diff --git a/README.md b/README.md
index b2329ec..763c85c 100644
--- a/README.md
+++ b/README.md
@@ -332,6 +332,13 @@ server, disconnect Desktop before running the restart command. No restart
 notice is printed when the catalog and managed provider configuration are
 unchanged or when a newer daemon has already loaded them.
 
+After setup or any required reconnect, start a new Codex task before selecting
+a Webster model. Codex persists the model provider with each task, so a task
+created before setup remains on the built-in `openai` provider even if its
+model is later changed in the picker. Selecting a Webster model in that older
+task sends it to the ChatGPT backend and produces a "model is not supported
+when using Codex with a ChatGPT account" error.
+
 It is not wired into `install.sh`, because it needs both a Webster secret and an
 existing Codex login. Run it separately after the bootstrap.
 
diff --git a/codex-setup.sh b/codex-setup.sh
index 0cc8882..be38044 100755
--- a/codex-setup.sh
+++ b/codex-setup.sh
@@ -124,11 +124,15 @@ report_codex_client_reload() {
 
   if [ "$state" = managed ] && [ -f "$pid_file" ] && [ "$pid_file" -nt "$RELOAD_MARKER" ]; then
     [ "$CHECK_ONLY" = 1 ] || rm -f "$RELOAD_MARKER"
+    printf '\nStart a new Codex task before selecting a Webster model. '
+    printf 'Existing tasks keep the model provider they started with.\n'
     return 0
   fi
   if [ "$state" = stopped ]; then
     [ "$CHECK_ONLY" = 1 ] || rm -f "$RELOAD_MARKER"
     printf '\nCodex app-server is not running; the new models will load on its next start.\n'
+    printf 'Start a new Codex task before selecting a Webster model. '
+    printf 'Existing tasks keep the model provider they started with.\n'
     return 0
   fi
 
@@ -151,6 +155,8 @@ report_codex_client_reload() {
     printf 'Then disconnect and reconnect this machine in Codex Desktop '
     printf '(or fully quit and reopen Desktop).\n'
   fi
+  printf 'Start a new Codex task before selecting a Webster model. '
+  printf 'Existing tasks keep the model provider they started with.\n'
 }
 
 record_failure() {
diff --git a/test.sh b/test.sh
index e49b657..cc2b6ae 100755
--- a/test.sh
+++ b/test.sh
@@ -1249,6 +1249,7 @@ if ! (
   unmanaged_notice=$(report_codex_client_reload 2>&1)
   printf '%s' "$unmanaged_notice" | grep -q '1. Disconnect'
   printf '%s' "$unmanaged_notice" | grep -q '3. Reconnect'
+  printf '%s' "$unmanaged_notice" | grep -q 'Start a new Codex task'
   mkdir -p "$CODEX_DIR/app-server-daemon"
   touch -t 202001010000 "$CODEX_DIR/app-server-daemon/app-server.pid"
   touch -t 202101010000 "$RELOAD_MARKER"
@@ -1256,8 +1257,13 @@ if ! (
   reload_notice=$(report_codex_client_reload 2>&1)
   printf '%s' "$reload_notice" | grep -q 'codex app-server daemon restart'
   printf '%s' "$reload_notice" | grep -q 'disconnect and reconnect'
+  printf '%s' "$reload_notice" | grep -q 'Existing tasks keep the model provider'
+  export TEST_CODEX_DAEMON_STATE=stopped
+  stopped_notice=$(report_codex_client_reload 2>&1)
+  printf '%s' "$stopped_notice" | grep -q 'new models will load on its next start'
+  printf '%s' "$stopped_notice" | grep -q 'Start a new Codex task'
 ); then
-  echo "FAIL: Codex stale-daemon detection omitted restart/reconnect guidance" >&2
+  echo "FAIL: Codex stale-daemon detection omitted restart/reconnect/new-task guidance" >&2
   exit 1
 fi