diff --git a/AmazonBedrock/anthropic/08_Avoiding_Hallucinations.ipynb b/AmazonBedrock/anthropic/08_Avoiding_Hallucinations.ipynb
index 8bf0e68..7fefd59 100755
--- a/AmazonBedrock/anthropic/08_Avoiding_Hallucinations.ipynb
+++ b/AmazonBedrock/anthropic/08_Avoiding_Hallucinations.ipynb
@@ -207,9 +207,16 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "How do we fix this? Well, a great way to reduce hallucinations on long documents is to **make Claude gather evidence first.** \n",
+    "How do we fix this? In practice, the strongest version of this pattern uses **two separate safeguards together**:\n",
     "\n",
-    "In this case, we **tell Claude to first extract relevant quotes, then base its answer on those quotes**. Telling Claude to do so here makes it correctly notice that the quote does not answer the question."
+    "1. **Evidence gathering**: have Claude pull the most relevant quote before answering.\n",
+    "2. **Information-sufficiency check**: have Claude explicitly decide whether that quote actually answers the question or whether it still lacks the needed detail.\n",
+    "\n",
+    "The prompt below intentionally combines both. That matters because they do different jobs:\n",
+    "- The quote extraction step improves grounding and makes the model show its work.\n",
+    "- The sufficiency check gives Claude permission to say that the document does **not** fully answer the question, instead of guessing from a nearby distractor.\n",
+    "\n",
+    "If you only ask for quotes, Claude can still over-interpret a near match. If you only ask whether the document is sufficient, Claude is less likely to hallucinate, but you lose the explicit evidence trail. **Using both together is the most reliable teaching example here.**\n"
    ]
   },
   {
@@ -376,8 +383,8 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### Exercise 8.1 - Prospectus Hallucination\n",
-    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations. The correct answer is that subscribers went up 49x."
+    "### Exercise 8.2 - Prospectus Hallucination\n",
+    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations **and** telling it to check whether those citations actually support the final answer. The correct answer is that subscribers went up 49x.\n"
    ]
   },
   {
diff --git a/AmazonBedrock/boto3/08_Avoiding_Hallucinations.ipynb b/AmazonBedrock/boto3/08_Avoiding_Hallucinations.ipynb
index a303f1b..62a8712 100755
--- a/AmazonBedrock/boto3/08_Avoiding_Hallucinations.ipynb
+++ b/AmazonBedrock/boto3/08_Avoiding_Hallucinations.ipynb
@@ -213,9 +213,16 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "How do we fix this? Well, a great way to reduce hallucinations on long documents is to **make Claude gather evidence first.** \n",
+    "How do we fix this? In practice, the strongest version of this pattern uses **two separate safeguards together**:\n",
     "\n",
-    "In this case, we **tell Claude to first extract relevant quotes, then base its answer on those quotes**. Telling Claude to do so here makes it correctly notice that the quote does not answer the question."
+    "1. **Evidence gathering**: have Claude pull the most relevant quote before answering.\n",
+    "2. **Information-sufficiency check**: have Claude explicitly decide whether that quote actually answers the question or whether it still lacks the needed detail.\n",
+    "\n",
+    "The prompt below intentionally combines both. That matters because they do different jobs:\n",
+    "- The quote extraction step improves grounding and makes the model show its work.\n",
+    "- The sufficiency check gives Claude permission to say that the document does **not** fully answer the question, instead of guessing from a nearby distractor.\n",
+    "\n",
+    "If you only ask for quotes, Claude can still over-interpret a near match. If you only ask whether the document is sufficient, Claude is less likely to hallucinate, but you lose the explicit evidence trail. **Using both together is the most reliable teaching example here.**\n"
    ]
   },
   {
@@ -382,8 +389,8 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### Exercise 8.1 - Prospectus Hallucination\n",
-    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations. The correct answer is that subscribers went up 49x."
+    "### Exercise 8.2 - Prospectus Hallucination\n",
+    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations **and** telling it to check whether those citations actually support the final answer. The correct answer is that subscribers went up 49x.\n"
    ]
   },
   {
diff --git a/Anthropic 1P/08_Avoiding_Hallucinations.ipynb b/Anthropic 1P/08_Avoiding_Hallucinations.ipynb
index 7e3df42..5b12bf6 100644
--- a/Anthropic 1P/08_Avoiding_Hallucinations.ipynb	
+++ b/Anthropic 1P/08_Avoiding_Hallucinations.ipynb	
@@ -201,9 +201,16 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "How do we fix this? Well, a great way to reduce hallucinations on long documents is to **make Claude gather evidence first.** \n",
+    "How do we fix this? In practice, the strongest version of this pattern uses **two separate safeguards together**:\n",
     "\n",
-    "In this case, we **tell Claude to first extract relevant quotes, then base its answer on those quotes**. Telling Claude to do so here makes it correctly notice that the quote does not answer the question."
+    "1. **Evidence gathering**: have Claude pull the most relevant quote before answering.\n",
+    "2. **Information-sufficiency check**: have Claude explicitly decide whether that quote actually answers the question or whether it still lacks the needed detail.\n",
+    "\n",
+    "The prompt below intentionally combines both. That matters because they do different jobs:\n",
+    "- The quote extraction step improves grounding and makes the model show its work.\n",
+    "- The sufficiency check gives Claude permission to say that the document does **not** fully answer the question, instead of guessing from a nearby distractor.\n",
+    "\n",
+    "If you only ask for quotes, Claude can still over-interpret a near match. If you only ask whether the document is sufficient, Claude is less likely to hallucinate, but you lose the explicit evidence trail. **Using both together is the most reliable teaching example here.**\n"
    ]
   },
   {
@@ -370,8 +377,8 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### Exercise 8.1 - Prospectus Hallucination\n",
-    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations. The correct answer is that subscribers went up 49x."
+    "### Exercise 8.2 - Prospectus Hallucination\n",
+    "Modify the `PROMPT` to fix Claude's hallucination issue by asking for citations **and** telling it to check whether those citations actually support the final answer. The correct answer is that subscribers went up 49x.\n"
    ]
   },
   {
diff --git a/scripts/check_hallucination_lesson_wording.py b/scripts/check_hallucination_lesson_wording.py
new file mode 100644
index 0000000..b0d27be
--- /dev/null
+++ b/scripts/check_hallucination_lesson_wording.py
@@ -0,0 +1,39 @@
+#!/usr/bin/env python3
+import json
+from pathlib import Path
+import sys
+
+ROOT = Path(__file__).resolve().parents[1]
+NOTEBOOKS = [
+    ROOT / 'Anthropic 1P/08_Avoiding_Hallucinations.ipynb',
+    ROOT / 'AmazonBedrock/anthropic/08_Avoiding_Hallucinations.ipynb',
+    ROOT / 'AmazonBedrock/boto3/08_Avoiding_Hallucinations.ipynb',
+]
+
+required_fragments = [
+    'Evidence gathering',
+    'Information-sufficiency check',
+    'Using both together is the most reliable teaching example here.',
+]
+exercise_heading = '### Exercise 8.2 - Prospectus Hallucination'
+exercise_phrase = 'asking for citations **and** telling it to check whether those citations actually support the final answer'
+
+failed = False
+for notebook in NOTEBOOKS:
+    data = json.loads(notebook.read_text())
+    lesson = ''.join(data['cells'][9]['source'])
+    exercise = ''.join(data['cells'][17]['source'])
+
+    missing = [frag for frag in required_fragments if frag not in lesson]
+    if missing:
+        failed = True
+        print(f'{notebook}: missing lesson fragments: {missing}')
+
+    if exercise_heading not in exercise or exercise_phrase not in exercise:
+        failed = True
+        print(f'{notebook}: exercise wording not updated as expected')
+
+if failed:
+    sys.exit(1)
+
+print(f'validated {len(NOTEBOOKS)} chapter 8 notebooks')