diff --git a/README.md b/README.md
index 7bc27e3..f25729a 100644
--- a/README.md
+++ b/README.md
@@ -13,7 +13,13 @@ This downloads the latest shared `liblbug` binary (via upstream
 The Python package is installed directly from `src_py/`, so the standalone
 workflow no longer depends on `./build/ladybug`.
 
-Run tests with:
+In order to run tests you'll need to pull data first with:
+
+```bash
+git submodule update --init --recursive
+```
+
+Then run tests with:
 
 ```bash
 uv run pytest
diff --git a/src_cpp/py_query_result_converter.cpp b/src_cpp/py_query_result_converter.cpp
index 5401938..24ea7d6 100644
--- a/src_cpp/py_query_result_converter.cpp
+++ b/src_cpp/py_query_result_converter.cpp
@@ -78,6 +78,7 @@ void NPArrayWrapper::appendElement(Value* value) {
             ((int64_t*)dataBuffer)[numElements] =
                 Interval::getNanoseconds(value->getValue<interval_t>());
         } break;
+        case LogicalTypeID::JSON:
         case LogicalTypeID::STRING: {
             auto val = value->getValue<std::string>();
             py::str result(val);
@@ -167,6 +168,7 @@ py::dtype NPArrayWrapper::convertToArrayType(const LogicalType& type) {
     case LogicalTypeID::INTERVAL: {
         dtype = "timedelta64[ns]";
     } break;
+    case LogicalTypeID::JSON:
     case LogicalTypeID::DECIMAL:
     case LogicalTypeID::UNION:
     case LogicalTypeID::BLOB:
diff --git a/test/test_json.py b/test/test_json.py
index 924a6a5..4f83e4c 100644
--- a/test/test_json.py
+++ b/test/test_json.py
@@ -155,3 +155,87 @@ def test_to_json_python_param_with_homogeneous_list_uses_typed_binding(
 
     assert normalized_query == query
     assert normalized_parameters == parameters
+
+
+def test_get_as_df_json_scalar(conn_db_empty: ConnDB) -> None:
+    """
+    Scalar JSON values convert through get_as_df() as Python strings.
+
+    Covers non-null values, typed JSON nulls, and a mixed column in a single query.
+    """
+    conn, _ = conn_db_empty
+    df = conn.execute(
+        "UNWIND ["
+        'CAST(\'{"a": 1, "b": [2, 3]}\' AS JSON), '
+        "CAST(NULL AS JSON), "
+        "CAST('[1, 2, 3]' AS JSON)"
+        "] AS j RETURN j"
+    ).get_as_df()
+
+    assert str(df["j"].dtype) == "object"
+    assert df["j"].isna().tolist() == [False, True, False]
+
+    first = df["j"].iloc[0]
+    assert isinstance(first, str)
+    assert json.loads(first) == {"a": 1, "b": [2, 3]}
+
+    third = df["j"].iloc[2]
+    assert isinstance(third, str)
+    assert json.loads(third) == [1, 2, 3]
+
+
+def test_get_as_df_json_empty_result(conn_db_empty: ConnDB) -> None:
+    """
+    An empty result over a JSON column builds the column without crashing.
+
+    convertToArrayType() runs during NPArrayWrapper construction, before any
+    rows are iterated, so a zero-row JSON result is the minimal reproduction
+    for the original dtype-selection crash.
+    """
+    conn, _ = conn_db_empty
+    conn.execute("CREATE NODE TABLE t (id SERIAL PRIMARY KEY, data JSON)")
+
+    df = conn.execute("MATCH (n:t) RETURN n.data AS data").get_as_df()
+
+    assert len(df) == 0
+    assert str(df["data"].dtype) == "object"
+
+
+def test_get_as_df_json_extract(conn_db_empty: ConnDB) -> None:
+    """json_extract() produces a scalar JSON result that converts via get_as_df()."""
+    conn, _ = conn_db_empty
+    conn.execute("INSTALL json; LOAD json;")
+    conn.execute("CREATE NODE TABLE t (id SERIAL PRIMARY KEY, data JSON)")
+
+    data = {"name": {"first": "Alice", "last": "Smith"}}
+    conn.execute(
+        "CREATE (n:t {data: to_json($data)})",
+        parameters={"data": json.dumps(data)},
+    )
+
+    df = conn.execute(
+        "MATCH (n:t) RETURN json_extract(n.data, '$.name') AS name"
+    ).get_as_df()
+
+    assert str(df["name"].dtype) == "object"
+    val = df["name"].iloc[0]
+    assert isinstance(val, str)
+    assert json.loads(val) == {"first": "Alice", "last": "Smith"}
+
+
+def test_get_as_df_json_list(conn_db_empty: ConnDB) -> None:
+    """JSON[] (LIST of JSON) columns keep their existing pandas behavior."""
+    conn, _ = conn_db_empty
+    conn.execute("INSTALL json; LOAD json;")
+    conn.execute("CREATE NODE TABLE t (id SERIAL PRIMARY KEY, data JSON[])")
+
+    data = [{"x": 1}, {"x": 2}, {"x": 3}]
+    conn.execute("CREATE (n:t {data: $d})", parameters={"d": data})
+
+    df = conn.execute("MATCH (n:t) RETURN n.data AS data").get_as_df()
+
+    assert str(df["data"].dtype) == "object"
+    val = df["data"].iloc[0]
+    assert isinstance(val, list)
+    assert len(val) == 3
+    assert all(isinstance(e, str) for e in val)