How does Rypl write a reply?
Updated
For each new comment, Rypl reads what the person wants, decides whether the agent should answer, and writes the reply from your instructions, tone and knowledge base, in the language of the comment, unless the channel's Reply language is set to a fixed one.
Reads the comment
Rypl gives it an intent label, such as Question or Complaint, and a sentiment.
Moderates it
Spam and hate are hidden first, by your banned words, rules and filters. Hidden comments get no reply.
Decides whether to answer
By default the agent answers questions, purchases and requests, and your reply rules can change that. A complaint that only states a problem isn't answered; if it asks something, the agent answers. Legal threats and emergencies always wait for a person.
Writes the reply
From the agent's instructions, tone and knowledge base. On Pro and Ultra it can also read the post the comment is under, and on Ultra it remembers earlier messages from the same person.
Posts it
Within seconds, or after the delay set in How soon. Every reply shows up in Conversations.
Is every Rypl customer's AI the same?
No. Each agent answers from its own instructions and knowledge base, so two businesses get different replies to the same question. On Ultra, the agent also uses Rypl's most powerful AI. The more your knowledge base covers, the more exact the replies: if an answer is missing or off, add the fact there.
Related
What is an agent
An agent is the AI that answers your comments and messages. It holds your instructions, tone and knowledge base, and you assign it to channels.
Brand voice and tone
Pick one of four tones and write instructions of up to 4,000 characters. Specific instructions and a few example answers work best.