diff --git a/packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/span_utils.py b/packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/span_utils.py
index e80147d968..f7d71b18ab 100644
--- a/packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/span_utils.py
+++ b/packages/opentelemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/span_utils.py
@@ -211,10 +211,17 @@ def set_chat_response(span: Span, response: LLMResult) -> None:
     for generations in response.generations:
         for generation in generations:
             prefix = f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}"
+
+            if hasattr(generation, "message") and generation.message and hasattr(generation.message, "type"):
+                role = _message_type_to_role(generation.message.type)
+            else:
+                # For non-chat completions (Generation objects), default to assistant
+                role = "assistant"
+
             _set_span_attribute(
                 span,
                 f"{prefix}.role",
-                _message_type_to_role(generation.type),
+                role,
             )
 
             # Try to get content from various sources
diff --git a/packages/opentelemetry-instrumentation-langchain/tests/test_generation_role_extraction.py b/packages/opentelemetry-instrumentation-langchain/tests/test_generation_role_extraction.py
new file mode 100644
index 0000000000..60068843a3
--- /dev/null
+++ b/packages/opentelemetry-instrumentation-langchain/tests/test_generation_role_extraction.py
@@ -0,0 +1,142 @@
+"""
+Tests for generation role extraction in completion spans.
+
+This tests the fix for generation.type returning class names ("ChatGeneration", "Generation")
+instead of message types ("ai", "tool", etc.), which caused completion roles to appear as "unknown"
+in observability traces.
+"""
+
+import pytest
+from unittest.mock import Mock
+from langchain_core.outputs import LLMResult, ChatGeneration, Generation
+from langchain_core.messages import AIMessage, ToolMessage
+from opentelemetry.semconv._incubating.attributes import gen_ai_attributes as GenAIAttributes
+from opentelemetry.instrumentation.langchain.span_utils import set_chat_response
+
+
+class TestCompletionRoleExtraction:
+    """Test that completion roles are correctly extracted from generation objects."""
+
+    @pytest.fixture
+    def mock_span(self):
+        """Create a mock span for testing."""
+        span = Mock()
+        span.is_recording.return_value = True
+        span.attributes = {}
+
+        def set_attribute(key, value):
+            span.attributes[key] = value
+
+        span.set_attribute = set_attribute
+        return span
+
+    def test_chat_generation_with_ai_message_role(self, mock_span, monkeypatch):
+        """Test that ChatGeneration with AIMessage correctly extracts 'assistant' role."""
+        # Mock should_send_prompts to return True
+        monkeypatch.setattr(
+            "opentelemetry.instrumentation.langchain.span_utils.should_send_prompts",
+            lambda: True
+        )
+
+        # Create ChatGeneration with AIMessage
+        generation = ChatGeneration(message=AIMessage(content="Hello!"))
+        llm_result = LLMResult(generations=[[generation]])
+
+        # Call the function
+        set_chat_response(mock_span, llm_result)
+
+        # Assert role is 'assistant', not 'unknown'
+        role_key = f"{GenAIAttributes.GEN_AI_COMPLETION}.0.role"
+        assert role_key in mock_span.attributes
+        assert mock_span.attributes[role_key] == "assistant"
+
+    def test_chat_generation_with_tool_message_role(self, mock_span, monkeypatch):
+        """Test that ChatGeneration with ToolMessage correctly extracts 'tool' role."""
+        # Mock should_send_prompts to return True
+        monkeypatch.setattr(
+            "opentelemetry.instrumentation.langchain.span_utils.should_send_prompts",
+            lambda: True
+        )
+
+        # Create ChatGeneration with ToolMessage
+        generation = ChatGeneration(
+            message=ToolMessage(content="Tool result", tool_call_id="123")
+        )
+        llm_result = LLMResult(generations=[[generation]])
+
+        # Call the function
+        set_chat_response(mock_span, llm_result)
+
+        # Assert role is 'tool', not 'unknown'
+        role_key = f"{GenAIAttributes.GEN_AI_COMPLETION}.0.role"
+        assert role_key in mock_span.attributes
+        assert mock_span.attributes[role_key] == "tool"
+
+    def test_generation_without_message_defaults_to_assistant(self, mock_span, monkeypatch):
+        """Test that Generation (non-chat) defaults to 'assistant' role."""
+        # Mock should_send_prompts to return True
+        monkeypatch.setattr(
+            "opentelemetry.instrumentation.langchain.span_utils.should_send_prompts",
+            lambda: True
+        )
+
+        # Create Generation without message (legacy completion)
+        generation = Generation(text="This is a completion")
+        llm_result = LLMResult(generations=[[generation]])
+
+        # Call the function
+        set_chat_response(mock_span, llm_result)
+
+        # Assert role defaults to 'assistant', not 'unknown'
+        role_key = f"{GenAIAttributes.GEN_AI_COMPLETION}.0.role"
+        assert role_key in mock_span.attributes
+        assert mock_span.attributes[role_key] == "assistant"
+
+    def test_multiple_generations_with_different_roles(self, mock_span, monkeypatch):
+        """Test that multiple generations with different message types are handled correctly."""
+        # Mock should_send_prompts to return True
+        monkeypatch.setattr(
+            "opentelemetry.instrumentation.langchain.span_utils.should_send_prompts",
+            lambda: True
+        )
+
+        # Create multiple generations with different message types
+        gen1 = ChatGeneration(message=AIMessage(content="AI response"))
+        gen2 = ChatGeneration(message=ToolMessage(content="Tool result", tool_call_id="123"))
+        gen3 = Generation(text="Legacy completion")
+
+        llm_result = LLMResult(generations=[[gen1], [gen2], [gen3]])
+
+        # Call the function
+        set_chat_response(mock_span, llm_result)
+
+        # Assert all roles are correctly set
+        assert mock_span.attributes[f"{GenAIAttributes.GEN_AI_COMPLETION}.0.role"] == "assistant"
+        assert mock_span.attributes[f"{GenAIAttributes.GEN_AI_COMPLETION}.1.role"] == "tool"
+        assert mock_span.attributes[f"{GenAIAttributes.GEN_AI_COMPLETION}.2.role"] == "assistant"
+
+    def test_generation_type_attribute_is_not_used(self, mock_span, monkeypatch):
+        """Test that generation.type (which returns class name) is not used directly."""
+        # Mock should_send_prompts to return True
+        monkeypatch.setattr(
+            "opentelemetry.instrumentation.langchain.span_utils.should_send_prompts",
+            lambda: True
+        )
+
+        # Create ChatGeneration - note that generation.type would be "ChatGeneration"
+        generation = ChatGeneration(message=AIMessage(content="Test"))
+
+        # Verify the bug scenario: generation.type returns class name, not message type
+        assert generation.type == "ChatGeneration"  # This is the bug
+        assert generation.message.type == "ai"  # This is what we should use
+
+        llm_result = LLMResult(generations=[[generation]])
+
+        # Call the function
+        set_chat_response(mock_span, llm_result)
+
+        # Assert role is 'assistant', not 'unknown'
+        # If the bug existed, passing generation.type directly to _message_type_to_role
+        # would return 'unknown' because "ChatGeneration" doesn't match any message type
+        role_key = f"{GenAIAttributes.GEN_AI_COMPLETION}.0.role"
+        assert mock_span.attributes[role_key] == "assistant"