diff --git a/agentplatform/_genai/_evals_data_converters.py b/agentplatform/_genai/_evals_data_converters.py
index 2934ba92ea..e4ec6a3bf3 100644
--- a/agentplatform/_genai/_evals_data_converters.py
+++ b/agentplatform/_genai/_evals_data_converters.py
@@ -230,6 +230,18 @@ def convert(self, raw_data: list[dict[str, Any]]) -> types.EvaluationDataset:
                     "Invalid prompt type for case %s: %s" % (i, type(prompt_data))
                 )
 
+            if (
+                isinstance(conversation_history_data, str)
+                and conversation_history_data.strip()
+            ):
+                try:
+                    conversation_history_data = json.loads(conversation_history_data)
+                except json.JSONDecodeError:
+                    logger.warning(
+                        "Could not decode conversation_history JSON string for case %s.",
+                        eval_case_id,
+                    )
+
             conversation_history: Optional[list[types.evals.Message]] = None
             if isinstance(conversation_history_data, list):
                 conversation_history = []
@@ -242,6 +254,11 @@ def convert(self, raw_data: list[dict[str, Any]]) -> types.EvaluationDataset:
                             )
                         )
                     elif isinstance(content, dict):
+                        if isinstance(content.get("content"), str):
+                            content = {
+                                "role": content.get("role"),
+                                "parts": [{"text": content["content"]}],
+                            }
                         try:
                             validated_content = genai_types.Content.model_validate(
                                 content
diff --git a/tests/unit/agentplatform/genai/test_evals.py b/tests/unit/agentplatform/genai/test_evals.py
index 314f0859d4..1072e75d17 100644
--- a/tests/unit/agentplatform/genai/test_evals.py
+++ b/tests/unit/agentplatform/genai/test_evals.py
@@ -45,6 +45,9 @@
     types as agentplatform_genai_types,
 )
 from agentplatform._genai.types import common as common_types
+from vertexai._genai import (
+    _evals_data_converters as vertexai_evals_data_converters,
+)
 from vertexai._genai import types as vertexai_genai_types
 from google.genai import client
 from google.genai import errors as genai_errors
@@ -5867,6 +5870,68 @@ def test_convert_with_conversation_history(self):
             eval_case.conversation_history[1].content.parts[0].text == "Old model msg"
         )
 
+    @pytest.mark.parametrize(
+        "converters",
+        [_evals_data_converters, vertexai_evals_data_converters],
+        ids=["agent_platform", "vertexai"],
+    )
+    @pytest.mark.parametrize(
+        "history, model_role",
+        [
+            (
+                '[{"role": "user", "parts": [{"text": "Old user msg"}]},'
+                ' {"role": "model", "parts": [{"text": "Old model msg"}]}]',
+                "model",
+            ),
+            (
+                '[{"role": "user", "content": "Old user msg"},'
+                ' {"role": "assistant", "content": "Old model msg"}]',
+                "assistant",
+            ),
+            (
+                [
+                    {"role": "user", "content": "Old user msg"},
+                    {"role": "assistant", "content": "Old model msg"},
+                ],
+                "assistant",
+            ),
+        ],
+        ids=["json_string", "openai_json_string", "openai_dicts"],
+    )
+    def test_convert_parses_string_and_openai_conversation_history(
+        self, converters, history, model_role
+    ):
+        result_dataset = converters._FlattenEvalDataConverter().convert(
+            [
+                {
+                    "prompt": "Current prompt",
+                    "response": "A response",
+                    "conversation_history": history,
+                }
+            ]
+        )
+
+        assert [
+            (message.content.role, message.content.parts[0].text)
+            for message in result_dataset.eval_cases[0].conversation_history
+        ] == [("user", "Old user msg"), (model_role, "Old model msg")]
+
+    @pytest.mark.parametrize(
+        "history, warns",
+        [("not json", True), ("  ", False)],
+        ids=["invalid_json", "blank"],
+    )
+    def test_convert_skips_unparsable_conversation_history(
+        self, history, warns, caplog
+    ):
+        with caplog.at_level("WARNING", logger=_evals_data_converters.logger.name):
+            result_dataset = self.converter.convert(
+                [{"prompt": "p", "response": "r", "conversation_history": history}]
+            )
+
+        assert result_dataset.eval_cases[0].conversation_history is None
+        assert ("Could not decode conversation_history" in caplog.text) == warns
+
     def test_convert_with_conversation_history_column_name(self):
         """Tests that 'conversation_history' is accepted as a column name alias for 'history'."""
         raw_data_df = pd.DataFrame(
diff --git a/vertexai/_genai/_evals_data_converters.py b/vertexai/_genai/_evals_data_converters.py
index f293152cbe..1e7beb0af0 100644
--- a/vertexai/_genai/_evals_data_converters.py
+++ b/vertexai/_genai/_evals_data_converters.py
@@ -230,6 +230,18 @@ def convert(self, raw_data: list[dict[str, Any]]) -> types.EvaluationDataset:
                     "Invalid prompt type for case %s: %s" % (i, type(prompt_data))
                 )
 
+            if (
+                isinstance(conversation_history_data, str)
+                and conversation_history_data.strip()
+            ):
+                try:
+                    conversation_history_data = json.loads(conversation_history_data)
+                except json.JSONDecodeError:
+                    logger.warning(
+                        "Could not decode conversation_history JSON string for case %s.",
+                        eval_case_id,
+                    )
+
             conversation_history: Optional[list[types.evals.Message]] = None
             if isinstance(conversation_history_data, list):
                 conversation_history = []
@@ -242,6 +254,11 @@ def convert(self, raw_data: list[dict[str, Any]]) -> types.EvaluationDataset:
                             )
                         )
                     elif isinstance(content, dict):
+                        if isinstance(content.get("content"), str):
+                            content = {
+                                "role": content.get("role"),
+                                "parts": [{"text": content["content"]}],
+                            }
                         try:
                             validated_content = genai_types.Content.model_validate(
                                 content