diff --git a/admin_manual/ai/overview.rst b/admin_manual/ai/overview.rst
index cb59310b04a..944b6803b0c 100644
--- a/admin_manual/ai/overview.rst
+++ b/admin_manual/ai/overview.rst
@@ -416,6 +416,39 @@ The complete logs of the workers can be checked with (replace 1 with the worker
 
    journalctl -xeu nextcloud-ai-worker@1.service -f
 
+If your Nextcloud runs inside a Docker container (for example with Nextcloud AIO), the worker has to run
+*inside* the container. Make the service unit shown above wait for Docker by extending its ``[Unit]``
+section:
+
+.. code-block:: ini
+
+   [Unit]
+   Description=Nextcloud AI worker %i
+   After=network.target docker.service
+   Requires=docker.service
+
+and call ``occ`` through ``docker exec`` in ``taskprocessing.sh`` instead of running it locally. The
+``[Unit]`` ordering above only waits for the Docker daemon, not for the ``nextcloud-aio-nextcloud``
+container, so the script waits until that container reports healthy before running ``occ`` (the AIO image
+ships a healthcheck for php-fpm and the database). Without this wait the worker would exit immediately
+after a host reboot and, with ``Restart=always``, hit systemd's start-rate limit and stay down until a
+manual reset:
+
+.. code-block:: bash
+
+   #!/bin/sh
+   echo "Starting Nextcloud AI Worker $1"
+   i=0
+   until [ "$(docker inspect -f '{{.State.Health.Status}}' nextcloud-aio-nextcloud 2>/dev/null)" = healthy ]; do
+       i=$((i + 1))
+       [ "$i" -ge 60 ] && echo "nextcloud-aio-nextcloud not healthy after 5 min" && exit 1
+       sleep 5
+   done
+   docker exec -i nextcloud-aio-nextcloud sudo -E -u www-data php occ taskprocessing:worker -v -t 60
+
+Use ``docker exec -i`` without ``-t``: systemd does not allocate a pseudo-TTY, so adding ``-t`` makes the
+command fail with ``the input device is not a TTY``.
+
 
 Frequently Asked Questions
 --------------------------