diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index ae2e30d6e..49efc7ef4 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -125,6 +125,9 @@ jobs:
         name: Install Python dependencies
         run: |
           pipenv sync --dev
+          pipenv run python3 -m nltk.downloader snowball_data
+          pipenv run python3 -m nltk.downloader stopwords
+          pipenv run python3 -m nltk.downloader punkt
       -
         name: List installed Python dependencies
         run: |
diff --git a/docker/docker-entrypoint.sh b/docker/docker-entrypoint.sh
index c97225007..14a0650f0 100755
--- a/docker/docker-entrypoint.sh
+++ b/docker/docker-entrypoint.sh
@@ -53,24 +53,6 @@ map_folders() {
 	export CONSUME_DIR="${PAPERLESS_CONSUMPTION_DIR:-/usr/src/paperless/consume}"
 }
 
-nltk_data () {
-	# Store the NLTK data outside the Docker container
-	local nltk_data_dir="${DATA_DIR}/nltk"
-
-	# Download or update the snowball stemmer data
-	python3 -m nltk.downloader -d "${nltk_data_dir}" snowball_data
-
-	# Download or update the stopwords corpus
-	python3 -m nltk.downloader -d "${nltk_data_dir}" stopwords
-
-	# Download or update the punkt tokenizer data
-	python3 -m nltk.downloader -d "${nltk_data_dir}" punkt
-
-	# Set env so nltk can find the downloaded data
-	export NLTK_DATA="${nltk_data_dir}"
-
-}
-
 initialize() {
 
 	# Setup environment from secrets before anything else
@@ -123,8 +105,6 @@ initialize() {
 	done
 	set -e
 
-	nltk_data
-
 	"${gosu_cmd[@]}" /sbin/docker-prepare.sh
 }
 
diff --git a/docker/docker-prepare.sh b/docker/docker-prepare.sh
index c4e45c032..75504ee8c 100755
--- a/docker/docker-prepare.sh
+++ b/docker/docker-prepare.sh
@@ -89,6 +89,24 @@ superuser() {
 	fi
 }
 
+nltk_data () {
+	# Store the NLTK data outside the Docker container
+	local nltk_data_dir="${DATA_DIR}/nltk"
+
+	# Download or update the snowball stemmer data
+	python3 -m nltk.downloader -d "${nltk_data_dir}" snowball_data
+
+	# Download or update the stopwords corpus
+	python3 -m nltk.downloader -d "${nltk_data_dir}" stopwords
+
+	# Download or update the punkt tokenizer data
+	python3 -m nltk.downloader -d "${nltk_data_dir}" punkt
+
+	# Set env so nltk can find the downloaded data
+	export NLTK_DATA="${nltk_data_dir}"
+
+}
+
 do_work() {
 	if [[ "${PAPERLESS_DBENGINE}" == "mariadb" ]]; then
 		wait_for_mariadb
@@ -100,6 +118,8 @@ do_work() {
 
 	migrations
 
+	nltk_data
+
 	search_index
 
 	superuser