diff --git a/backend/src/plugins/llm_analysis/service.py b/backend/src/plugins/llm_analysis/service.py
index 70bebe54..8d5ec1fe 100644
--- a/backend/src/plugins/llm_analysis/service.py
+++ b/backend/src/plugins/llm_analysis/service.py
@@ -953,12 +953,19 @@ class LLMClient:
     # @POST Returns a parsed JSON dictionary.
     # @SIDE_EFFECT Calls external LLM API.
     def _should_retry(exception: Exception) -> bool:
-        """Custom retry predicate that excludes authentication errors."""
+        """Custom retry predicate that excludes non-recoverable errors."""
         # Don't retry on authentication errors
         if isinstance(exception, OpenAIAuthenticationError):
             return False
-        # Retry on rate limit errors and other exceptions
-        return isinstance(exception, (RateLimitError, Exception))
+        # Don't retry on null content / model errors — retrying won't help
+        msg = str(exception).lower()
+        if "null content" in msg or "none" in msg:
+            return False
+        # Retry on rate limit errors
+        if isinstance(exception, RateLimitError):
+            return True
+        # For other exceptions, limit retries
+        return True
 
     @retry(
         stop=stop_after_attempt(5),
@@ -1058,6 +1065,10 @@ class LLMClient:
             content = response.choices[0].message.content
             logger.debug(f"[get_json_completion] Raw content to parse: {content}")
 
+            # LLM returned null content — likely content filter or rate limit
+            if content is None:
+                raise RuntimeError("LLM returned null content (content filter or rate limit)")
+
             try:
                 return json.loads(content)
             except json.JSONDecodeError: