot, or build a custom API integration, AI Actions make it possible in minutes.
What Is an AI Action?
An AI Action is an automated workflow attached to your chat widget. You define what the action does, when it should trigger, and what data it needs. During a conversation, when your AI chatbot detects that the visitor's request matches an action's trigger conditions, it executes the action automatically. The visitor sees the result - a booked meeting, a confirmation message, a button to click, or a direct handoff to a human agent - without ever leaving the chat window.
Each action is configured through the Social Intents dashboard. You do not need to write code, although advanced users can build custom API integrations. The system handles everything: detecting when to fire the action, collecting required data from the conversation, executing the workflow, and presenting results back to the visitor.
How AI Actions Differ from Chatbot Responses
A standard chatbot response is pure text - the AI reads the visitor's message, checks its training data, and generates a written answer. An AI Action goes further. It performs a real operation: sending data to an external system, embedding a scheduling widget, routing the conversation to a different channel, or displaying interactive elements. Think of chatbot responses as the AI talking, and AI Actions as the AI doing.
For example, if a visitor asks "Can I schedule a demo?", a standard chatbot might respond with "Sure, you can schedule a demo at calendly.com/yourcompany." With an AI Action, the chatbot instead triggers a Calendly action that embeds the scheduling widget directly in the chat. The visitor books the meeting without opening a new tab or leaving the conversation.
Como funcionam as ações de IA
Every AI Action has three core components that determine its behavior:
1. The Trigger (When to Use)
The trigger is a natural language description that tells the AI engine when this action should fire. You write it in plain English - there are no complex rule builders or decision trees. The AI reads the conversation context and compares it against your trigger description to decide whether the action is relevant.
For example, a trigger might say: "Use this action when the visitor wants to book a meeting, schedule a demo, or set up a consultation call." The AI engine evaluates each visitor message against this description and fires the action when it detects a match.
Good triggers are specific and descriptive. Instead of writing "when someone asks about meetings," write "when the visitor explicitly requests to schedule, book, or arrange a meeting, demo, consultation, or call." The more context you provide, the more accurately the AI matches visitor intent.
2. The Action (What Happens)
The action is the operation that runs when the trigger matches. This could be an API call to an external system, an embedded Calendly widget, a routing command that sends the chat to a specific Slack channel or Teams channel, a button that links to a page, or a command that closes the chat session. The action type determines what configuration fields are available.
3. The Response (What the Visitor Sees)
The triggered response is the message the chatbot sends to the visitor when the action fires. This provides context for what just happened. For example, "Great, I've created a support ticket for you. Your ticket number is #12345." or "Let me connect you with our sales team right away." You configure this in the Triggered response field.
Action Types Overview
Social Intents provides nine distinct action types. Each serves a different automation purpose. Here is a summary of all available types:
| Action Type | Label | What It Does |
|---|---|---|
| Call API Request | api | Sends an HTTP request (GET, POST, PUT, DELETE) to any external API endpoint. Ideal for creating records, querying databases, triggering webhooks, and integrating with third-party tools. |
| Escalate Chat to Human | escalate_route | Routes the conversation to a specific channel or space in your connected platform (Slack channel, Teams channel, Google Chat space, or Webex room). Includes optional confirmation and context instructions for agents. |
| Show Button or Iframe Only | button | Displays a clickable button, link, or embedded iframe in the chat. Use this for Calendly widgets, payment forms, external links, or any embeddable content. |
| Auto Trigger on Chat Start | auto | Fires automatically when a new chat session begins. Useful for greeting workflows, initial data collection, or pre-chat surveys. |
| Auto Trigger on Chat Escalation | escalate | Fires automatically when the chatbot escalates to a human agent. Use this to capture lead data or notify external systems when escalation occurs. |
| Auto Trigger on Chat End | autoend | Fires automatically when a chat session ends. Ideal for post-chat surveys, satisfaction scores, or syncing conversation data to external systems. |
| Auto Trigger on Every Message | autoall | Fires on every single message in the conversation. Use sparingly - this is meant for real-time logging, sentiment analysis, or continuous data enrichment. |
| Agent Manually Triggered | agent | Only fires when a human agent manually initiates it during a chat. The agent clicks the action from their chat interface. Useful for agent-assisted workflows like discount offers or account lookups. |
| End the Chat | endchat | Closes the chat session with an optional farewell message. The AI triggers this when the conversation appears complete. |
Each action type is covered in detail in its own article. For a side-by-side comparison to help you choose the right type, see Choosing the Right Action Type.
Creating Your First AI Action
Follow these steps to create an AI Action from your Social Intents dashboard:
Navigate to AI Actions
Log in to socialintents.com and go to My Widgets. Select the widget you want to add the action to. Click the AI Actions tab in the left sidebar, then click Add Action.
Choose an Action Type
Select the action type from the dropdown. This determines what fields appear in the configuration form. Start with Call API Request for a custom integration, Escalate Chat to Human for routing, or Show Button or Iframe Only for buttons and embedded content.
Name Your Action
Enter a descriptive action name. This name helps the AI identify the action, so make it clear and specific. Examples: Route_To_Sales, Book_Demo_Meeting, Create_HubSpot_Lead. Spaces and hyphens are automatically converted to underscores.
Write the Trigger Prompt
In the When to use field, describe in natural language when this action should fire. Be specific about visitor intent, keywords, and scenarios. The AI engine reads this description to decide whether to trigger the action during a conversation.
Set the Triggered Response
In the Triggered response field, enter the message the chatbot sends to the visitor when the action fires. This provides context about what happened - "I've scheduled your demo" or "Let me connect you with our billing team."
Configure Action-Specific Settings
Depending on your action type, fill in the relevant fields: API endpoint URL, target channel for routing, Calendly URL for meeting booking, button labels and URLs, or any other type-specific settings. Each action type article covers these fields in detail.
Set Optional Behaviors
Configure optional settings like Trigger only once per session (prevents the action from firing multiple times), Show action button in response (displays a clickable button), or Automatically close the chat after the action runs.
Save and Test
Click Save Action. Open your chat widget in a browser and test the action by typing a message that matches your trigger description. Verify that the action fires correctly and the visitor sees the expected response.
Action Configuration Fields Reference
Every AI Action shares a set of common fields. Here is a reference for the fields you will see across all action types:
| Field | Required | Description |
|---|---|---|
| Action Type | Sim | The type of action. Determines which additional fields appear. |
| Action Name | Sim | A descriptive identifier. Must be alphanumeric with underscores. Used by the AI engine to identify the action. |
| When to use | Recommended | Natural language description of when the AI should trigger this action. The more specific, the better. |
| Triggered response | Recommended | The message sent to the visitor when the action fires. |
| Trigger only once per session | Não | When checked, the action fires only once per chat session, even if the trigger matches again. |
| Show action button in response | Não | Displays a clickable button in the chat. Requires Button Label and Button URL fields. |
| Automatically close the chat | Não | Closes the chat session after the action runs. You can include a farewell message. |
Data Collection: How the AI Gathers Parameters
Many actions require data from the visitor - an email address for a CRM lead, an order number for a status lookup, or a name for a meeting booking. Social Intents handles this through the Collect data inputs configuration.
When you add parameters to an action, you define the parameter name, data type (Text, Number, or Date), a description of what the parameter represents, an optional default value, whether it accepts an array of values, and whether it is required. The AI agent uses this information to collect the data conversationally. If the visitor has already mentioned their email in the chat, the AI extracts it from the conversation history. If not, the AI asks the visitor for the missing information before executing the action.
For example, a CRM lead capture action might have these parameters:
| Nome | Type | Description | Required |
|---|---|---|---|
| Text | The visitor's email address | Sim | |
| first_name | Text | The visitor's first name | Sim |
| sobrenome | Text | The visitor's last name | Não |
| empresa | Text | The visitor's company or organization | Não |
The AI reads the conversation, finds any data that has already been shared, and prompts for anything that is still missing and marked as required. This creates a natural, conversational data collection flow rather than a rigid form experience.
Multiple Actions on One Widget
You can attach multiple AI Actions to a single chat widget. The AI engine evaluates all available actions during each conversation turn and selects the most appropriate one based on the visitor's current intent. For example, you might have a Calendly booking action, a CRM lead capture action, an escalation routing action for billing questions, and a separate routing action for technical support - all on the same widget.
There is no hard limit on the number of actions per widget, but we recommend keeping the list focused. If you have more than ten actions, the AI engine has more trigger descriptions to evaluate, which can occasionally lead to less precise matching. Group related functionality into a single action when possible, and write clear, distinct trigger descriptions so the AI can differentiate between similar actions.
Testing and Debugging Actions
After creating an action, test it thoroughly before going live with real visitors. Here are the steps we recommend:
- Open your chat widget on your website or use the preview mode in the Social Intents dashboard.
- Type a message that matches your action's trigger description. Use the exact kind of language a real visitor would use.
- Verify the triggered response appears in the chat. Confirm the message text is correct.
- Check the action result - if it is an API call, verify the external system received the request. If it is escalation routing, verify the message appeared in the correct channel. If it is a button or iframe, verify the element renders correctly.
- Test edge cases - try messages that should not trigger the action and confirm it stays quiet. Try variations of the trigger phrase to make sure the AI matches reliably.
- Test data collection - if your action has required parameters, start a chat without providing the data upfront. Verify the AI asks for the missing fields before executing the action.
Best Practices
- Write descriptive action names - Names like
Create_Support_TicketeBook_Sales_Demoare far better thanAction_1ouTest. The AI uses the name as additional context for matching. - Be specific in trigger descriptions - Vague triggers like "when someone needs help" will fire too often. Specific triggers like "when the visitor asks to create a support ticket, report a bug, or file an issue" are much more reliable.
- Use triggered responses to set expectations - Tell the visitor what happened. "I've booked your demo for Tuesday at 2pm" is better than silence after an action fires.
- Enable trigger-once for one-time workflows - If your action should only fire once per chat (like lead capture), check the Trigger only once per session box to prevent duplicate executions.
- Test before going live - Always test new actions with sample conversations before exposing them to real visitors. Verify both positive matches (should trigger) and negative matches (should not trigger).
- Keep each action focused - One action should do one thing. Do not try to combine meeting booking, lead capture, and escalation routing into a single action. Use separate actions with clear triggers for each workflow.
What to Read Next
Now that you understand what AI Actions are and how they work, dive into the specific action types:
- Choosing the Right Action Type - Compare all nine types side by side.
- Escalation Routing - Route chats to specific teams and channels.
- Booking Meetings with Calendly - Embed calendar scheduling in chat.
- Capturing Leads to HubSpot, Salesforce & Dynamics 365 - Sync leads to your CRM automatically.
- Custom API Actions - Call any REST API from your chatbot.
- Showing Buttons, Links & Embedded Content - Display interactive elements in chat.
Perguntas frequentes
Do I need coding skills to set up AI Actions?
No. Most action types - escalation routing, Calendly booking, buttons, and CRM integrations with pre-built templates - require zero coding. Custom API actions require basic understanding of REST APIs (HTTP methods, JSON, and headers), but the Social Intents interface handles the configuration through a visual form rather than code.
How many actions can I add to one widget?
There is no hard limit. However, we recommend keeping the number of actions focused and manageable - typically under ten per widget. Each action adds a trigger description the AI must evaluate, so a more focused set of actions produces more reliable matching.
Can one conversation trigger multiple actions?
Yes. Different actions can fire at different points in a single conversation. A visitor might start by asking a product question (chatbot responds with text), then request a demo (Calendly action fires), and then ask to speak with sales (escalation routing fires). Each action fires independently when its trigger condition matches.
Do AI Actions work with all AI engines?
Yes. AI Actions work with ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google). The action system is independent of the AI engine - the engine handles natural language understanding and trigger matching, while the action system handles execution.
What happens if no agents are online when an escalation action fires?
Escalation routing sends the conversation to the specified channel or space in your connected platform. If no agents are actively monitoring that channel, the message will be waiting for them when they return. The visitor sees a message indicating they are being connected. You can customize the triggered response to set appropriate expectations - for example, "I'm connecting you with our sales team. If they're not available right now, they'll respond as soon as possible."
Can I disable an action without deleting it?
Currently, you can remove actions from a widget by deleting them. If you want to temporarily disable an action, you can change the When to use field to something that will never match (like "DO NOT USE - DISABLED") as a workaround until a formal disable toggle is available.