diff --git a/satpy/tests/writer_tests/test_ninjogeotiff.py b/satpy/tests/writer_tests/test_ninjogeotiff.py
index 31fcb25efc..a9b283cf11 100644
--- a/satpy/tests/writer_tests/test_ninjogeotiff.py
+++ b/satpy/tests/writer_tests/test_ninjogeotiff.py
@@ -20,6 +20,7 @@
 import datetime
 import logging
 import os
+from unittest.mock import Mock
 
 import dask.array as da
 import numpy as np
@@ -247,6 +248,12 @@ def test_image_small_arctic_P(test_area_tiny_stereographic_wgs84):
             "start_time": datetime.datetime(2027, 8, 2, 8, 20),
             "area": test_area_tiny_stereographic_wgs84,
             "mode": "P"})
+    # simulate an enhancement history such as palettize may add
+    arr.attrs["enhancement_history"] = [
+            {"scale": np.float64(0.01),
+            "offset": np.float64(0.0),
+            "colormap": Mock()}]
+
     return to_image(arr)
 
 
@@ -577,8 +584,8 @@ def test_write_and_read_file_P(test_image_small_arctic_P, tmp_path):
         test_image_small_arctic_P,
         filename=fn,
         fill_value=255,
-        PhysicUnit="N/A",
-        PhysicValue="N/A",
+        PhysicUnit="satdata",
+        PhysicValue="satdata",
         SatelliteNameID=6400014,
         ChannelID=900015,
         DataType="PPRN",
@@ -591,8 +598,8 @@ def test_write_and_read_file_P(test_image_small_arctic_P, tmp_path):
     tgs = src.tags()
     assert tgs["ninjo_FileName"] == fn
     assert tgs["ninjo_DataSource"] == "dowsing rod"
-    assert "ninjo_Gradient" not in tgs
-    assert "ninjo_AxisIntercept" not in tgs
+    assert tgs["ninjo_Gradient"] == "1.0"
+    assert tgs["ninjo_AxisIntercept"] == "0.0"
 
 
 def test_write_and_read_file_units(
diff --git a/satpy/writers/ninjogeotiff.py b/satpy/writers/ninjogeotiff.py
index 5f88cc52ed..1d5cfb69ac 100644
--- a/satpy/writers/ninjogeotiff.py
+++ b/satpy/writers/ninjogeotiff.py
@@ -74,11 +74,13 @@
 NinJo has a functionality to read the corresponding quantity (example:
 brightness temperature or reflectance).  To make this possible, the writer
 adds the tags ``Gradient`` and ``AxisIntercept``.  Those tags are added if
-and only if the image has mode ``L`` or ``LA`` and ``PhysicUnit`` is not set
+and only if the image has mode ``L``, ``P``, or ``LA`` and ``PhysicUnit`` is not set
 to ``"N/A"``.  In other words, to suppress those tags for images with mode
 ``L`` or ``LA`` (for example, for the composite ``vis_with_ir``, where the
 physical interpretation of individual pixels is lost), one should set
 ``PhysicUnit`` to ``"N/A"``, ``"n/a"``, ``"1"``, or ``""`` (empty string).
+If the image has mode ``P``, ``Gradient`` is set to ``1.0`` and ``AxisIntercept``
+to ``0.0`` (as expected by NinJo).
 """
 
 import copy
@@ -204,11 +206,22 @@ def save_image(  # noqa: D417
             overviews_minsize=overviews_minsize,
             overviews_resampling=overviews_resampling,
             tags={**(tags or {}), **ninjo_tags},
-            scale_offset_tags=(self.scale_offset_tag_names
-                               if self._check_include_scale_offset(image, PhysicUnit)
-                               else None),
+            scale_offset_tags=self._get_scale_offset_tags(image, PhysicUnit),
             **gdal_opts)
 
+    def _get_scale_offset_tags(self, image, unit):
+        """Get scale offset tags (tuple or dict)."""
+        if self._check_include_scale_offset(image, unit):
+            # image.mode cannot be trusted https://github.com/pytroll/satpy/issues/2300
+            try:
+                mod = image.data.attrs["mode"]
+            except KeyError:
+                mod = image.mode
+            if mod == "P":
+                return dict(zip(self.scale_offset_tag_names, (1, 0)))
+            return self.scale_offset_tag_names
+        return None  # explicit is better than implicit
+
     def _fix_units(self, image, quantity, unit):
         """Adapt units between °C and K.
 
@@ -236,7 +249,7 @@ def _fix_units(self, image, quantity, unit):
 
     def _check_include_scale_offset(self, image, unit):
         """Check if scale-offset tags should be included."""
-        if image.mode.startswith("L") and unit.lower() not in ("n/a", "1", ""):
+        if image.mode[0] in "LP" and unit.lower() not in ("n/a", "1", ""):
             return True
         return False