Skip to main content
Use this playbook when an AI agent is helping a support team review SiteGPT conversations and maintain the support queue.

​
Required access

Recommended scopes: 
sitegpt login \
  --profile support-agent \
  --scope account:read \
  --scope chatbots:read \
  --scope conversations:read \
  --scope conversations:write \
  --scope leads:read \
  --scope leads:write
Add these when the agent must manage tags or delete records: 
--scope conversations:delete --scope leads:delete

​
1. Establish context

sitegpt whoami --json
sitegpt chatbots list --json
sitegpt tags list --json
Pick the chatbot: 
CHATBOT_ID=<chatbot-id>

​
2. Review the active support queue

sitegpt conversations list --chatbot $CHATBOT_ID --status open --json
sitegpt conversations list --chatbot $CHATBOT_ID --escalated true --json
sitegpt conversations list --chatbot $CHATBOT_ID --important true --json
sitegpt conversations list --chatbot $CHATBOT_ID --read false --json
Prioritize:
  1. Escalated conversations.
  2. Important conversations.
  3. Unread conversations.
  4. Failed or negative-feedback conversations, if surfaced in the thread.

​
3. Inspect a thread

sitegpt conversations get --chatbot $CHATBOT_ID <thread-id> --json
sitegpt messages list --chatbot $CHATBOT_ID <thread-id> --json
Summarize:
  • Visitor goal
  • Last user question
  • Last AI answer
  • Whether a human should take over
  • Any missing knowledge that caused confusion
  • Relevant lead/contact information

​
4. Apply tags and state

Create tags if needed: 
sitegpt tags add "Billing"
sitegpt tags add "Bug report"
sitegpt tags add "Sales"
Assign tags to a thread: 
sitegpt conversations update --chatbot $CHATBOT_ID <thread-id> --tag <tag-id> --tag <another-tag-id>
Mark important or resolved: 
sitegpt conversations star --chatbot $CHATBOT_ID <thread-id>
sitegpt conversations resolve --chatbot $CHATBOT_ID <thread-id>
sitegpt conversations read --chatbot $CHATBOT_ID <thread-id>
Escalate to human support: 
sitegpt conversations escalate --chatbot $CHATBOT_ID <thread-id> --message "A support teammate will follow up shortly."
Return to AI mode after a human support window: 
sitegpt conversations switch-to-ai --chatbot $CHATBOT_ID <thread-id> --message "The AI assistant is available again."

​
5. Maintain leads

List and inspect leads: 
sitegpt leads list --chatbot $CHATBOT_ID --status open --json
sitegpt leads get --chatbot $CHATBOT_ID <lead-id> --json
Update lead details: 
sitegpt leads update --chatbot $CHATBOT_ID <lead-id> --name "Jane Doe" --phone "+15551234567"
sitegpt leads star --chatbot $CHATBOT_ID <lead-id>
sitegpt leads archive --chatbot $CHATBOT_ID <lead-id>
Bulk archive handled leads: 
sitegpt leads bulk --chatbot $CHATBOT_ID --action archive <lead-id> <lead-id>

​
6. Use messages carefully

Start a new test conversation: 
sitegpt messages send --chatbot $CHATBOT_ID "What can you help with?" --json
Send a visitor-style message into an existing thread: 
sitegpt messages send --chatbot $CHATBOT_ID <thread-id> "Can I talk to support?" --json
The CLI sends visitor messages. It does not let you impersonate internal agent or system senders.

​
7. Identify knowledge gaps

When a conversation shows a weak or incorrect answer, capture:
  • Thread ID
  • User question
  • Bot answer
  • Missing or outdated source material
  • Suggested new document, link, or Custom Response
Then hand off to the knowledge-maintenance workflow: 
sitegpt knowledge custom-responses add \
  --chatbot $CHATBOT_ID \
  --question "Exact question visitors are asking" \
  --answer "Approved answer"

​
8. Daily support report prompt

An agent can generate a daily report after running: 
sitegpt conversations list --chatbot $CHATBOT_ID --status open --json
sitegpt conversations list --chatbot $CHATBOT_ID --escalated true --json
sitegpt leads list --chatbot $CHATBOT_ID --status open --json
sitegpt tags list --json
Report format:
  • Open conversations count
  • Escalated conversations count
  • Unread conversations count
  • New or important leads
  • Repeated topics
  • Suggested Custom Responses or new knowledge documents
  • Threads that need human review

​
Safety rules for agents

  • Do not delete conversations or leads unless explicitly asked.
  • Prefer resolve over delete for completed support work.
  • Use --json before making decisions that depend on IDs or nested fields.
  • When updating tags, remember that repeated --tag values replace the assigned tag set.
  • Do not claim a customer was contacted unless an external support system confirms it.