diff --git a/packages/gooddata-eval/src/gooddata_eval/core/agentic/alert_skill.py b/packages/gooddata-eval/src/gooddata_eval/core/agentic/alert_skill.py
index d1a6e22f5..6c5a55d67 100644
--- a/packages/gooddata-eval/src/gooddata_eval/core/agentic/alert_skill.py
+++ b/packages/gooddata-eval/src/gooddata_eval/core/agentic/alert_skill.py
@@ -39,6 +39,7 @@
 from gooddata_eval.core.models import (
     AgenticAssertionError,
     AgenticEvalOutcome,
+    ChatResult,
     LoopExit,
     ReasoningStepEvent,
     ToolCallEvent,
@@ -928,6 +929,18 @@ def render_alert_proposal(proposal: dict) -> str:
     return f"{cta}\n\nAlert proposal:\n{json.dumps(summary, indent=2, sort_keys=True)}"
 
 
+def render_turn_for_simulated_user(chat_result: ChatResult) -> str:
+    """Render the whole agent turn the simulated user answers: prose, every other part, and the last proposal.
+
+    A ``clarifyingQuestions`` part carries the options the agent asks the user to pick from, so the
+    simulated user cannot answer a question whose options it never reads.
+    """
+    chunks = [render_answer_text(chat_result)]
+    if chat_result.alert_proposals:
+        chunks.append(render_alert_proposal(chat_result.alert_proposals[-1]))
+    return "\n\n".join(chunk for chunk in chunks if chunk)
+
+
 def run_agentic_alert_skill(
     host: str,
     token: str,
@@ -1027,11 +1040,7 @@ def _run_once(conv_id: str) -> AlertRunResult:
                     alert_id_to_delete = alert_id
                     exit_reason = LoopExit.SUCCESS
                     break
-                response_text = (chat_result.text_response or "").strip()
-                if not response_text and chat_result.alert_proposals:
-                    response_text = render_alert_proposal(chat_result.alert_proposals[-1])
-                if not response_text:
-                    response_text = render_answer_text(chat_result)
+                response_text = render_turn_for_simulated_user(chat_result)
                 # Stop if agent gave a completely empty response (stuck)
                 if not response_text and not chat_result.tool_call_events:
                     exit_reason = LoopExit.AGENT_SILENT
diff --git a/packages/gooddata-eval/src/gooddata_eval/core/agentic/metric_skill.py b/packages/gooddata-eval/src/gooddata_eval/core/agentic/metric_skill.py
index e77832986..464cf592e 100644
--- a/packages/gooddata-eval/src/gooddata_eval/core/agentic/metric_skill.py
+++ b/packages/gooddata-eval/src/gooddata_eval/core/agentic/metric_skill.py
@@ -378,9 +378,7 @@ def _execute_single_metric_run(
                 metric_result = candidate
                 exit_reason = LoopExit.SUCCESS
                 break
-            response_text = (chat_result.text_response or "").strip()
-            if not response_text:
-                response_text = render_answer_text(chat_result)
+            response_text = render_answer_text(chat_result)
             if not response_text and not chat_result.tool_call_events:
                 exit_reason = LoopExit.AGENT_SILENT
                 break
diff --git a/packages/gooddata-eval/src/gooddata_eval/core/chat/render.py b/packages/gooddata-eval/src/gooddata_eval/core/chat/render.py
index ca3f3d278..dd3d924aa 100644
--- a/packages/gooddata-eval/src/gooddata_eval/core/chat/render.py
+++ b/packages/gooddata-eval/src/gooddata_eval/core/chat/render.py
@@ -26,10 +26,10 @@ def render_search_results(part: dict) -> str:
 
 
 def render_unhandled_part(part: dict) -> str:
-    """Best-effort rendering of a part gd-eval does not model, truncated to a sane size."""
+    """Render an unmodelled part, preserving questionnaires and capping other payloads."""
     ptype = str(part.get("type") or "unknown")
     body = json.dumps({k: v for k, v in part.items() if k != "type"}, ensure_ascii=False)
-    if len(body) > _MAX_UNHANDLED_PART_CHARS:
+    if ptype != "clarifyingQuestions" and len(body) > _MAX_UNHANDLED_PART_CHARS:
         body = body[:_MAX_UNHANDLED_PART_CHARS] + "… (truncated)"
     return f"[{ptype}]\n{body}"
 
diff --git a/packages/gooddata-eval/tests/test_agentic_alert_skill.py b/packages/gooddata-eval/tests/test_agentic_alert_skill.py
index b0ff5481d..09cfab387 100644
--- a/packages/gooddata-eval/tests/test_agentic_alert_skill.py
+++ b/packages/gooddata-eval/tests/test_agentic_alert_skill.py
@@ -23,6 +23,7 @@
     evaluate_agentic_alert_skill,
     generate_simulated_alert_response,
     render_alert_proposal,
+    render_turn_for_simulated_user,
     run_agentic_alert_skill,
 )
 from gooddata_eval.core.chat.sse_client import ChatError
@@ -821,15 +822,67 @@ def test_render_alert_proposal_falls_back_to_default_cta():
     assert render_alert_proposal({}).startswith("Should I create this alert?")
 
 
-def test_run_agentic_alert_skill_answers_proposal_only_confirmation_turn():
-    """GDAI-2032 regression: confirmation turn has no text part, only an alertProposal.
+_QUESTIONS_PART = {
+    "type": "clarifyingQuestions",
+    "questions": [
+        {
+            "text": "Which date?",
+            "control": {
+                "type": "singleChoice",
+                "options": [{"text": "{dataset/date}"}, {"text": "{dataset/created}"}],
+            },
+        },
+    ],
+}
 
-    Without the fallback the simulated user is handed an empty agent message, so the agent
-    never receives an explicit "yes" and create_metric_alert is never called.
-    """
+
+@pytest.mark.parametrize(
+    ("turn", "must_contain"),
+    [
+        ({"text_response": "Pick both.", "unhandled_parts": [_QUESTIONS_PART]}, ["Pick both.", "{dataset/date}"]),
+        ({"text_response": "Here it is.", "alert_proposals": [_PROPOSAL]}, ["Here it is.", "GREATER_THAN"]),
+        ({"alert_proposals": [_PROPOSAL]}, ["Should I create this alert?", "GREATER_THAN"]),
+        ({"text_response": "Which metric?"}, ["Which metric?"]),
+    ],
+    ids=["text_and_questions", "text_and_proposal", "proposal_only", "text_only"],
+)
+def test_simulated_user_reads_every_part_of_the_turn(turn, must_contain):
+    rendered = render_turn_for_simulated_user(ChatResult(**turn))
+    for fragment in must_contain:
+        assert fragment in rendered
+
+
+def test_simulated_user_reads_nothing_of_a_silent_turn():
+    assert render_turn_for_simulated_user(ChatResult()) == ""
+
+
+def test_run_agentic_alert_skill_passes_question_options_to_simulated_user() -> None:
+    client = MagicMock()
+    client.send_message.return_value = ChatResult(text_response="Pick one.", unhandled_parts=[_QUESTIONS_PART])
+    with _patched(client, simulated_reply="Use the created date.") as mock_sim:
+        run_agentic_alert_skill(
+            host="http://host",
+            token="tok",
+            workspace_id="ws1",
+            question="Create an alert",
+            expected_output={"operator": "GREATER_THAN", "threshold": 500},
+            k=1,
+            max_iterations=2,
+            initial_conversation_id="conv-1",
+        )
+
+    agent_message = mock_sim.call_args.args[0]
+    assert "Pick one." in agent_message
+    assert "{dataset/date}" in agent_message
+    assert "{dataset/created}" in agent_message
+
+
+@pytest.mark.parametrize("text_response", [None, "Here it is."])
+def test_run_agentic_alert_skill_answers_proposal_confirmation_turn(text_response: str | None) -> None:
+    """Confirmation keeps the proposal fields visible alongside any prose."""
     proposal_turn = ChatResult.model_validate(
         {
-            "text_response": None,
+            "text_response": text_response,
             "alertProposals": [_PROPOSAL],
             "tool_call_events": [
                 {"functionName": "prepare_metric_alert_proposal", "functionArguments": "{}", "result": None}
@@ -866,6 +919,8 @@ def test_run_agentic_alert_skill_answers_proposal_only_confirmation_turn():
         )
 
     agent_message = mock_sim.call_args.args[0]
+    if text_response:
+        assert text_response in agent_message
     assert "Should I create this alert?" in agent_message
     assert "admin@gooddata.com" in agent_message
     assert summary.best.eval.alert_created is True
diff --git a/packages/gooddata-eval/tests/test_agentic_metric_skill.py b/packages/gooddata-eval/tests/test_agentic_metric_skill.py
index 65f23a39f..a86a2e9c3 100644
--- a/packages/gooddata-eval/tests/test_agentic_metric_skill.py
+++ b/packages/gooddata-eval/tests/test_agentic_metric_skill.py
@@ -331,6 +331,42 @@ def test_run_agentic_metric_skill_creates_conversation(monkeypatch):
     mock_client.close.assert_called_once()
 
 
+def test_run_agentic_metric_skill_passes_question_options_to_simulated_user() -> None:
+    client = _client()
+    client.send_message.return_value = ChatResult(
+        text_response="Pick one.",
+        unhandled_parts=[
+            {
+                "type": "clarifyingQuestions",
+                "questions": [
+                    {
+                        "text": "Which metric?",
+                        "control": {
+                            "type": "singleChoice",
+                            "options": [{"text": "{metric/revenue}"}, {"text": "{metric/profit}"}],
+                        },
+                    }
+                ],
+            }
+        ],
+    )
+    with _patched(client, simulated_reply="Use profit.") as (mock_sim, _):
+        run_agentic_metric_skill(
+            host="http://host",
+            token="tok",
+            workspace_id="ws1",
+            question="Create a metric",
+            expected_output={"maql": "SELECT {metric/profit}"},
+            k=1,
+            max_iterations=2,
+        )
+
+    agent_message = mock_sim.call_args.args[0]
+    assert "Pick one." in agent_message
+    assert "{metric/revenue}" in agent_message
+    assert "{metric/profit}" in agent_message
+
+
 def test_run_agentic_metric_skill_closes_client_on_no_result():
     mock_client = _client()
     mock_client.send_message.return_value = ChatResult.model_validate(
diff --git a/packages/gooddata-eval/tests/test_chat_render.py b/packages/gooddata-eval/tests/test_chat_render.py
index 29904158f..17834e2e7 100644
--- a/packages/gooddata-eval/tests/test_chat_render.py
+++ b/packages/gooddata-eval/tests/test_chat_render.py
@@ -155,6 +155,39 @@ def test_a_huge_unmodelled_part_is_truncated():
     assert len(rendered) < 3_000
 
 
+def test_long_questionnaire_keeps_every_offered_option() -> None:
+    questions: list[dict] = []
+    for question_index in range(4):
+        options = [
+            {"text": "{dashboard/00000000-0000-4000-8000-" + f"{question_index * 6 + option_index:012d}" + "}"}
+            for option_index in range(6)
+        ]
+        questions.append(
+            {
+                "text": (
+                    f"Which dashboard should receive the alert for metric {question_index}, considering the department, "
+                    "reporting period, audience, and the business context in my original request? "
+                    "For example: choose the closest match."
+                ),
+                "control": {"type": "singleChoice", "options": options},
+            }
+        )
+    result = parse_sse_lines(
+        _multipart_lines(
+            {"type": "text", "text": "Pick one for each metric."},
+            {"type": "clarifyingQuestions", "questions": questions},
+        )
+    )
+
+    rendered = render_answer_text(result)
+
+    assert "Pick one for each metric." in rendered
+    for question in questions:
+        assert question["text"] in rendered
+        for option in question["control"]["options"]:
+            assert option["text"] in rendered
+
+
 @pytest.mark.parametrize("ptype", ["dashboard", "dashboardPatch", "report", "kda", "whatIf", "clarifyingQuestions"])
 def test_no_known_part_type_is_silently_dropped(ptype):
     result = parse_sse_lines(_multipart_lines({"type": ptype, "payload": {"a": 1}}))