diff --git a/core/segment_tts.py b/core/segment_tts.py
index 7a78555..01e0c13 100644
--- a/core/segment_tts.py
+++ b/core/segment_tts.py
@@ -119,6 +119,12 @@ def analyze(self, text: str, index: int, total: int) -> Tuple[str, float]:
             if re.search(rule["pattern"], text):
                 candidates.append((rule["type"], rule["confidence"]))
 
+        # 位置规则兜底：最后一个 segment 如果没有被关键词/正则覆盖，默认 cta
+        if index == total - 1 and not any(t == "cta" for t, _ in candidates):
+            pos_rules = self.rules.get("position_rules", {})
+            if "last" in pos_rules:
+                candidates.append((pos_rules["last"]["type"], pos_rules["last"]["confidence"]))
+
         if not candidates:
             # 默认：中间段为 narrative
             if 0 < index < total - 1:
@@ -127,6 +133,13 @@ def analyze(self, text: str, index: int, total: int) -> Tuple[str, float]:
 
         # 选择置信度最高的类型
         best_type, best_conf = max(candidates, key=lambda x: x[1])
+
+        # Guard：非最后一个 segment 不允许被推断为 cta
+        # CTA 只能出现在视频结尾，避免中间段被误标为行动号召
+        if best_type == "cta" and index < total - 1:
+            # 降级为 narrative，置信度中等
+            return "narrative", 0.5
+
         return best_type, best_conf
 
 
diff --git a/extensions/storyboard/storyboard_ai.py b/extensions/storyboard/storyboard_ai.py
index b27d0c3..fe85f6c 100644
--- a/extensions/storyboard/storyboard_ai.py
+++ b/extensions/storyboard/storyboard_ai.py
@@ -102,6 +102,20 @@ def _infer_segment_type(self, text: str, index: int, total: int) -> str:
             if candidates:
                 return max(candidates, key=lambda x: x[1])[0]
 
+        # 位置规则兜底
+        pos_rules = self.rules.get("position_rules", {})
+        if index == 0 and "first" in pos_rules:
+            candidates.append((pos_rules["first"]["type"], pos_rules["first"]["confidence"]))
+        if index == total - 1 and "last" in pos_rules:
+            candidates.append((pos_rules["last"]["type"], pos_rules["last"]["confidence"]))
+
+        if candidates:
+            best_type, _ = max(candidates, key=lambda x: x[1])
+            # Guard：非最后一个 segment 不允许被推断为 cta
+            if best_type == "cta" and index < total - 1:
+                return "narrative"
+            return best_type
+
         # 简单回退
         if index == 0:
             return "hook"
diff --git a/rules/segment_types.json b/rules/segment_types.json
index 9630de4..9808705 100644
--- a/rules/segment_types.json
+++ b/rules/segment_types.json
@@ -9,8 +9,8 @@
     },
     "last": {
       "type": "cta",
-      "confidence": 0.6,
-      "description": "最后一段默认为 cta，但如果有 CTA 关键词则 confidence 更高"
+      "confidence": 0.9,
+      "description": "最后一段默认为 cta，高 confidence 确保结尾引导"
     }
   },
   "keyword_rules": [
diff --git a/scripts/fix_cta_overuse.py b/scripts/fix_cta_overuse.py
new file mode 100755
index 0000000..77031a5
--- /dev/null
+++ b/scripts/fix_cta_overuse.py
@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+"""
+修复 segments.json 中 CTA 过多的问题
+
+用法：
+    python scripts/fix_cta_overuse.py output/rsi_cn/segments.json
+
+逻辑：
+    - 只有最后一个 segment 允许类型为 "cta"
+    - 其他被误标为 "cta" 的 segment 降级为 "narrative"
+"""
+
+import json
+import sys
+from pathlib import Path
+
+
+def fix_cta_overuse(segments_path: str):
+    path = Path(segments_path)
+    with open(path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+
+    segments = data.get("segments", [])
+    if not segments:
+        print("No segments found")
+        return
+
+    total = len(segments)
+    fixed = 0
+
+    for i, seg in enumerate(segments):
+        if seg.get("segment_type") == "cta" and i < total - 1:
+            seg["segment_type"] = "narrative"
+            seg["segment_type_confidence"] = 0.5
+            fixed += 1
+            print(f"  Fixed {seg['id']}: cta -> narrative")
+
+    if fixed > 0:
+        # 备份原文件
+        backup_path = path.with_suffix(".json.bak")
+        path.rename(backup_path)
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(data, f, ensure_ascii=False, indent=2)
+        print(f"\nFixed {fixed} segment(s). Original backed up to {backup_path.name}")
+    else:
+        print("No overuse CTA segments found. All good!")
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print(f"Usage: python {sys.argv[0]} <segments.json>")
+        sys.exit(1)
+    fix_cta_overuse(sys.argv[1])