diff --git a/followthemoney/proxy.py b/followthemoney/proxy.py
index 3c94216a7..4172ecfe7 100644
--- a/followthemoney/proxy.py
+++ b/followthemoney/proxy.py
@@ -86,7 +86,13 @@ def __init__(
             if key not in self.schema.properties:
                 continue
             if cleaned:
-                self.add(key, values, cleaned=cleaned)
+                # This does not call `self.add` as it might be called millions of times
+                # in some context and we want to avoid the performance overhead of doing so.
+                seen = set()
+                seen_add = seen.add
+                unique_values = [v for v in values if not (v in seen or seen_add(v))]
+                self._properties[key] = unique_values
+                self._size += sum([len(v) for v in unique_values])
             else:
                 self.add(key, values, quiet=True)
 
diff --git a/tests/test_proxy.py b/tests/test_proxy.py
index 3d82f0eda..febb8f2d3 100644
--- a/tests/test_proxy.py
+++ b/tests/test_proxy.py
@@ -315,3 +315,15 @@ def test_value_deduplication(self):
         assert proxy.get("name") == ["ACME, Inc."]
         proxy.add("name", "ACME, Inc.")
         assert proxy.get("name") == ["ACME, Inc."]
+
+    def test_value_deduplication_cleaned(self):
+        proxy = EntityProxy.from_dict(model, {
+            "id": "acme-inc",
+            "schema": "Company",
+            "properties": {
+                "name": ["ACME, Inc.", "ACME, Inc."],
+            },
+        }, cleaned=True)
+
+        assert proxy.get("name") == ["ACME, Inc."]
+        
\ No newline at end of file