diff --git a/.github/CHANGELOG.md b/.github/CHANGELOG.md
index ad0271b5cf..7b591fe79b 100644
--- a/.github/CHANGELOG.md
+++ b/.github/CHANGELOG.md
@@ -49,6 +49,9 @@
 
 ### Improvements
 
+* Update the `lightning.tensor` Python layer unit tests, as `lightning.tensor` cannot be cleaned up like other state-vector devices because the data is attached to the graph. It is recommended to use one device per circuit for `lightning.tensor`.
+  [(#971)](https://github.com/PennyLaneAI/pennylane-lightning/pull/971)
+
 *  Fix PTM stable-latest.
   [(#961)](https://github.com/PennyLaneAI/pennylane-lightning/pull/961)
 
@@ -129,6 +132,9 @@
 
 ### Documentation
 
+* Update `lightning.tensor` usage suggestions.
+  [(#971)](https://github.com/PennyLaneAI/pennylane-lightning/pull/971)
+
 * Update ``lightning.tensor`` documentation to include all the new features added since pull request #756. The new features are: 1, Finite-shot measurements; 2. Expval-base quantities; 3. Support for ``qml.state()`` and ``qml.stateprep()``; 4. Support for all gates support via Matrix Product Operator (MPO).
   [(#909)](https://github.com/PennyLaneAI/pennylane-lightning/pull/909)
 
diff --git a/doc/lightning_tensor/device.rst b/doc/lightning_tensor/device.rst
index cf125c07cc..7a364f0245 100644
--- a/doc/lightning_tensor/device.rst
+++ b/doc/lightning_tensor/device.rst
@@ -53,6 +53,7 @@ Check out the :doc:`/lightning_tensor/installation` guide for more information.
 
 .. seealso:: `DefaultTensor <https://docs.pennylane.ai/en/latest/code/api/pennylane.devices.default_tensor.DefaultTensor.html>`__ for a CPU only tensor network simulator device.
 
+Note that as `lightning.tensor` cannot be cleaned up like other state-vector devices because the data is attached to the graph. It is recommended to create a new ``lightning.tensor`` device per circuit to ensure resources are correctly handled.
 Operations and observables support
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
diff --git a/tests/test_gates.py b/tests/test_gates.py
index da414c3789..c3d361836c 100644
--- a/tests/test_gates.py
+++ b/tests/test_gates.py
@@ -104,7 +104,7 @@ def test_gate_unitary_correct(op, op_name):
 
     if op_name == "QubitUnitary" and device_name == "lightning.tensor":
         pytest.skip(
-            "Skipping QubitUnitary on lightning.tensor. It can't be decomposed into 1-wire or 2-wire gates"
+            "Skipping QubitUnitary on lightning.tensor. as `lightning.tensor` cannot be cleaned up like other state-vector devices because the data is attached to the graph. It is recommended to use one device per circuit for `lightning.tensor`."
         )
 
     dev = qml.device(device_name, wires=wires)
@@ -153,6 +153,40 @@ def output(input):
     assert np.allclose(unitary, unitary_expected)
 
 
+@pytest.mark.parametrize("op_name", ld.operations)
+def test_gate_unitary_correct_lt(op, op_name):
+    """Test if lightning device correctly applies gates by reconstructing the unitary matrix and
+    comparing to the expected version"""
+
+    if op_name in ("BasisState", "QubitStateVector", "StatePrep"):
+        pytest.skip("Skipping operation because it is a state preparation")
+    if op == None:
+        pytest.skip("Skipping operation.")
+
+    wires = len(op[2]["wires"])
+
+    if wires == 1 and device_name == "lightning.tensor":
+        pytest.skip("Skipping single wire device on lightning.tensor.")
+
+    unitary = np.zeros((2**wires, 2**wires), dtype=np.complex128)
+
+    for i, input in enumerate(itertools.product([0, 1], repeat=wires)):
+        dev = qml.device(device_name, wires=wires)
+
+        @qml.qnode(dev)
+        def output(input):
+            qml.BasisState(input, wires=range(wires))
+            op[0](*op[1], **op[2])
+            return qml.state()
+
+        out = output(np.array(input))
+        unitary[:, i] = out
+
+    unitary_expected = qml.matrix(op[0](*op[1], **op[2]))
+
+    assert np.allclose(unitary, unitary_expected)
+
+
 @pytest.mark.parametrize("op_name", ld.operations)
 def test_inverse_unitary_correct(op, op_name):
     """Test if lightning device correctly applies inverse gates by reconstructing the unitary matrix