Enhance AI-generated bot messages

AI label, citation, feedback buttons, and sensitivity label in your bot’s messages improve user engagement and foster transparency and trust.

• AI label enables users to identify that the message was generated using AI.
• Citation enables users to refer to the source of the bot message through in-text citations and references.
• Feedback buttons enable users to provide positive or negative feedback to the bot messages.
• Sensitivity label enables users to understand the confidentiality of the bot message.

The following screenshots show how bot messages can be enhanced with AI label, citation, feedback buttons, and sensitivity label:

AI label

An AI label on your bot message indicates it’s generated by AI. While AI-powered bots using Large Language Models (LLMs) are generally reliable, the label helps prevent confusion about the information's source.

Add AI label

If you're using Teams AI library to build yout bot, AI label is automatically enabled for all AI-powered bot messages in the ai module within the PredictedSayCommand action. For more information, see AIEntity interface.

If you're using Microsoft Bot Framework SDK to build your bot, include additionalType under the entities array of the message entity object. Following is an example code snippet:

await context.sendActivity({
  type: ActivityTypes.Message,
  text: `Hey! I'm a friendly AI bot. This message is generated by AI.`,
  entities: [
    {
     type: "https://schema.org/Message",
     "@type": "Message",
     "@context": "https://schema.org",
     additionalType: ["AIGeneratedContent"], // Enables AI label
    }
  ]
});

After adding an AI label, your bot’s message shows an "AI generated" label next to the bot’s name. A disclaimer hovers stating, "AI-generated content may be incorrect." The AI label and disclaimer cannot be customized for AI-powered bots.

Error handling

Citations

Citing sources in bot messages helps users ask follow-up questions or conduct independent research. Cite data sources like files, messages, emails, and work items to provide valuable insights. Citations are crucial for bots using techniques like Retrieval-Augmented Generation (RAG).

Citations to your bot message include in-text citation, details of citation reference, and sensitivity label for the content referenced.

• In-text citation denotes the citation numbers added to the bot message in the [#] format, each corresponding to a reference. A citation can be inserted anywhere within the text.
• Details of citation reference include the title, keywords, abstract, hyperlink, and sensitivity information. References appear as pop-up windows for each in-text citation.
• Sensitivity labels to citation indicate the confidentiality of the citation content referenced and aren't added automatically. To add a sensitivity label for a citation, see add sensitivity label.

Add citations

If you're using Teams AI library to build your bot, citations are added to an AI-powered bot message automatically through PredictedSayCommand action. You can also modify the PredictedSayCommand action to add citations to your bot message. For more information, see ClientCitation interface.

If you're using Bot Framework SDK to build your bot, include citation under the entities array. Following is an example code snippet:

await context.sendActivity({
  type: ActivityTypes.Message,
  text: `Hey I'm a friendly AI bot. This message is generated through AI [1]`, // cite with [1],
  entities: [
  {
    type: "https://schema.org/Message",
    "@type": "Message",
    "@context": "https://schema.org",
    citation: [
    {
      "@type": "Claim",
      position: 1, // Required. Must match the [1] in the text above
      appearance: {
        "@type": "DigitalDocument",
        name: "AI bot", // Title
        url: "https://example.com/claim-1", // Hyperlink on the title
        abstract: "Excerpt description", // Appears in the citation pop-up window
        keywords: ["keyword 1", "keyword 2", "keyword 3"], // Appears in the citation pop-up window
        },
      },
    ],
  },
],
})

After you enable citations, the bot message includes in-text citations and references. The in-text citations display the reference details when users hover over the citation.

Error handling

Feedback buttons

Feedback buttons in bot messages are essential for tracking user engagement, identifying errors, and gaining insights into bot performance. These insights enable targeted enhancements of the bot’s conversational capabilities. Enable feedback buttons to allow users to like or dislike messages and provide detailed feedback.

When the user selects a feedback button, a respective feedback form appears based on the user's selection.

Feedback buttons are located at the footer of the bot’s message and include a 👍 (thumbs up) and a 👎 (thumbs down) button that the user to choose. You can collect feedback on bot responses from one-on-one and group chats.

Add feedback buttons

For a bot built using Teams AI library, Teams enables feedback buttons to all bot messages when enable_feedback_loop is set to true in the ai module.

export const app = new Application<ApplicationTurnState>({
    ai: {
        planner: planner,
        enable_feedback_loop: true
    },

For more information, see the const app variable.

After you enable feedback buttons, all SAY commands from the bot have feedbackLoopEnabled automatically set to true in the channelData object.

To enable feedback buttons in a bot built using Bot Framework SDK, add a channelData object in your bot message and set the value of feedbackLoopEnabled to true.

await context.sendActivity({
  type: ActivityTypes.Message,
  text: `Hey! I'm a friendly AI bot!`,
  channelData: {
    feedbackLoopEnabled: true // Enable feedback buttons
  },
});

Handle feedback

The bot receives user input from the feedback form through a bot invoke flow. For bots built using Teams AI library, the bot invoke request is automatically handled. To manage feedback, use the app.feedbackLoop method to register a feedback loop handler when the user provides feedback.

app.feedbackLoop(async (context, state, feedbackLoopData) => {
  // custom logic here...
});

For more information, see the asynchronous callback function.

For a bot built using Bot Framework SDK, you must have an onInvokeActivity handler to process the feedback. Ensure that you return a status code 200 with an empty JSON object as a response.

The following code snippet shows how to handle feedback received in a bot invoke and return a response with the status code 200:

public async onInvokeActivity(context: TurnContext): Promise<InvokeResponse> {
    try {
      switch (context.activity.name) {
        case "message/submitAction":
           console.log('Your feedback is ' + JSON.stringify(context.activity.value))
           // Your feedback is {"actionName":"feedback","actionValue":{"reaction":"like","feedback":"{\"feedbackText\":\"This is my feedback.\"}"}}
           return CreateInvokeResponse(200, {});
        default:
          return {
            status: 200,
            body: `Unknown invoke activity handled as default- ${context.activity.name}`,
          };
      }
    } catch (err) {
      console.log(`Error in onInvokeActivity: ${err}`);
      return {
        status: 500,
        body: `Invoke activity received- ${context.activity.name}`,
      };
    }
  }

 export const CreateInvokeResponse = (
  status: number,
  body?: unknown
 ): InvokeResponse => {
    return { status, body };
 };

Store the feedback by saving message IDs and content of messages your bot sends and receives. When your bot gets an invoke request with feedback, match the message ID with the corresponding feedback.

If a user uninstalls your bot and still has access to the bot chat, Teams removes the feedback buttons from the bot messages to prevent the user from providing feedback to the bot.

Error handling

Sensitivity label

Bot responses might contain confidential information or be accessible only to certain individuals within the organization. Add a sensitivity label to help users identify the confidentiality of a message, enabling them to exercise caution when sharing it.

Add sensitivity label

For bots built using Teams AI library, sensitivity label can be added through PredictedSayCommand action. For more information, see SensitivityUsageInfo interface.

For bots built using Bot Framework SDK, add a sensitivity label to your bot message by modifying the message to include usageInfo in the entities object.

The following code snippet shows how to add sensitivity labels to both bot messages and citation reference:

await context.sendActivity({
  type: ActivityTypes.Message,
  text: `Hey, I'm a friendly AI bot. This message is generated through AI [1]`,
  entities: [
    {
      type: "https://schema.org/Message",
      "@type": "Message",
      "@context": "https://schema.org",
      usageInfo: {
        "@type": "CreativeWork",
        name: "Sensitivity title",
        description: "Sensitivity description",
      },
    },
  ],
});

await context.sendActivity({
  type: ActivityTypes.Message,
  text: `Hey, I'm a friendly AI bot. This message is generated through AI [1]`,
  entities: [
    {
      type: "https://schema.org/Message",
      "@type": "Message",
      "@context": "https://schema.org",
      usageInfo: {
        "@type": "CreativeWork",
        name: "Sensitivity title",
        description: "Sensitivity description",
      },
      citation: [
        {
          "@type": "Claim",
          position: 1,
          appearance: {
            "@type": "DigitalDocument",
            name: "AI bot",
            url: "https://example.com/claim-1",
            usageInfo: {
              "@type": "CreativeWork",
              "@id": "sensitivity1",
              name: "Sensitivity title",
              description: "Sensitivity description",
            },
          },
        },
      ],
    },
  ],
});

After adding the sensitivity label, your bot message displays a shield icon. Users can hover over the icon to see a disclaimer about the message's sensitivity.

Error handling

Modify PredictedSayCommand

For a bot built using Teams AI library, the PredictedSayCommand provides control on how AI label, citation, feedback button, and sensitivity label are added to the bot's activity. Following is the code snippet to modify PredictedSayCommand:

app.ai.action<PredictedSayCommand>(AI.SayCommandActionName, async (context, state, data, action) => {
  // custom logic here...
  await context.sendActivity(data.content);
  return "";
});

For more information about PredictedSayCommand, see PredictedSayCommand interface.

Code sample

See also

• Bot activity handlers
• Format your bot messages
• Get started with Teams AI library
• Streaming UX in bots