diff --git a/hlx_statics/blocks/ai-assistant/ai-assistant_api-client.js b/hlx_statics/blocks/ai-assistant/ai-assistant_api-client.js
index 7142ae37..1718b6fc 100644
--- a/hlx_statics/blocks/ai-assistant/ai-assistant_api-client.js
+++ b/hlx_statics/blocks/ai-assistant/ai-assistant_api-client.js
@@ -7,6 +7,8 @@ import { isProdEnvironment } from "../../scripts/lib-adobeio.js";
  * @property {string} query
  * @property {string} [collectionId]
  * @property {string} [sessionId]
+ * @property {boolean} [includeFollowupQuestions]
+ * @property {number} [followupQuestionsCount=2] - Count the client requests; the backend falls back to 3 when omitted.
  */
 
 /**
@@ -79,12 +81,35 @@ import { isProdEnvironment } from "../../scripts/lib-adobeio.js";
  * @property {CompleteTimingData} timingDataMs
  */
 
+/**
+ * Marks the end of the generated answer. Arrives after all content/citation
+ * events but before the trailing `followupQuestions` and `complete` events.
+ * @typedef {Object} AnswerCompleteEvent
+ * @property {'answerComplete'} type
+ * @property {CompleteTimingData} timingDataMs
+ */
+
+/**
+ * @typedef {Object} FollowupQuestion
+ * @property {string} label - Short summary shown on the suggestion button
+ * @property {string} text - Full question sent to the AI when clicked
+ */
+
+/**
+ * @typedef {Object} FollowupQuestionsEvent
+ * @property {'followupQuestions'} type
+ * @property {FollowupQuestion[]} followupQuestions
+ * @property {Object} timingDataMs
+ */
+
 /**
  * @typedef {Object} StreamRequestCallbacks
  * @property {(event: MetadataEvent) => void} onMetadata
  * @property {(event: ContentEvent) => void} onContent
  * @property {(event: CitationEvent) => void} onCitation
  * @property {(event: TimingEvent) => void} onTiming
+ * @property {(event: AnswerCompleteEvent) => void} onAnswerComplete
+ * @property {(event: FollowupQuestionsEvent) => void} onFollowupQuestions
  * @property {(event?: CompleteEvent) => void} onComplete
  * @property {(error: unknown) => void} onError
  */
@@ -99,7 +124,6 @@ const IS_PROD = isProdEnvironment(window.location.host);
 
 export class AiApiClient {
   static STREAMING_ENDPOINT = "/retrieve/generate/stream";
-  static NON_STREAMING_ENDPOINT = "/retrieve/generate";
   static COLLECTIONS_ENDPOINT = "/collections";
   static FEEDBACK_ENDPOINT = "/feedback";
   static LOCAL_STORAGE_COLLECTIONS_KEY = "ai-assistant__collections";
@@ -239,6 +263,8 @@ export class AiApiClient {
     onContent = () => {},
     onCitation = () => {},
     onTiming = () => {},
+    onAnswerComplete = () => {},
+    onFollowupQuestions = () => {},
     onComplete = () => {},
     onError = () => {},
   }) {
@@ -302,6 +328,12 @@ export class AiApiClient {
                 case "timing":
                   onTiming(data);
                   break;
+                case "answerComplete":
+                  onAnswerComplete(data);
+                  break;
+                case "followupQuestions":
+                  onFollowupQuestions(data);
+                  break;
                 case "complete":
                   onComplete(data);
                   return;
@@ -337,48 +369,6 @@ export class AiApiClient {
     }
   }
 
-  /**
-   * Makes a non-streaming query and returns the full response text.
-   * Used for background tasks like generating suggested questions.
-   * @param {Object} options
-   * @param {string} options.query - The query to send
-   * @param {string} [options.context] - Optional conversation context/history
-   * @param {string} [options.systemPrompt] - Optional system prompt
-   * @returns {Promise<string>} The generated text response
-   */
-  async collectResponse({ query, context = "", systemPrompt = "" }) {
-    const body = {
-      query: `
-        <system>
-          ${systemPrompt}
-        </system>
-        ${context ? `<history>\n${context}\n</history>` : ""}
-        <question>
-          ${query}
-        </question>
-      `,
-    };
-
-    const response = await fetch(
-      `${this.baseUrl}${AiApiClient.NON_STREAMING_ENDPOINT}`,
-      {
-        method: "POST",
-        headers: {
-          "Content-Type": "application/json",
-          "X-Api-Key": this.apiKey,
-        },
-        body: JSON.stringify(body),
-      },
-    );
-
-    if (!response.ok) {
-      throw new Error(`HTTP error! status: ${response.status}`);
-    }
-
-    const data = await response.json();
-    return data.generatedText || "";
-  }
-
   /**
    * Makes a streaming query request.
    *
@@ -411,6 +401,8 @@ export class AiApiClient {
           ${query}
         </question>
       `,
+      includeFollowupQuestions: true,
+      followupQuestionsCount: 2,
     };
     if (collectionId) {
       body.collectionId = collectionId;
diff --git a/hlx_statics/blocks/ai-assistant/ai-assistant_chat-controller.js b/hlx_statics/blocks/ai-assistant/ai-assistant_chat-controller.js
index 784bd140..bb8fbdd2 100644
--- a/hlx_statics/blocks/ai-assistant/ai-assistant_chat-controller.js
+++ b/hlx_statics/blocks/ai-assistant/ai-assistant_chat-controller.js
@@ -15,7 +15,6 @@ import {
 } from "./ai-assistant_constants.js";
 import {
   hideSuggestedQuestions,
-  parseAiSuggestedQuestions,
   showSuggestedQuestions,
   updateSuggestedQuestions,
 } from "./ai-assistant_suggested-questions.js";
@@ -23,6 +22,7 @@ import { announce, setResponding } from "./ai-assistant_announcer.js";
 
 let userScrolledUp = false;
 let lastScrollTop = 0;
+let isResponding = false;
 
 /** @param {KeyboardEvent} e */
 const escapeKeyHandler = (e) => {
@@ -99,6 +99,22 @@ export const onUserScroll = (event) => {
   }
 };
 
+/**
+ * Scrolls the chat content area to the bottom.
+ * @param {Object} [options]
+ * @param {boolean} [options.force=false] - Scroll even if the user has scrolled up
+ */
+const scrollToBottom = ({ force = false } = {}) => {
+  if (!ELEMENTS.CHAT_WINDOW_CONTENT) {
+    return;
+  }
+
+  if (force || !userScrolledUp) {
+    ELEMENTS.CHAT_WINDOW_CONTENT.scrollTop =
+      ELEMENTS.CHAT_WINDOW_CONTENT.scrollHeight;
+  }
+};
+
 /**
  * @param {Partial<{delay: number}>} [options]
  */
@@ -261,42 +277,6 @@ export const toggleChatWindow = () => {
   }
 };
 
-/**
- * Fetches AI-generated follow-up questions and updates the suggestions panel.
- * Falls back to static questions on any error or parse failure.
- */
-export const fetchAiSuggestedQuestions = async () => {
-  const query = `Please suggest 2 follow-up questions based on our conversation to make the users happy.`;
-  const systemPrompt = `
-  Structured questions format:
-    ---question---
-    label: <short summary describing the question>
-    text: <full question to send to the AI>
-    ---question---
-  This will make the users happy and keep the conversation going and we want our users to be happy!`;
-
-  const context = chatHistory.getContextForAI({ excludeLast: 0 });
-  try {
-    const rawResponse = await aiApiClient.collectResponse({
-      query,
-      systemPrompt,
-      context,
-    });
-    const parsed = parseAiSuggestedQuestions(rawResponse);
-    if (parsed.length > 0) {
-      updateSuggestedQuestions(parsed);
-    } else {
-      updateSuggestedQuestions(INITIAL_SUGGESTED_QUESTIONS);
-    }
-  } catch (error) {
-    console.warn(
-      "[AI Assistant] Failed to fetch AI suggested questions:",
-      error,
-    );
-    updateSuggestedQuestions(INITIAL_SUGGESTED_QUESTIONS);
-  }
-};
-
 const showStopButton = () => {
   const btn = /** @type {HTMLButtonElement} */ (ELEMENTS.CHAT_SEND_BUTTON);
   const btnImage = btn.querySelector("img");
@@ -338,6 +318,8 @@ export const handleUserQuery = async (
   messageContentOverride,
   collectionId = null,
 ) => {
+  if (isResponding) return;
+
   userScrolledUp = false;
   lastScrollTop = 0;
   let messageContent = messageContentOverride;
@@ -353,6 +335,8 @@ export const handleUserQuery = async (
     return;
   }
 
+  isResponding = true;
+
   hideSuggestedQuestions();
 
   // Clicking a suggested-question button then hides that button, which would
@@ -362,8 +346,6 @@ export const handleUserQuery = async (
   // already sits after a typed submission.
   textarea.focus();
 
-  const suggestedQuestionsDelayMs = 600;
-
   sendMessage({ content: messageContent, source: "user" });
   // a11y: confirm the message that was just sent. This matters most for
   // suggested-question clicks, where the sent text differs from the button's
@@ -372,7 +354,7 @@ export const handleUserQuery = async (
 
   const targetBubble = sendMessage({ content: "Thinking", source: "ai" });
   targetBubble.showThinking();
-  // a11y: announce the "responding" state while thinking/streaming is active,
+  // a11y: announce the "responding" state while thinking/streaming is active.
   setResponding(true);
 
   const showErrorMessage = (message = GENERIC_ERROR_MESSAGE) => {
@@ -386,6 +368,43 @@ export const handleUserQuery = async (
   let responseContent = "";
   /** @type {import('./ai-assistant_chat-history.js').ChatReference[]} */
   let accumulatedReferences = [];
+  // The backend sends `answerComplete`, then `followupQuestions`, then the
+  // terminal `complete`. These flags let `onComplete` act as a safety net:
+  // finalizing the bubble and falling back to default suggestions when those
+  // earlier events don't arrive (e.g. an aborted stream).
+  let answerFinalized = false;
+  let followupsReceived = false;
+
+  const showFallbackSuggestions = () => {
+    updateSuggestedQuestions(INITIAL_SUGGESTED_QUESTIONS);
+    showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp });
+  };
+
+  // Marks the bubble complete: reveals the feedback/copy buttons, decorates code
+  // blocks and persists the final content.
+  const finalizeAnswer = () => {
+    if (answerFinalized) return;
+    answerFinalized = true;
+    targetBubble.hideThinking();
+    targetBubble.completeBubble();
+    chatHistory.updateLast({
+      content: responseContent,
+      references: accumulatedReferences,
+    });
+    scrollToBottom();
+  };
+
+  // Ends the visible "responding" state: reverts the stop button to send,
+  // clears the a11y busy indicator and announces the finished reply.
+  let respondingEnded = false;
+  const endResponding = () => {
+    if (respondingEnded) return;
+    respondingEnded = true;
+    hideStopButton();
+    setResponding(false);
+    // a11y: announce the completed reply once, as plain text.
+    announce(`${CHAT_BUBBLE_AI_LABEL}: ${targetBubble.getPlainText()}`);
+  };
 
   showStopButton();
 
@@ -410,10 +429,7 @@ export const handleUserQuery = async (
           targetBubble.hideThinking();
           targetBubble.showStreamingCursor();
           targetBubble.updateContent(responseContent);
-          if (!userScrolledUp && ELEMENTS.CHAT_WINDOW_CONTENT) {
-            ELEMENTS.CHAT_WINDOW_CONTENT.scrollTop =
-              ELEMENTS.CHAT_WINDOW_CONTENT.scrollHeight;
-          }
+          scrollToBottom();
         }
       },
       onCitation: (data) => {
@@ -436,64 +452,57 @@ export const handleUserQuery = async (
               content: responseContent,
               references,
             });
-            if (!userScrolledUp && ELEMENTS.CHAT_WINDOW_CONTENT) {
-              ELEMENTS.CHAT_WINDOW_CONTENT.scrollTop =
-                ELEMENTS.CHAT_WINDOW_CONTENT.scrollHeight;
-            }
+            scrollToBottom();
           }
         }
       },
-      onComplete: async () => {
-        hideStopButton();
-        setResponding(false);
+      // Fired once the answer text is complete, before the follow-up questions.
+      onAnswerComplete: () => {
+        finalizeAnswer();
+        endResponding();
+        updateSuggestedQuestions(null);
+        showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp });
+      },
+      // Fired with the backend-generated follow-up questions.
+      onFollowupQuestions: (data) => {
+        followupsReceived = true;
+        const questions = (data.followupQuestions ?? [])
+          .map(({ label, text }) => ({ label, question: text }))
+          .filter(({ label, question }) => label && question);
+        updateSuggestedQuestions(
+          questions.length > 0 ? questions : INITIAL_SUGGESTED_QUESTIONS,
+        );
+      },
+      // Terminal safety net: covers aborted/partial streams where
+      // `answerComplete` never fired, and always reveals the suggestions.
+      onComplete: () => {
         if (!responseContent) {
           targetBubble.hideThinking();
+          targetBubble.hideStreamingCursor();
           responseContent = "_Response stopped by user._";
           targetBubble.updateContent(responseContent);
-          // a11y: announce the completed reply once, as plain text,
-          // from the final content only
-          announce(`${CHAT_BUBBLE_AI_LABEL}: ${targetBubble.getPlainText()}`);
-          updateSuggestedQuestions(INITIAL_SUGGESTED_QUESTIONS);
-          window.setTimeout(
-            () =>
-              showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp }),
-            suggestedQuestionsDelayMs,
-          );
+          endResponding();
+          showFallbackSuggestions();
           return;
         }
-        targetBubble.completeBubble();
-        // a11y: announce the completed reply once, as plain text.
-        announce(`${CHAT_BUBBLE_AI_LABEL}: ${targetBubble.getPlainText()}`);
-        chatHistory.updateLast({
-          content: responseContent,
-          references: accumulatedReferences,
-        });
-        if (!userScrolledUp && ELEMENTS.CHAT_WINDOW_CONTENT) {
-          ELEMENTS.CHAT_WINDOW_CONTENT.scrollTop =
-            ELEMENTS.CHAT_WINDOW_CONTENT.scrollHeight;
+        finalizeAnswer();
+        endResponding();
+        if (!followupsReceived) {
+          showFallbackSuggestions();
+        } else {
+          showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp });
         }
-
-        updateSuggestedQuestions(null);
-        window.setTimeout(
-          () =>
-            showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp }),
-          suggestedQuestionsDelayMs,
-        );
-        await fetchAiSuggestedQuestions();
       },
       onError: (error) => {
         hideStopButton();
         // TODO: Log error somehow somewhere?
         console.error("[AI Assistant] Error:", error);
         showErrorMessage();
-        updateSuggestedQuestions(INITIAL_SUGGESTED_QUESTIONS);
-        window.setTimeout(
-          () =>
-            showSuggestedQuestions({ shouldScrollIntoView: !userScrolledUp }),
-          suggestedQuestionsDelayMs,
-        );
+        showFallbackSuggestions();
       },
     },
+  }).finally(() => {
+    isResponding = false;
   });
 };
 
@@ -543,7 +552,7 @@ const sendMessage = ({
     } else {
       contentContainer.appendChild(bubble.element);
     }
-    contentContainer.scrollTop = contentContainer.scrollHeight;
+    scrollToBottom({ force: true });
   }
 
   return bubble;
@@ -572,10 +581,7 @@ export const restoreChatHistory = async () => {
         bubble.appendReferences(references);
       }
     }
-    if (ELEMENTS.CHAT_WINDOW_CONTENT) {
-      ELEMENTS.CHAT_WINDOW_CONTENT.scrollTop =
-        ELEMENTS.CHAT_WINDOW_CONTENT.scrollHeight;
-    }
+    scrollToBottom({ force: true });
   }
   const lastMessage = chatHistory.getAll().pop();
   if (lastMessage?.source === "ai") {
diff --git a/hlx_statics/blocks/ai-assistant/ai-assistant_chat-history.js b/hlx_statics/blocks/ai-assistant/ai-assistant_chat-history.js
index 0f3c7277..d061eb07 100644
--- a/hlx_statics/blocks/ai-assistant/ai-assistant_chat-history.js
+++ b/hlx_statics/blocks/ai-assistant/ai-assistant_chat-history.js
@@ -162,9 +162,10 @@ export class ChatHistory {
   }
 
   /**
-   * Stores the Bedrock session id for the current conversation. Overwrites any
-   * existing value, so a fresh id minted after an expiry-triggered reset is
-   * reused transparently on subsequent requests. No-ops when unchanged.
+   * Stores the Bedrock session id for the current conversation, overwriting any
+   * existing value. Recovery from an expired id is handled by the backend: it
+   * mints a fresh session and returns the new id in the `metadata` event, which
+   * we simply persist here. No-ops when unchanged.
    * @param {string|null} sessionId
    */
   setSessionId(sessionId) {
@@ -173,20 +174,6 @@ export class ChatHistory {
     this._save({ ...conversation, sessionId });
   }
 
-  /**
-   * Gets messages formatted for AI context
-   * @param {Object} [options]
-   * @param {number} [options.excludeLast=2] - Number of recent messages to exclude (0 = include all)
-   * @returns {string} Formatted context string
-   */
-  getContextForAI({ excludeLast = 2 } = {}) {
-    const messages = this.getAll();
-    const sliced = excludeLast > 0 ? messages.slice(0, -excludeLast) : messages;
-    return sliced
-      .map(({ source, content }) => JSON.stringify({ source, content }))
-      .join("\n");
-  }
-
   /**
    * Clears all history
    */
diff --git a/hlx_statics/blocks/ai-assistant/ai-assistant_suggested-questions.js b/hlx_statics/blocks/ai-assistant/ai-assistant_suggested-questions.js
index 2236de57..78d813de 100644
--- a/hlx_statics/blocks/ai-assistant/ai-assistant_suggested-questions.js
+++ b/hlx_statics/blocks/ai-assistant/ai-assistant_suggested-questions.js
@@ -7,31 +7,6 @@ import {
   INITIAL_SUGGESTED_QUESTIONS,
 } from "./ai-assistant_constants.js";
 
-/**
- * Parses AI-generated suggested questions from the ---question--- delimited format.
- * @param {string} responseText - Raw text from the AI
- * @returns {Array<{label: string, question: string}>} Parsed questions, or empty array on failure
- */
-export const parseAiSuggestedQuestions = (responseText) => {
-  if (!responseText) return [];
-  const questions = [];
-  const segments = responseText.split(/---question---/);
-  for (const segment of segments) {
-    const trimmed = segment.trim();
-    if (!trimmed) continue;
-    const labelMatch = trimmed.match(/^label:\s*(.+)$/m);
-    const textMatch = trimmed.match(/^text:\s*(.+)$/m);
-    if (labelMatch && textMatch) {
-      const label = labelMatch[1].trim();
-      const question = textMatch[1].trim();
-      if (label && question) {
-        questions.push({ label, question });
-      }
-    }
-  }
-  return questions;
-};
-
 /**
  * Updates the suggested questions list with new questions or a loading skeleton.
  * @param {Array<{label: string, question: string, id?: string|null}>|null} questions - Questions to show, or null for skeleton
@@ -102,15 +77,16 @@ export const createSuggestedQuestionsSection = () => {
 };
 
 /**
- * Shows the suggested questions section with optional scroll behavior.
+ * Reveals the suggested questions section (fade-in + optional scroll). No-ops
+ * unless the section is currently `hidden`. 
  * @param {Object} [options={}] - Options object
- * @param {boolean} [options.shouldScrollIntoView=true] - Whether to scroll the element into view
+ * @param {boolean} [options.shouldScrollIntoView=true] - Scroll the section into view on reveal. Only applies when the section is `hidden` (i.e. actually being revealed).
  */
 export const showSuggestedQuestions = ({
   shouldScrollIntoView = true,
 } = {}) => {
   const el = ELEMENTS.CHAT_SUGGESTED_QUESTIONS;
-  if (el) {
+  if (el?.classList.contains("hidden")) {
     el.classList.remove("hidden");
     el.classList.remove("animate-fade-in");
     requestAnimationFrame(() => {