diff --git a/Changelog.rst b/Changelog.rst
index e112ae4b38..b351b250dc 100644
--- a/Changelog.rst
+++ b/Changelog.rst
@@ -1,3 +1,14 @@
+Version NEXTVERSION
+--------------
+
+**2026-10-??**
+
+* Fix problems creating CFA variables with `cf.write` from
+  aggregation-promoted field ancillaries
+  (https://github.com/NCAS-CMS/cf-python/issues/971)
+
+----
+
 Version 3.21.0
 --------------
 
diff --git a/cf/aggregate.py b/cf/aggregate.py
index 149be58d67..f70c22db6a 100644
--- a/cf/aggregate.py
+++ b/cf/aggregate.py
@@ -3845,41 +3845,13 @@ def _create_hash_and_first_values(
 
         # ------------------------------------------------------------
         # Field ancillaries
+        #
+        # No need to calculate actual hash values for field
+        # ancillaries (which can be very slow), because they always
+        # get concatenated regardless of their values.
         # ------------------------------------------------------------
-        if donotchecknonaggregatingaxes:
-            for anc in m.field_anc.values():
-                anc["hash_value"] = (None,)
-        else:
-            for anc in m.field_anc.values():
-                key = anc["key"]
-                canonical_units = anc["units"]
-
-                field_anc = constructs[key]
-
-                c_axes = anc["axes"]
-                canonical_axes = anc["canonical_axes"]
-                if c_axes != canonical_axes:
-                    # Transpose the field ancillary so that it has the
-                    # canonical axis order
-                    iaxes = [c_axes.index(axis) for axis in canonical_axes]
-                    field_anc = field_anc.transpose(iaxes)
-
-                sort_indices, needs_sorting = _sort_indices(m, canonical_axes)
-
-                # Get the hash of the data array
-                h = _get_hfl(
-                    field_anc,
-                    canonical_units,
-                    sort_indices,
-                    needs_sorting,
-                    False,
-                    False,
-                    hfl_cache,
-                    rtol,
-                    atol,
-                )
-
-                anc["hash_value"] = (h,)
+        for anc in m.field_anc.values():
+            anc["hash_value"] = (None,)
 
         # ------------------------------------------------------------
         # Domain ancillaries
@@ -5048,6 +5020,14 @@ def _fix_promoted_field_ancillaries(output_meta, axes_aggregated):
             fa.squeeze(squeeze, inplace=True)
             fa_axes = [a for i, a in enumerate(fa_axes) if i not in squeeze]
 
+            # Note that this field ancillary can be written as a
+            # CF-netCDF aggregation variable. (The previous indexing
+            # and squeezing will likely have set the aggregation write
+            # status to False, but we know that promoted field
+            # ancillaries, which only contain a single broadcast
+            # value, are safe in this regard.)
+            fa.data._nc_set_aggregation_write_status(True)
+
             # Record the field ancillary as being able to be written
             # as a CF-netCDF aggregation 'value' variable
             fa.data._nc_set_aggregation_fragment_type("value")
diff --git a/cf/test/test_CFA.py b/cf/test/test_CFA.py
index 324fe38bf8..f2ee8fef34 100644
--- a/cf/test/test_CFA.py
+++ b/cf/test/test_CFA.py
@@ -181,7 +181,7 @@ def test_CFA_constructs(self):
         f = cf.read(tmpfile1, cfa_write="all")[0]
 
         # No constructs
-        cf.write(f, tmpfile2, cfa={"constructs": []})
+        cf.write(f, tmpfile2, cfa=[])
         nc = netCDF4.Dataset(tmpfile2, "r")
         for var in nc.variables.values():
             attrs = var.ncattrs()
@@ -311,6 +311,9 @@ def test_CFA_cfa(self):
         cf.write(f, tmpfile2, cfa="field")
         g = cf.read(tmpfile2, cfa_write="field")[0]
 
+        cf.write(f, tmpfile2, cfa=["field"])
+        g = cf.read(tmpfile2, cfa_write="field")[0]
+
         # Default of cfa="auto" - check that aggregation variable
         # gets written
         cf.write(g, cfa_file)
@@ -345,7 +348,7 @@ def test_CFA_cfa(self):
         nc.close()
 
         # Check bad values of cfa
-        for cfa in (False, True, (), []):
+        for cfa in (False, True):
             with self.assertRaises(ValueError):
                 cf.write(g, cfa_file, cfa=cfa)
 
diff --git a/cf/test/test_aggregate.py b/cf/test/test_aggregate.py
index e4684d6f87..27af7a2505 100644
--- a/cf/test/test_aggregate.py
+++ b/cf/test/test_aggregate.py
@@ -1,6 +1,7 @@
 import datetime
 import faulthandler
 import os
+import tempfile
 import unittest
 import warnings
 
@@ -10,6 +11,14 @@
 
 import cf
 
+n_tmpfiles = 1
+tmpfiles = [
+    tempfile.mkstemp("_test_aggregate.nc", dir=os.getcwd())[1]
+    for i in range(n_tmpfiles)
+]
+[tmpfile] = tmpfiles
+
+
 # To facilitate the testing of logging outputs (see test_aggregate_verbosity)
 log_name = __name__
 logger = cf.logging.getLogger(log_name)
@@ -345,6 +354,8 @@ def test_aggregate_promote_field_ancillaries(self):
         self.assertTrue((anc[:2] == "bar_a").all())
         self.assertTrue((anc[2:] == "bar_b").all())
 
+        cf.write(c, tmpfile, cfa="field_ancillary")
+
     def test_aggregate_cells(self):
         """Test the 'cells' keyword of cf.aggregate"""
         f = cf.example_field(0)