In some embodiments, the user assistance system may score the candidate message based on user feedback. To facilitate this, the user assistance system may store or otherwise access and obtain a history of user feedback and may refer to the history of user feedback as a basis for scoring the candidate message. User feedback may take various forms, such as messages received from users in response to suggested messages sent by agents and/or user-submitted reviews of agent performance. In an example implementation, the user assistance system may be configured to parse one or more messages received from the user in response to a suggested message that the agent added to the conversation, such as the first message that was received from the user after the suggested message was added. Then, using natural language processing and/or other techniques, the user assistance system may evaluate the user's feedback regarding the suggested message and may score future candidate messages based on the evaluation. For example, if the agent adds a suggested message that reads “I will try and find your lost luggage and get back to you by the end of the day” and the user then responds with a message that reads “No! Find my luggage NOW! !! !”, the user assistance system may determine from the user's response that the suggested message was not acceptable to the user and may thus take this into account when scoring a candidate message that has a similar or identical template as the suggested message (e.g., may decrease the score of the candidate message). On the other hand, if the user instead responds with a message that reads “Thank you very much!”, the user assistance system may determine from the user's response that the suggested message was acceptable to the user and may thus take this into account when scoring a candidate message that has a similar or identical template as the suggested message (e.g., may increase the score of the candidate message). Other examples are possible as well.