diff --git a/src/pipecat/services/google/gemini_live/llm.py b/src/pipecat/services/google/gemini_live/llm.py
index c1af46745..5c2871fdc 100644
--- a/src/pipecat/services/google/gemini_live/llm.py
+++ b/src/pipecat/services/google/gemini_live/llm.py
@@ -1040,7 +1040,7 @@ class GeminiLiveLLMService(LLMService):
                 context
             )  # Take a static snapshot guaranteed not to change
 
-            if self._context_update_requires_reconnect(diff):
+            if self._context_update_requires_reconnect(diff, messages_programmatically_edited):
                 # Reconnect
                 logger.debug("Context update requires reconnect. Reconnecting...")
 
@@ -1057,41 +1057,46 @@ class GeminiLiveLLMService(LLMService):
                 # Trigger "initial" response with new connection
                 await self._create_initial_response()
             else:
-                # If messages were appended, they might be:
-                # - Messages generated by Gemini Live itself, which the remote
-                #   service and its internal context management is already
-                #   aware of. We can ignore these.
-                # - Messages appended programmatically by the user (e.g. via
-                #   an LLMMessagesAppendFrame). We need to send these to the
-                #   remote Gemini Live service.
-                # - Function call results, which are handled separately below.
-                if diff.messages_appended and messages_programmatically_edited:
-                    # Messages were programmatically edited - send to the API
-                    logger.debug(
-                        "Context update includes programmatically-appended messages. Creating incremental response..."
-                    )
-                    await self._create_incremental_response(diff.messages_appended)
-
                 # Send results for newly-completed function calls, if any.
                 logger.debug("Checking for newly-completed function call results...")
                 await self._process_completed_function_calls(send_new_results=True)
 
-    def _context_update_requires_reconnect(self, diff: LLMContextDiff) -> bool:
+    def _context_update_requires_reconnect(
+        self, diff: LLMContextDiff, messages_programmatically_edited: bool
+    ) -> bool:
         """Check if an update to our LLM context requires reconnection.
 
         Args:
             diff: The context diff representing the update
+            messages_programmatically_edited: Whether context messages were
+                programmatically edited (e.g. with LLMMessagesAppendFrame)
+
         Returns:
             True if reconnection is required, False otherwise.
         """
-        if diff.history_edited or diff.tools_diff.has_changes():
+        # We need to reconnect in 3 cases:
+        # 1. If the conversation history was edited
+        # 2. If the tools available to the model were changed
+        # 3. If messages were appended programmatically by the user, e.g. with
+        #    LLMMessagesAppendFrame (NOT if they originated from Gemini Live
+        #    itself, since the service is already aware of those)
+        #
+        # Note that *ideally* in this 3rd case we would just send the newly
+        # appended messages without reconnecting, but in all my testing so far,
+        # I haven't been able to get that to work reliably.
+        # TODO: add more detail about what's been attempted.
+        if (
+            diff.history_edited
+            or diff.tools_diff.has_changes()
+            or (diff.messages_appended and messages_programmatically_edited)
+        ):
             if diff.history_edited:
                 print("[pk] Context diff: history edited.")
             if diff.tools_diff.has_changes():
                 print("[pk] Context diff: tools changed.")
+            if diff.messages_appended and messages_programmatically_edited:
+                print(f"[pk] Context diff: messages appended: {diff.messages_appended}")
             return True
-        elif diff.messages_appended:
-            print(f"[pk] Context diff: messages appended: {diff.messages_appended}")
 
         return False
 
@@ -1481,32 +1486,6 @@ class GeminiLiveLLMService(LLMService):
         if not self._inference_on_context_initialization:
             self._needs_turn_complete_message = True
 
-    async def _create_incremental_response(self, new_messages: List[LLMContextMessage]):
-        """Create a new response mid-conversation with newly-appended messages.
-
-        Args:
-            new_messages: The new messages to send.
-        """
-        if self._disconnecting or not self._session or not new_messages:
-            return
-
-        # Create a throwaway context just for the purpose of getting messages
-        adapter: GeminiLLMAdapter = self.get_llm_adapter()
-        messages = adapter.get_llm_invocation_params(
-            LLMContext(messages=new_messages), strip_function_messages=True
-        ).get("messages", [])
-        if not messages:
-            return
-
-        logger.debug(f"Creating response with new messages: {messages}")
-
-        await self.start_ttfb_metrics()
-
-        try:
-            await self._session.send_client_content(turns=messages, turn_complete=True)
-        except Exception as e:
-            await self._handle_send_error(e)
-
     async def _create_single_response(self, messages_list):
         """Create a single response from a list of messages."""
         if self._disconnecting or not self._session: