diff --git a/dist.py b/dist.py
index fd018b3..5d90f82 100755
--- a/dist.py
+++ b/dist.py
@@ -401,13 +401,14 @@ def build_linux(
                 WHEEL_LINUX_CONFIGS[cuda_version]['system_packages']
 
             if kind == 'cuda':
-                long_description_tmpl = WHEEL_LONG_DESCRIPTION_CUDA
+                long_description = WHEEL_LONG_DESCRIPTION_CUDA.format(
+                    version=platform_version,
+                    wheel_suffix=''.join(platform_version.split('.')))
             elif kind == 'rocm':
-                long_description_tmpl = WHEEL_LONG_DESCRIPTION_ROCM
+                long_description = WHEEL_LONG_DESCRIPTION_ROCM.format(
+                    version=platform_version)
             else:
                 raise AssertionError('Unreachable')
-            long_description = long_description_tmpl.format(
-                version=platform_version)
 
             # Rename wheels to manylinux.
             asset_name = wheel_name(
@@ -611,7 +612,8 @@ def build_windows(
             'platform_version', cuda_version)
         package_name = WHEEL_WINDOWS_CONFIGS[cuda_version]['name']
         long_description = WHEEL_LONG_DESCRIPTION_CUDA.format(
-            version=platform_version)
+            version=platform_version,
+            wheel_suffix=''.join(platform_version.split('.')))
         asset_name = wheel_name(
             package_name, version, python_version, 'win_amd64')
         asset_dest_name = asset_name
diff --git a/dist_config.py b/dist_config.py
index f3f39da..e76ee5e 100644
--- a/dist_config.py
+++ b/dist_config.py
@@ -309,10 +309,13 @@ class _WheelWindowsConfig(TypedDict):
 
 
 # Long description of the CUDA wheel package in reST syntax.
-# `{version}` will be replaced by the CUDA version (e.g., `9.0`).
+# `{version}` and `{wheel_suffix}` will be replaced by the CUDA version (e.g., `9.0`).
 WHEEL_LONG_DESCRIPTION_CUDA = _long_description_header + '''\
 This is a CuPy wheel (precompiled binary) package for CUDA {version}.
-You need to install `CUDA Toolkit {version} <https://developer.nvidia.com/cuda-toolkit-archive>`_ to use these packages.
+You need to install `CUDA Toolkit {version} <https://developer.nvidia.com/cuda-toolkit-archive>`_ locally to use these packages.
+Alternatively, you can install this package together with all needed CUDA components from PyPI by passing the ``[ctk]`` tag::
+
+   $ pip install cupy-cuda{wheel_suffix}[ctk]
 
 If you have another version of CUDA, or want to build from source, refer to the `Installation Guide <https://docs.cupy.dev/en/latest/install.html>`_ for instructions.
 '''  # NOQA