ai-draft-edited tag whenever they have to rewrite a suggested answer (and customer-callback-fix if a customer calls back to correct a mistake).
ai-draft-edited tags: once an intent passes 85% unedited replies with zero customer correction callbacks, flip that branch to Auto-Pilot and keep human review only on exception tags.
Set up a 50-thread Suggestive-to-Auto-Pilot evaluation workflow in sub-account <YOUR_LOCATION_ID>: 1. In Settings > Private Integrations, scope the agent token to contact tags, notes, and standard calendar slots while leaving delete and billing scopes disabled. 2. In AI Agents > Conversation AI and Automation > Workflow Builder, run new intents in Suggestive mode and track ai-draft-edited and customer-callback-fix tags on contact records. 3. In Automation > Workflow > Execution Logs, promote an intent branch to Auto-Pilot once 85% or more of its last 50 drafts send without human edits.
# Audit private integration token scopes and active AI workflows in the sub-account ghl-agency user-role-governance --context <YOUR_CONTEXT> --location <YOUR_LOCATION_ID> ghl-crm workflows list --context <YOUR_CONTEXT> --location <YOUR_LOCATION_ID>
Official GHL MCP vs. ghl-tools: HighLevel’s official hosted MCP (services.leadconnectorhq.com/mcp) covers public v2 CRM records (contacts, opportunities, appointments, conversations). ghl-tools MCP complements it with workflow graph inspection & recipes, funnel/form builders, DNS/BIMI checks, and ghl-crm doctor audits.
| Step | Menu Path in GHL | Action |
|---|---|---|
| Step 1 | Settings > Private Integrations | Create a scoped token for one specific job (such as standard FAQ replies or appointment rescheduling) with delete, billing, and custom-quote permissions unchecked so the agent can act inside clear limits. |
| Step 2 | AI Agents > Conversation AI | Run the new intent in Suggestive mode for the first 50 requests and instruct reps to apply an ai-draft-edited tag whenever they have to rewrite a suggested answer (and customer-callback-fix if a customer calls back to correct a mistake). |
| Step 3 | Automation > Workflow > Execution Logs | Review the weekly count of unedited sends versus ai-draft-edited tags: once an intent passes 85% unedited replies with zero customer correction callbacks, flip that branch to Auto-Pilot and keep human review only on exception tags. |
Connect Claude Code, Cursor, or OpenAI Codex to our ghl-tools MCP Server & Rust CLI to build workflows, forms, and funnels from your editor, or start your GoHighLevel plan through our partner link to unlock $612+ in ready-to-import kits (including GHL AI Agent Oversight & Evaluation Kit) and weekly live clinics.