CuraCourseWhat this block does
AI Chat is a conversation activity with an instructor-defined purpose. It supports AI, simulated, and hybrid modes. Use a system prompt to establish a role, or write simulated trigger-and-response pairs for a scripted exercise.
Configure a conversation
Insert AI Chat, write the student instructions, and choose the mode. Configure the system prompt or simulated responses and fallback message. Set the message limit and decide whether attachments are enabled, including the maximum number allowed across the conversation.
Set grading objectives
- Participation: points require the configured minimum number of student messages. Assistant replies do not count.
- Keyword in user message: checks student messages for the configured keyword, case-insensitively, with up to one letter of typo tolerance per word. An empty keyword awards no points.
- Follow-up question: requires at least two student messages and a question mark in a later message. It does not judge whether the question is thoughtful.
- Format request: checks for the configured format words, such as list, table, bullets, or outline.
Each criterion awards its full points or zero when submitted. Objective titles are internal; the keyword is not shown to students. Explain the intended learning task in the prompt rather than treating these checks as a semantic assessment of conversation quality.
Example: practice an interview
Define the assistant’s role as an interviewee on a course topic. Ask students to pose an opening question, ask a follow-up, and request a short summary in a list. Set participation and prompting objectives to match those observable actions.
Set boundaries
Choose strict, moderate, or minimal safety filtering, add a topic restriction, and configure custom blocked phrases. Strict uses broader topic keyword filters; moderate filters profanity; minimal disables the built-in word filter. Topic restrictions and custom phrases apply at every level. These are keyword controls, not a provider moderation API or a guarantee of safety.
Troubleshooting
If a follow-up objective fails, check the number of student messages and whether a later message contains a question mark. If scripted replies do not trigger, review the trigger type and text, then test the fallback.