Does a reply feel off the mark, too flat, too direct, or factually wrong? Here is how to fix it, in the moment and then for good.
In the moment
Take over the conversation (turn off its automation), reply yourself, then hand it back to the AI. It will take your messages into account; they are part of the history.
Fixing it for good: the source of the problem
Reply quality flows directly from what you have configured. Depending on the symptom:
The tone does not fit, the style sounds fake: refine the model's profile, especially the conversation style, favorite expressions, emoji usage and spelling (see "Fill in your model's profile").
The AI is vague on a topic: the matching profile field is probably empty. An empty field is a topic it has nothing to say about.
It brings up a topic that should not exist: fill in the no-go areas and topics to avoid in the profile.
It describes sold content wrong (wrong outfit, wrong color): the script step's description is imprecise or missing. That is its only source of truth about the scene (see "Media descriptions").
It sells too hard, or not enough: review your prices and reduced prices (the negotiation margin), and the number of messages before the next step in your scripts.
Nudges land at the wrong time: adjust the delays in Follow-up Settings.
It forgot or distorted a detail about a fan: fix the fan's note directly in the conversation; the AI will take it into account.
If abnormal behavior repeats
Some rules are firm commitments of the AI (never announce a media without sending it, never invent a service or a price, never sell below the minimum price: see "What the AI never does"). If you see one of these rules broken, or the same flaw coming back, report it to support via the Messenger with the conversation link and, if possible, a screenshot. These reports are handled as a priority and feed directly into improving the product.