diff --git a/tests/test_app.py b/tests/test_app.py
index 61fb083..f58fbfa 100644
--- a/tests/test_app.py
+++ b/tests/test_app.py
@@ -97,6 +97,10 @@ async def test_redaction_analyze(pdf_bytes: bytes):
         assert data["gap"]["w"] > 0
         assert data["font"]["name"]
         assert data["font"]["size"] > 0
+        assert "offset_x" in data
+        assert "offset_y" in data
+        assert isinstance(data["offset_x"], (int, float))
+        assert isinstance(data["offset_y"], (int, float))
 
 
 @pytest.mark.anyio
diff --git a/unredact/app.py b/unredact/app.py
index 1b1bbf9..68b750b 100644
--- a/unredact/app.py
+++ b/unredact/app.py
@@ -184,6 +184,15 @@ async def analyze_redaction(req: AnalyzeRequest):
     left_text = "".join(c.text for c in left_chars).rstrip()
     right_text = "".join(c.text for c in right_chars).lstrip()
 
+    # Compute initial offset guess: align end of left text with gap start
+    pil_font = font_match.to_pil_font()
+    if left_text:
+        left_rendered_width = pil_font.getlength(left_text)
+        offset_x = float(rx - left_rendered_width - best_line.x)
+    else:
+        offset_x = 0.0
+    offset_y = 0.0
+
     if left_text:
         lx = left_chars[0].x
         lw = (left_chars[-1].x + left_chars[-1].w) - lx
@@ -216,6 +225,8 @@ async def analyze_redaction(req: AnalyzeRequest):
             "text": best_line.text,
         },
         "chars": chars_json,
+        "offset_x": round(offset_x, 1),
+        "offset_y": round(offset_y, 1),
     }