diff --git a/backend/app/services/ai_service.py b/backend/app/services/ai_service.py
index 44a1d2d..9747598 100644
--- a/backend/app/services/ai_service.py
+++ b/backend/app/services/ai_service.py
@@ -33,7 +33,7 @@ Your task: extract every question and return ONLY a JSON object in this exact fo
"question_type": "mcq",
"options": ["", " ", " ", " ", " "],
"correct_answer": "",
- "explanation": "\\n\\nCritique: \\n\\nContent Specifications: ",
+ "explanation": "",
"page_reference": {page_ref}
}}
]}}
@@ -44,9 +44,12 @@ CRITICAL RULES — follow exactly:
NEVER store just the letter. NEVER guess. If not found, set correct_answer to null.
Example: options=["alpha","beta","gamma","delta","epsilon"] and "Correct Answer: C" → correct_answer="gamma"
-2. EXPLANATION: Copy the explanation paragraph, then the Critique section, then Content Specifications
- ALL VERBATIM — do NOT summarize, shorten, or paraphrase anything.
- Format: "\\n\\nCritique: \\n\\nContent Specifications: "
+2. EXPLANATION: Copy EVERYTHING that appears after the correct answer line and before the next question,
+ verbatim and in full — including the explanation paragraph, any Critique section, Learning Points,
+ Content Specifications, American Board of Pediatrics specifications, Suggested Readings, and any
+ other content. Do NOT summarize, shorten, skip, or paraphrase a single word.
+ Keep all section headers (e.g. "Critique:", "Content Specifications:", "Suggested Reading:").
+ The explanation field should be the complete, unaltered text block.
3. QUESTIONS: Extract ALL questions, even if partially cut off.
A new question starts with "Item NNN" or "ltem NNN" (OCR artifact — lowercase l instead of I).