diff --git a/backend/app/services/video_renderer.py b/backend/app/services/video_renderer.py
index 6a8832a..557badd 100644
--- a/backend/app/services/video_renderer.py
+++ b/backend/app/services/video_renderer.py
@@ -511,10 +511,11 @@ class VideoRendererService:
         logger.info(f"Starting pause-insert render for {source_video_path}")
         placements = analysis.get("placements", [])
 
-        # Sort placements by pause_point time
+        # Sort placements by pause_point time, with ad_cue_index as secondary key
+        # to ensure consolidated cues (sharing same pause_point) maintain VTT order
         sorted_placements = sorted(
             [p for p in placements if p.get("pause_point") is not None],
-            key=lambda p: p["pause_point"]
+            key=lambda p: (p["pause_point"], p.get("ad_cue_index", 0))
         )
 
         # Debug logging for pause points (midpoint algorithm with silence buffers)