diff --git a/mainline/README.md b/mainline/README.md
index c7e93a4..2a12d8d 100644
--- a/mainline/README.md
+++ b/mainline/README.md
@@ -18,7 +18,10 @@
    In general `mainline_modules_sdks-userdebug` is the slowest build to
    complete, so look for that to be on the safe side.
 
-3. Run:
+3. Edit the list returned by `install_entries` in `update.py` in this directory
+   to only cover the updates you need (this change shouldn't be submitted).
+
+4. Run:
 
    ```
    prebuilts/runtime/mainline/update.py \
@@ -33,13 +36,13 @@
    corresponding CI builds so they may not be complete - see `update.py` for
    details.
 
-4. Try to minimize the updates in the created CLs and ensure they are safe. This
-   usually means reverting some prebuilts back to their old contents, and
-   reverting specific hunks in the Android.bp files. Also, please improve the
-   autogenerated CL descriptions to make them more accurate and describe the
+5. Try to further minimize the updates in the created CLs and ensure they are
+   safe. This usually means reverting some prebuilts back to their old contents,
+   and reverting specific hunks in the `Android.bp` files. Also, please improve
+   the autogenerated CL descriptions to make them more accurate and describe the
    reason for the update. Please avoid doing any other changes - in particular
    there must be no local change in any prebuilt file (not counting
-   Android.bp's). mast@google.com is happy to help with review.
+   `Android.bp`'s). mast@google.com is happy to help with reviews.
 
 Note that the script takes a `--local-dist` argument that you can use to test
 with prebuilts from a local build in another tree. However, CLs created that way