diff --git a/docs/changelog.rst b/docs/changelog.rst
index d725cb25..70b7ee80 100644
--- a/docs/changelog.rst
+++ b/docs/changelog.rst
@@ -2,11 +2,12 @@ Changelog
 ~~~~~~~~~
 
 
-0.44.0 (Unreleased)
--------------------
+Unreleased
+----------
 
 * Implement crosstalk noise model for AerBackend.
-
+* Don't include ``SimplifyInitial`` in default passes; instead make it an option
+  to ``process_circuits()``.
 
 0.44.0 (September 2023)
 -----------------------
diff --git a/docs/intro.txt b/docs/intro.txt
index b256d0f5..64d50456 100644
--- a/docs/intro.txt
+++ b/docs/intro.txt
@@ -170,22 +170,17 @@ Every :py:class:`Backend` in pytket has its own ``default_compilation_pass`` met
      - self.rebase_pass [2]
      - `SynthesiseTket <https://cqcl.github.io/tket/pytket/api/passes.html#pytket.passes.SynthesiseTket>`_
    * -
-     - `SimplifyInitial <https://cqcl.github.io/tket/pytket/api/passes.html#pytket.passes.SimplifyInitial>`_ [4]
      - self.rebase_pass [2]
    * - 
      -
      - `RemoveRedundancies <https://cqcl.github.io/tket/pytket/api/passes.html#pytket.passes.RemoveRedundancies>`_ 
-   * - 
-     - 
-     - `SimplifyInitial <https://cqcl.github.io/tket/pytket/api/passes.html#pytket.passes.SimplifyInitial>`_ [4]
 
 * [1] If no value is specified then ``optimisation_level`` defaults to a value of 2.
 * [2] self.rebase_pass is a rebase to the gateset supported by the backend, For IBM quantum devices that is {X, SX, Rz, CX}.
 * [3] Here :py:class:`CXMappingPass` maps program qubits to the architecture using a `NoiseAwarePlacement <https://cqcl.github.io/tket/pytket/api/placement.html#pytket.placement.NoiseAwarePlacement>`_
-* [4] :py:class:`SimplifyInitial` has arguments ``allow_classical=False`` and ``create_all_qubits=True``.
 
 
-**Note:** The ``default_compilation_pass`` for :py:class:`AerBackend` is the same as above except it doesn't use :py:class:`SimplifyInitial`.
+**Note:** The ``default_compilation_pass`` for :py:class:`AerBackend` is the same as above.
 
 
 Backend Predicates
diff --git a/pytket/extensions/qiskit/backends/ibm.py b/pytket/extensions/qiskit/backends/ibm.py
index 5db91cd0..f2090074 100644
--- a/pytket/extensions/qiskit/backends/ibm.py
+++ b/pytket/extensions/qiskit/backends/ibm.py
@@ -428,10 +428,6 @@ def default_compilation_pass(
 
         if self._supports_rz:
             passlist.extend([self.rebase_pass(), RemoveRedundancies()])
-        if optimisation_level > 0:
-            passlist.append(
-                SimplifyInitial(allow_classical=False, create_all_qubits=True)
-            )
         return SequencePass(passlist)
 
     @property
@@ -451,7 +447,13 @@ def process_circuits(
     ) -> List[ResultHandle]:
         """
         See :py:meth:`pytket.backends.Backend.process_circuits`.
-        Supported kwargs: `postprocess`.
+
+        Supported `kwargs`:
+        - `postprocess`: apply end-of-circuit simplifications and classical
+          postprocessing to improve fidelity of results (bool, default False)
+        - `simplify_initial`: apply the pytket ``SimplifyInitial`` pass to improve
+          fidelity of results assuming all qubits initialized to zero (bool, default
+          False)
         """
         circuits = list(circuits)
 
@@ -464,6 +466,9 @@ def process_circuits(
         handle_list: List[Optional[ResultHandle]] = [None] * len(circuits)
         circuit_batches, batch_order = _batch_circuits(circuits, n_shots_list)
 
+        postprocess = kwargs.get("postprocess", False)
+        simplify_initial = kwargs.get("simplify_initial", False)
+
         batch_id = 0  # identify batches for debug purposes only
         for (n_shots, batch), indices in zip(circuit_batches, batch_order):
             for chunk in itertools.zip_longest(
@@ -475,8 +480,6 @@ def process_circuits(
                 if valid_check:
                     self._check_all_circuits(batch_chunk)
 
-                postprocess = kwargs.get("postprocess", False)
-
                 qcs, ppcirc_strs = [], []
                 for tkc in batch_chunk:
                     if postprocess:
@@ -484,6 +487,10 @@ def process_circuits(
                         ppcirc_rep = ppcirc.to_dict()
                     else:
                         c0, ppcirc_rep = tkc, None
+                    if simplify_initial:
+                        SimplifyInitial(
+                            allow_classical=False, create_all_qubits=True
+                        ).apply(c0)
                     qcs.append(tk_to_qiskit(c0))
                     ppcirc_strs.append(json.dumps(ppcirc_rep))
                 if self._MACHINE_DEBUG:
diff --git a/tests/backend_test.py b/tests/backend_test.py
index f9dc3f29..1d10da3c 100644
--- a/tests/backend_test.py
+++ b/tests/backend_test.py
@@ -1035,10 +1035,6 @@ def test_compilation_correctness(perth_backend: IBMQBackend) -> None:
         m_ini = lift_perm(ini)
         m_inv_fin = lift_perm(inv_fin)
 
-        # Note that we do not expect unitary equivalence, because the pass includes
-        # SimplifyInitial which may remove initial gates that do not affect the final
-        # result. However the first columns of the unitaries (i.e. the final
-        # statevectors arising from an initial all-zero state) should match.
         assert compare_statevectors(u[:, 0], (m_inv_fin @ compiled_u @ m_ini)[:, 0])