diff --git a/csv_diff/__init__.py b/csv_diff/__init__.py
index 59a2eaf..b222da9 100644
--- a/csv_diff/__init__.py
+++ b/csv_diff/__init__.py
@@ -35,8 +35,13 @@ def load_json(fp, key=None):
     if key:
         keyfn = lambda r: r[key]
     else:
+        # Missing fields and explicit nulls have the same row identity,
+        # independent of the other rows' columns. Keep nested values intact.
         keyfn = lambda r: hashlib.sha1(
-            json.dumps(r, sort_keys=True).encode("utf8")
+            json.dumps(
+                {field: value for field, value in r.items() if value is not None},
+                sort_keys=True,
+            ).encode("utf8")
         ).hexdigest()
     return {keyfn(r): _simplify_json_row(r, common_keys) for r in raw_list}
 
diff --git a/tests/test_json_normalization.py b/tests/test_json_normalization.py
new file mode 100644
index 0000000..86b72eb
--- /dev/null
+++ b/tests/test_json_normalization.py
@@ -0,0 +1,102 @@
+from csv_diff import compare, load_json
+import io
+import json
+import pytest
+
+
+@pytest.mark.parametrize("key", [None, "id"])
+def test_missing_and_null_values_compare_equal(key):
+    previous = [{"id": 1}, {"id": 2, "name": "Cleo"}]
+    current = [{"id": 1, "name": None}, {"id": 2, "name": "Cleo"}]
+    result = compare(
+        load_json(io.StringIO(json.dumps(previous)), key=key),
+        load_json(io.StringIO(json.dumps(current)), key=key),
+    )
+    assert result == {
+        "added": [],
+        "removed": [],
+        "changed": [],
+        "columns_added": [],
+        "columns_removed": [],
+    }
+
+
+def test_equivalent_normalized_rows_are_deduplicated_without_key():
+    rows = [{"id": 1}, {"id": 1, "name": None}]
+    result = load_json(io.StringIO(json.dumps(rows)))
+    assert list(result.values()) == [{"id": 1, "name": None}]
+
+
+def test_missing_explicit_key_is_not_filled_with_null():
+    rows = [{"id": 1}, {"name": "Cleo"}]
+    with pytest.raises(KeyError):
+        load_json(io.StringIO(json.dumps(rows)), key="id")
+
+
+@pytest.mark.parametrize("value", [{"name": "Cleo"}, ["Cleo", None]])
+def test_nested_values_remain_distinct_from_json_strings(value):
+    rows = [{"value": value}, {"value": json.dumps(value)}]
+    result = load_json(io.StringIO(json.dumps(rows)))
+    assert len(result) == 2
+    assert list(result.values()) == [{"value": json.dumps(value)}] * 2
+
+
+def test_unrelated_column_changes_do_not_replace_identical_rows():
+    result = compare(
+        load_json(io.StringIO('[{"id": 1}, {"a": 0}]')),
+        load_json(io.StringIO('[{"id": 1}, {"b": 0}]')),
+    )
+    assert result == {
+        "added": [{"b": 0, "id": None}],
+        "removed": [{"a": 0, "id": None}],
+        "changed": [],
+        "columns_added": ["b"],
+        "columns_removed": ["a"],
+    }
+
+
+@pytest.mark.parametrize("value", [None, 0, False, "", [], {}])
+def test_new_column_preserves_non_null_row_differences(value):
+    previous = [{"id": 1}]
+    current = [{"id": 1, "extra": value}]
+    result = compare(
+        load_json(io.StringIO(json.dumps(previous))),
+        load_json(io.StringIO(json.dumps(current))),
+    )
+    displayed = json.dumps(value) if isinstance(value, (list, dict)) else value
+    assert result == {
+        "added": [] if value is None else [{"id": 1, "extra": displayed}],
+        "removed": [] if value is None else previous,
+        "changed": [],
+        "columns_added": ["extra"],
+        "columns_removed": [],
+    }
+
+
+def test_nested_null_fields_are_not_ignored():
+    result = compare(
+        load_json(io.StringIO('[{"value": {}}]')),
+        load_json(io.StringIO('[{"value": {"field": null}}]')),
+    )
+    assert result == {
+        "added": [{"value": '{"field": null}'}],
+        "removed": [{"value": "{}"}],
+        "changed": [],
+        "columns_added": [],
+        "columns_removed": [],
+    }
+
+
+def test_explicit_key_still_uses_last_duplicate_row():
+    rows = [{"id": 1, "value": "old"}, {"id": 1, "value": None}]
+    assert load_json(io.StringIO(json.dumps(rows)), key="id") == {
+        1: {"id": 1, "value": None}
+    }
+
+
+def test_empty_key_argument_still_uses_content_keys():
+    source = '[{"": "same", "value": 0}, {"": "same", "value": false}]'
+    result = load_json(io.StringIO(source), key="")
+    assert result == load_json(io.StringIO(source))
+    assert len(result) == 2
+    assert "same" not in result