diff --git a/patterns/find_hidden_message/system.md b/patterns/find_hidden_message/system.md
index e7e40a5..3b82578 100644
--- a/patterns/find_hidden_message/system.md
+++ b/patterns/find_hidden_message/system.md
@@ -18,9 +18,7 @@ You consume input and cynically evaluate what's being said to find the true mess
 
 - In a section called HIDDEN MESSAGE, output a single 15-word sentence that captures the true/hidden message of the input. This should be different than the OVERT MESSAGE they are portraying, and if you don't see any divergence between the two then output, "No hidden message detected."
 
-- In a section called MESSAGE SUMMARY, output a single 25-word sentence that combines the two into the following format:
-
-"They are pretending to say is ****\_****, but they are actually trying to get people to think is ****\_****."
+- In a section called MESSAGE SUMMARY, output a single 25-word sentence that combines the two into the following format: "In other words, the superficial message is [RESTATED OVERT MESSAGE], but the hidden message is [RESTATED HIDDEN MESSAGE]."
 
 - In a section called HIDDEN OPINIONS, output a bulleted list of 10-word true/hidden opinions extracted from the input.