diff --git a/src/google/adk/evaluation/agent_evaluator.py b/src/google/adk/evaluation/agent_evaluator.py
index 3304b4359ae..67d6eef5ec0 100644
--- a/src/google/adk/evaluation/agent_evaluator.py
+++ b/src/google/adk/evaluation/agent_evaluator.py
@@ -60,6 +60,7 @@
 from .eval_sets_manager import EvalSetsManager
 from .evaluator import EvalStatus
 from .in_memory_eval_sets_manager import InMemoryEvalSetsManager
+from .llm_as_judge_utils import get_text_from_content
 from .local_eval_sets_manager import convert_eval_set_to_pydantic_schema
 from .simulation.user_simulator_provider import UserSimulatorProvider
 
@@ -601,10 +602,7 @@ def _print_details(
 
   @staticmethod
   def _convert_content_to_text(content: Optional[genai_types.Content]) -> str:
-    if content and content.parts:
-      return "\n".join([p.text for p in content.parts if p.text])
-
-    return ""
+    return get_text_from_content(content) or ""
 
   @staticmethod
   def _convert_tool_calls_to_text(
diff --git a/src/google/adk/evaluation/final_response_match_v1.py b/src/google/adk/evaluation/final_response_match_v1.py
index 63559b26a9c..ccd0e218fab 100644
--- a/src/google/adk/evaluation/final_response_match_v1.py
+++ b/src/google/adk/evaluation/final_response_match_v1.py
@@ -32,6 +32,7 @@
 from .evaluator import EvaluationResult
 from .evaluator import Evaluator
 from .evaluator import PerInvocationResult
+from .llm_as_judge_utils import get_text_from_content
 
 
 class RougeEvaluator(Evaluator):
@@ -90,10 +91,7 @@ def evaluate_invocations(
 
 
 def _get_text_from_content(content: Optional[genai_types.Content]) -> str:
-  if content and content.parts:
-    return "\n".join([part.text for part in content.parts if part.text])
-
-  return ""
+  return get_text_from_content(content) or ""
 
 
 def _get_eval_status(score: float, threshold: float) -> EvalStatus:
diff --git a/src/google/adk/evaluation/hallucinations_v1.py b/src/google/adk/evaluation/hallucinations_v1.py
index 5e32389316f..987d32e47d5 100644
--- a/src/google/adk/evaluation/hallucinations_v1.py
+++ b/src/google/adk/evaluation/hallucinations_v1.py
@@ -46,6 +46,7 @@
 from .evaluator import PerInvocationResult
 from .llm_as_judge_utils import get_eval_status
 from .llm_as_judge_utils import get_text_from_content
+from .llm_as_judge_utils import get_text_parts
 from .llm_as_judge_utils import get_tool_declarations_as_json_str
 
 logger = logging.getLogger("google_adk." + __name__)
@@ -461,7 +462,7 @@ def _create_context_for_step(
           for part in event.content.parts
           if part.function_response
       ]
-      nl_responses = [part.text for part in event.content.parts if part.text]
+      nl_responses = get_text_parts(event.content)
 
       if nl_responses:
         context_parts.append("\n".join(nl_responses) + "\n")
@@ -650,11 +651,7 @@ def _get_steps_to_evaluate(self, actual: Invocation) -> list[EvaluationStep]:
 
     if self._criterion.evaluate_intermediate_nl_responses:
       for event in all_events:
-        nl_parts = (
-            [p.text for p in event.content.parts if p.text]
-            if event.content and event.content.parts
-            else []
-        )
+        nl_parts = get_text_parts(event.content)
         if nl_parts:
           context = self._create_context_for_step(
               actual.app_details, actual, events_for_context
diff --git a/src/google/adk/evaluation/llm_as_judge_utils.py b/src/google/adk/evaluation/llm_as_judge_utils.py
index 626cb9b3960..161673d11fe 100644
--- a/src/google/adk/evaluation/llm_as_judge_utils.py
+++ b/src/google/adk/evaluation/llm_as_judge_utils.py
@@ -47,6 +47,13 @@ class Label(enum.Enum):
   NOT_FOUND = "label field not found"
 
 
+def get_text_parts(content: Optional[genai_types.Content]) -> list[str]:
+  """Returns the visible text parts of a `Content`, excluding thoughts."""
+  if not content or not content.parts:
+    return []
+  return [p.text for p in content.parts if p.text and not p.thought]
+
+
 def get_text_from_content(
     content: Optional[Union[genai_types.Content, Invocation]],
     *,
@@ -87,7 +94,7 @@ def get_text_from_content(
     return "\n".join(parts) if parts else None
 
   if content and content.parts:
-    return "\n".join([p.text for p in content.parts if p.text])
+    return "\n".join(get_text_parts(content))
 
   return None
 
diff --git a/src/google/adk/evaluation/rubric_based_multi_turn_trajectory_evaluator.py b/src/google/adk/evaluation/rubric_based_multi_turn_trajectory_evaluator.py
index 0fba74f79eb..bd7592efeed 100644
--- a/src/google/adk/evaluation/rubric_based_multi_turn_trajectory_evaluator.py
+++ b/src/google/adk/evaluation/rubric_based_multi_turn_trajectory_evaluator.py
@@ -30,6 +30,7 @@
 from .evaluator import _validate_invocation_lengths
 from .evaluator import EvaluationResult
 from .evaluator import PerInvocationResult
+from .llm_as_judge_utils import get_text_parts
 from .rubric_based_evaluator import RubricBasedEvaluator
 
 logger = logging.getLogger("google_adk." + __name__)
@@ -193,7 +194,7 @@ def _assemble_dialogue_history(
     for turn_index, invocation in enumerate(actual_invocations):
       # USER TURN
       if invocation.user_content and invocation.user_content.parts:
-        text_parts = [p.text for p in invocation.user_content.parts if p.text]
+        text_parts = get_text_parts(invocation.user_content)
         if text_parts:
           dialogue_lines.append(
               f"USER TURN {turn_index + 1}: {' '.join(text_parts)}"
@@ -208,7 +209,7 @@ def _assemble_dialogue_history(
               else f"AGENT ({event.author})"
           )
           if event.content and event.content.parts:
-            text_parts = [p.text for p in event.content.parts if p.text]
+            text_parts = get_text_parts(event.content)
             if text_parts:
               dialogue_lines.append(
                   f"{role} TURN {turn_index + 1}: {' '.join(text_parts)}"
@@ -245,7 +246,7 @@ def _assemble_dialogue_history(
         ):
           agent_name = intermediate_data.invocation_events[0].author
         role = f"AGENT ({agent_name})"
-        text_parts = [p.text for p in invocation.final_response.parts if p.text]
+        text_parts = get_text_parts(invocation.final_response)
         if text_parts:
           dialogue_lines.append(
               f"{role} TURN {turn_index + 1}: {' '.join(text_parts)}"
diff --git a/src/google/adk/evaluation/vertex_ai_eval_facade.py b/src/google/adk/evaluation/vertex_ai_eval_facade.py
index 69d2b82f329..e418e12b67f 100644
--- a/src/google/adk/evaluation/vertex_ai_eval_facade.py
+++ b/src/google/adk/evaluation/vertex_ai_eval_facade.py
@@ -37,6 +37,7 @@
 from .evaluator import EvaluationResult
 from .evaluator import Evaluator
 from .evaluator import PerInvocationResult
+from .llm_as_judge_utils import get_text_from_content
 
 logger = logging.getLogger("google_adk." + __name__)
 
@@ -113,10 +114,7 @@ def evaluate_invocations(
     """
 
   def _get_text(self, content: Optional[genai_types.Content]) -> str:
-    if content and content.parts:
-      return "\n".join([p.text for p in content.parts if p.text])
-
-    return ""
+    return get_text_from_content(content) or ""
 
   def _get_score(self, eval_result: object) -> Optional[float]:
     summary_metrics: object = getattr(eval_result, "summary_metrics", None)
diff --git a/tests/unittests/evaluation/test_agent_evaluator.py b/tests/unittests/evaluation/test_agent_evaluator.py
index 3cc5eb256f8..6cfa6cf0eb6 100644
--- a/tests/unittests/evaluation/test_agent_evaluator.py
+++ b/tests/unittests/evaluation/test_agent_evaluator.py
@@ -488,6 +488,17 @@ def _make_result_with_invocation(
   )
 
 
+def test_convert_content_to_text_excludes_thought_parts():
+  content = genai_types.Content(
+      parts=[
+          genai_types.Part(text="Consider the options.", thought=True),
+          genai_types.Part(text="Paris"),
+      ]
+  )
+
+  assert AgentEvaluator._convert_content_to_text(content) == "Paris"
+
+
 def test_get_results_as_rows_flattens_metrics_and_invocations():
   eval_metric_results = {
       "response_match_score": [
diff --git a/tests/unittests/evaluation/test_final_response_match_v1.py b/tests/unittests/evaluation/test_final_response_match_v1.py
index cd7128f2aa3..c5bad7581b8 100644
--- a/tests/unittests/evaluation/test_final_response_match_v1.py
+++ b/tests/unittests/evaluation/test_final_response_match_v1.py
@@ -59,6 +59,62 @@ def _create_test_invocations(
   )
 
 
+@pytest.mark.parametrize(
+    "actual_parts, expected_parts, score",
+    [
+        (
+            [
+                genai_types.Part(
+                    text="Consider the capital of France.", thought=True
+                ),
+                genai_types.Part(text="Paris"),
+            ],
+            [genai_types.Part(text="Paris")],
+            1.0,
+        ),
+        (
+            [genai_types.Part(text="Paris", thought=False)],
+            [
+                genai_types.Part(
+                    text="Consider the capital of France.", thought=True
+                ),
+                genai_types.Part(text="Paris"),
+            ],
+            1.0,
+        ),
+        (
+            [genai_types.Part(text="Paris", thought=True)],
+            [genai_types.Part(text="Paris")],
+            0.0,
+        ),
+        (
+            [
+                genai_types.Part(text="Paris", thought=True),
+                genai_types.Part(text="London"),
+            ],
+            [genai_types.Part(text="Paris")],
+            0.0,
+        ),
+    ],
+)
+def test_response_match_scores_visible_text_only(
+    actual_parts, expected_parts, score
+):
+  """Thought summaries neither dilute correct answers nor credit wrong ones."""
+  actual, expected = _create_test_invocations("", "")
+  actual.final_response.parts = actual_parts
+  expected.final_response.parts = expected_parts
+  evaluator = _create_test_rouge_evaluator(threshold=0.5)
+
+  result = evaluator.evaluate_invocations([actual], [expected])
+
+  assert result.overall_score == pytest.approx(score)
+  assert result.per_invocation_results[0].score == pytest.approx(score)
+  assert result.overall_eval_status == (
+      EvalStatus.PASSED if score == 1.0 else EvalStatus.FAILED
+  )
+
+
 def test_calculate_rouge_1_scores_empty_candidate_and_reference():
   candidate = ""
   reference = ""
diff --git a/tests/unittests/evaluation/test_final_response_match_v2.py b/tests/unittests/evaluation/test_final_response_match_v2.py
index 2f6bc3b3385..3f865533d9c 100644
--- a/tests/unittests/evaluation/test_final_response_match_v2.py
+++ b/tests/unittests/evaluation/test_final_response_match_v2.py
@@ -268,6 +268,54 @@ def test_format_auto_rater_prompt_includes_intermediate_when_enabled():
   assert "reference intro\nreference final" in prompt
 
 
+def test_format_auto_rater_prompt_excludes_thought_parts():
+  evaluator = _create_test_evaluator_gemini(threshold=0.8)
+  actual_invocation, expected_invocation = _create_test_invocations(
+      "candidate text", "reference text"
+  )
+  actual_invocation.final_response.parts.insert(
+      0, genai_types.Part(text="Considering the response.", thought=True)
+  )
+  expected_invocation.final_response.parts.insert(
+      0, genai_types.Part(text="Considering the reference.", thought=True)
+  )
+
+  prompt = evaluator.format_auto_rater_prompt(
+      actual_invocation, expected_invocation
+  )
+
+  assert "Considering the response." not in prompt
+  assert "Considering the reference." not in prompt
+  assert "candidate text" in prompt
+  assert "reference text" in prompt
+
+
+def test_convert_auto_rater_response_to_score_ignores_judge_thought():
+  """The judge model's own thought text must not break verdict parsing."""
+  evaluator = _create_test_evaluator_gemini(threshold=0.8)
+  auto_rater_response = """```json
+{
+  "is_the_agent_response_valid": "valid",
+  "reasoning": "The response is valid."
+}
+```"""
+  llm_response = LlmResponse(
+      content=genai_types.Content(
+          parts=[
+              genai_types.Part(
+                  text="Let me evaluate this response.", thought=True
+              ),
+              genai_types.Part(text=auto_rater_response),
+          ],
+          role="model",
+      )
+  )
+  auto_rater_score = evaluator.convert_auto_rater_response_to_score(
+      llm_response
+  )
+  assert auto_rater_score == AutoRaterScore(score=1.0)
+
+
 def test_convert_auto_rater_response_to_score_valid():
   evaluator = _create_test_evaluator_gemini(threshold=0.8)
   auto_rater_response = """```json
diff --git a/tests/unittests/evaluation/test_hallucinations_v1.py b/tests/unittests/evaluation/test_hallucinations_v1.py
index b7a1c42f47d..e19b76c5bbb 100644
--- a/tests/unittests/evaluation/test_hallucinations_v1.py
+++ b/tests/unittests/evaluation/test_hallucinations_v1.py
@@ -577,6 +577,116 @@ def agent_tree_data():
   return invocation, expected_invocation
 
 
+class TestCreateContextExcludesThoughts:
+  """Test cases ensuring thought parts do not leak into the context."""
+
+  def test_create_context_excludes_thought_from_nl_response(
+      self, hallucinations_metric
+  ):
+    """A thought part must not appear in the assembled context string."""
+    app_details = AppDetails(
+        agent_details={
+            "root": AgentDetails(
+                name="root", instructions="Root agent instructions."
+            )
+        },
+    )
+    user_content = genai_types.Content(
+        parts=[genai_types.Part(text="User query.")]
+    )
+    events = [
+        InvocationEvent(
+            author="root",
+            content=genai_types.Content(
+                parts=[
+                    genai_types.Part(
+                        text="Considering the query.", thought=True
+                    ),
+                    genai_types.Part(text="Visible NL response."),
+                ]
+            ),
+        ),
+    ]
+    invocation = Invocation(
+        app_details=app_details,
+        user_content=user_content,
+        intermediate_data=InvocationEvents(invocation_events=events),
+    )
+
+    context = hallucinations_metric._create_context_for_step(
+        app_details, invocation, events
+    )
+
+    assert "Considering the query." not in context
+    assert "Visible NL response." in context
+
+
+class TestGetStepsToEvaluateExcludesThoughts:
+  """Test cases ensuring thought parts do not become their own eval steps."""
+
+  def test_thought_only_event_yields_no_step(self, hallucinations_metric):
+    """A thought-only intermediate event must not produce an evaluation step."""
+    app_details = AppDetails(agent_details={})
+    user_content = genai_types.Content(
+        parts=[genai_types.Part(text="User query.")]
+    )
+    events = [
+        InvocationEvent(
+            author="root",
+            content=genai_types.Content(
+                parts=[genai_types.Part(text="Just thinking.", thought=True)]
+            ),
+        ),
+    ]
+    invocation = Invocation(
+        app_details=app_details,
+        user_content=user_content,
+        intermediate_data=InvocationEvents(invocation_events=events),
+        final_response=genai_types.Content(
+            parts=[genai_types.Part(text="Final response.")]
+        ),
+    )
+
+    steps = hallucinations_metric._get_steps_to_evaluate(invocation)
+
+    assert [step.nl_response for step in steps] == ["Final response."]
+
+  def test_mixed_event_does_not_split_thought_into_its_own_step(
+      self, hallucinations_metric
+  ):
+    """A thought alongside visible text yields one step, not two."""
+    app_details = AppDetails(agent_details={})
+    user_content = genai_types.Content(
+        parts=[genai_types.Part(text="User query.")]
+    )
+    events = [
+        InvocationEvent(
+            author="root",
+            content=genai_types.Content(
+                parts=[
+                    genai_types.Part(text="Reasoning aloud.", thought=True),
+                    genai_types.Part(text="Visible NL response."),
+                ]
+            ),
+        ),
+    ]
+    invocation = Invocation(
+        app_details=app_details,
+        user_content=user_content,
+        intermediate_data=InvocationEvents(invocation_events=events),
+        final_response=genai_types.Content(
+            parts=[genai_types.Part(text="Final response.")]
+        ),
+    )
+
+    steps = hallucinations_metric._get_steps_to_evaluate(invocation)
+
+    assert [step.nl_response for step in steps] == [
+        "Visible NL response.",
+        "Final response.",
+    ]
+
+
 class TestEvaluateInvocationsAgentTree:
   """Test cases for agent tree."""
 
diff --git a/tests/unittests/evaluation/test_llm_as_judge_utils.py b/tests/unittests/evaluation/test_llm_as_judge_utils.py
index 6e6dd8772a4..f09a2692ca2 100644
--- a/tests/unittests/evaluation/test_llm_as_judge_utils.py
+++ b/tests/unittests/evaluation/test_llm_as_judge_utils.py
@@ -28,6 +28,7 @@
 from google.adk.evaluation.llm_as_judge_utils import get_eval_status
 from google.adk.evaluation.llm_as_judge_utils import get_grounding_metadata_as_json_str
 from google.adk.evaluation.llm_as_judge_utils import get_text_from_content
+from google.adk.evaluation.llm_as_judge_utils import get_text_parts
 from google.adk.evaluation.llm_as_judge_utils import get_tool_calls_and_responses_as_json_str
 from google.adk.evaluation.llm_as_judge_utils import get_tool_declarations_as_json_str
 from google.genai import types as genai_types
@@ -133,6 +134,82 @@ def test_get_text_from_content_with_invocation_include_intermediate_responses_in
   )
 
 
+def test_get_text_from_content_excludes_thought_parts():
+  """Tests get_text_from_content excludes parts marked as thoughts."""
+  content = genai_types.Content(
+      parts=[
+          genai_types.Part(text="Let me think about this.", thought=True),
+          genai_types.Part(text="Paris"),
+      ]
+  )
+  assert get_text_from_content(content) == "Paris"
+
+
+def test_get_text_from_content_with_only_thought_parts():
+  """Tests get_text_from_content returns empty string for thought-only content."""
+  content = genai_types.Content(
+      parts=[genai_types.Part(text="Just thinking.", thought=True)]
+  )
+  assert get_text_from_content(content) == ""
+
+
+def test_get_text_from_content_excludes_thoughts_from_intermediate_events():
+  """Tests thought parts are excluded from intermediate responses too."""
+  invocation = Invocation(
+      user_content=genai_types.Content(parts=[genai_types.Part(text="user")]),
+      intermediate_data=InvocationEvents(
+          invocation_events=[
+              InvocationEvent(
+                  author="agent",
+                  content=genai_types.Content(
+                      parts=[
+                          genai_types.Part(
+                              text="Considering options.", thought=True
+                          ),
+                          genai_types.Part(text="Let me check."),
+                      ]
+                  ),
+              ),
+          ]
+      ),
+      final_response=genai_types.Content(
+          parts=[genai_types.Part(text="Done.")]
+      ),
+  )
+
+  assert (
+      get_text_from_content(
+          invocation, include_intermediate_responses_in_final=True
+      )
+      == "Let me check.\nDone."
+  )
+
+
+def test_get_text_parts_with_none():
+  """Tests get_text_parts returns an empty list for None."""
+  assert get_text_parts(None) == []
+
+
+def test_get_text_parts_with_no_parts():
+  """Tests get_text_parts returns an empty list when parts is None."""
+  assert get_text_parts(genai_types.Content(parts=None)) == []
+
+
+def test_get_text_parts_excludes_thoughts():
+  """Tests get_text_parts excludes thought parts and non-text parts."""
+  content = genai_types.Content(
+      parts=[
+          genai_types.Part(text="Thinking...", thought=True),
+          genai_types.Part(text="Hello"),
+          genai_types.Part(
+              function_call=genai_types.FunctionCall(name="test_func")
+          ),
+          genai_types.Part(text="World"),
+      ]
+  )
+  assert get_text_parts(content) == ["Hello", "World"]
+
+
 def test_get_text_from_content_with_intermediate_data_full_response():
   invocation = Invocation(
       user_content=genai_types.Content(parts=[genai_types.Part(text="user")]),
diff --git a/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py b/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
index 27ed2d7ed17..9e398dc7105 100644
--- a/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
+++ b/tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py
@@ -112,6 +112,29 @@ def test_format_auto_rater_prompt_with_basic_invocation(
   ) in prompt
 
 
+def test_format_auto_rater_prompt_excludes_thought_parts(
+    evaluator: RubricBasedFinalResponseQualityV1Evaluator,
+):
+  """Tests format_auto_rater_prompt excludes thought parts from the response."""
+  invocation = Invocation(
+      user_content=genai_types.Content(
+          parts=[genai_types.Part(text="User input here.")]
+      ),
+      final_response=genai_types.Content(
+          parts=[
+              genai_types.Part(
+                  text="Considering how to respond.", thought=True
+              ),
+              genai_types.Part(text="Final agent response."),
+          ]
+      ),
+  )
+  prompt = evaluator.format_auto_rater_prompt(invocation, None)
+
+  assert "Considering how to respond." not in prompt
+  assert "Final agent response." in prompt
+
+
 def test_format_auto_rater_prompt_with_app_details(
     evaluator: RubricBasedFinalResponseQualityV1Evaluator,
 ):
diff --git a/tests/unittests/evaluation/test_rubric_based_multi_turn_trajectory_evaluator.py b/tests/unittests/evaluation/test_rubric_based_multi_turn_trajectory_evaluator.py
index e926d0b7203..ed94a18c589 100644
--- a/tests/unittests/evaluation/test_rubric_based_multi_turn_trajectory_evaluator.py
+++ b/tests/unittests/evaluation/test_rubric_based_multi_turn_trajectory_evaluator.py
@@ -285,6 +285,86 @@ async def test_app_details_instructions_and_tools(self, evaluator):
     assert "transfer_funds" in evaluator._formatted_tools
     assert "Transfer money between accounts." in evaluator._formatted_tools
 
+  @pytest.mark.asyncio
+  async def test_user_turn_excludes_thought_parts(self, evaluator):
+    """Tests that a thought part in the user turn is excluded from dialogue."""
+    invocations = [
+        Invocation(
+            user_content=genai_types.Content(
+                parts=[
+                    genai_types.Part(text="Considering...", thought=True),
+                    genai_types.Part(text="Hello"),
+                ]
+            ),
+            final_response=genai_types.Content(
+                parts=[genai_types.Part(text="Hi there!")]
+            ),
+            invocation_id="agent1",
+            rubrics=_RUBRICS,
+        ),
+    ]
+    evaluator._assemble_dialogue_history(invocations)
+
+    assert "Considering..." not in evaluator._formatted_dialogue
+    assert "USER TURN 1: Hello" in evaluator._formatted_dialogue
+
+  @pytest.mark.asyncio
+  async def test_intermediate_event_excludes_thought_parts(self, evaluator):
+    """Tests that a thought part in an intermediate event is excluded."""
+    intermediate_data = InvocationEvents(
+        invocation_events=[
+            InvocationEvent(
+                author="banking_agent",
+                content=genai_types.Content(
+                    parts=[
+                        genai_types.Part(
+                            text="Let me check the balance.", thought=True
+                        ),
+                        genai_types.Part(text="Checking your balance."),
+                    ]
+                ),
+            ),
+        ]
+    )
+    invocations = [
+        _make_invocation(
+            user_text="What is my balance?",
+            agent_text="Your balance is $100.",
+            invocation_id="banking_agent",
+            rubrics=_RUBRICS,
+            intermediate_data=intermediate_data,
+        ),
+    ]
+    evaluator._assemble_dialogue_history(invocations)
+
+    assert "Let me check the balance." not in evaluator._formatted_dialogue
+    assert "Checking your balance." in evaluator._formatted_dialogue
+
+  @pytest.mark.asyncio
+  async def test_final_response_excludes_thought_parts(self, evaluator):
+    """Tests that a thought part in the final response is excluded."""
+    invocations = [
+        Invocation(
+            user_content=genai_types.Content(
+                parts=[genai_types.Part(text="Hello")]
+            ),
+            final_response=genai_types.Content(
+                parts=[
+                    genai_types.Part(
+                        text="Deciding how to respond.", thought=True
+                    ),
+                    genai_types.Part(text="Hi there!"),
+                ]
+            ),
+            invocation_id="agent1",
+            rubrics=_RUBRICS,
+        ),
+    ]
+    evaluator._assemble_dialogue_history(invocations)
+
+    assert "Deciding how to respond." not in evaluator._formatted_dialogue
+    assert "AGENT (agent) TURN 1: Hi there!" in evaluator._formatted_dialogue
+
   @pytest.mark.asyncio
   async def test_invocation_without_user_content(self, evaluator):
     """Tests that invocations with no user text parts are handled gracefully."""
diff --git a/tests/unittests/evaluation/test_rubric_based_tool_use_quality_v1.py b/tests/unittests/evaluation/test_rubric_based_tool_use_quality_v1.py
index 64d5ff7a313..b340a7bcb73 100644
--- a/tests/unittests/evaluation/test_rubric_based_tool_use_quality_v1.py
+++ b/tests/unittests/evaluation/test_rubric_based_tool_use_quality_v1.py
@@ -79,6 +79,24 @@ def test_format_auto_rater_prompt_with_basic_invocation(
   assert "<response>\nNo intermediate steps were taken.\n</response>" in prompt
 
 
+def test_format_auto_rater_prompt_excludes_thought_parts(
+    evaluator: RubricBasedToolUseV1Evaluator,
+):
+  """Tests format_auto_rater_prompt excludes thought parts from user input."""
+  invocation = Invocation(
+      user_content=genai_types.Content(
+          parts=[
+              genai_types.Part(text="Considering the request.", thought=True),
+              genai_types.Part(text="User input here."),
+          ]
+      ),
+  )
+  prompt = evaluator.format_auto_rater_prompt(invocation, None)
+
+  assert "Considering the request." not in prompt
+  assert "User input here." in prompt
+
+
 def test_format_auto_rater_prompt_with_invocation_rubrics_only():
   """Tests prompt formatting when rubrics are defined on the invocation."""
   judge_model_options = JudgeModelOptions(
diff --git a/tests/unittests/evaluation/test_vertex_ai_eval_facade.py b/tests/unittests/evaluation/test_vertex_ai_eval_facade.py
index 285f8c70629..11df6c57cc8 100644
--- a/tests/unittests/evaluation/test_vertex_ai_eval_facade.py
+++ b/tests/unittests/evaluation/test_vertex_ai_eval_facade.py
@@ -36,6 +36,22 @@
 vertexai_types = vertexai.types
 
 
+def test_get_text_excludes_thought_parts(mocker):
+  """Tests _get_text excludes thought parts from the extracted text."""
+  mocker.patch("google.adk.dependencies.vertexai.vertexai.Client")
+  evaluator = _SingleTurnVertexAiEvalFacade(
+      threshold=0.8, metric_name=vertexai_types.PrebuiltMetric.COHERENCE
+  )
+  content = genai_types.Content(
+      parts=[
+          genai_types.Part(text="Consider the options.", thought=True),
+          genai_types.Part(text="Paris"),
+      ]
+  )
+
+  assert evaluator._get_text(content) == "Paris"
+
+
 class TestSingleTurnVertexAiEvalFacade:
   """A class to help organize "patch" that are applicable to all tests."""