diff --git a/torch_npu/_inductor/codegen/triton.py b/torch_npu/_inductor/codegen/triton.py
index 201136d3f6f072ac01597b46b31c50f7df24dbc8..ade172451312e6cdff9c298a65b7c96acd26fafe 100644
--- a/torch_npu/_inductor/codegen/triton.py
+++ b/torch_npu/_inductor/codegen/triton.py
@@ -1379,7 +1379,7 @@ class NPUIndexTritonKernel(TritonKernel):
         result_var.mask_vars = indexing.mask_vars  # type: ignore[assignment]
 
         if append_broadcast and append_broadcast != '[]':
-            line = f"tl.broadcast_to({result_var}, {append_broadcast})"
+            line = f"tl.reshape({result_var}, {append_broadcast})"
             result_var = self.cse.generate(load_buffer, line, dtype=dtype)
         # triton can handle broadcast
         elif index_analyze.need_permute: