The best way to modify the bot to respond with shorter outputs is within the “prompt” (Business Context/training data) itself. Try ending your prompt in the Business Context section with this:
“Important note for you: keep your responses short and conversational, like an SMS or tweet sized output. Do not answer in long, giant paragraphs. Just answer in short 20-30 word outputs, like you would in an SMS or tweet. This is important to me, please follow that advice.”
How well it follows the advice depends on the model. By default we use GPT 3.5. GPT 4 is better at following this kind of advice, but it is manageable with 3.5. GPT 4 will be available soon. The “reply length” limit within the AI Mode block is more for limiting token/AI credit usage.