Prompting the GHL AI Workflow Builder
Strategy
Create verbose, detailed prompts that give the GHL AI builder enough context to get 80% of the automation built correctly, then manually fix the remaining 20%.
Prompt Structure
Include in your prompt:
- Trigger: Exact trigger type and conditions (e.g., "appointment status changed to no-show")
- Actions: Every step in sequence with timing
- Wait nodes: Specific delays with business hour constraints
- Message copy: Exact SMS and email text with custom value placeholders
- Conditions/filters: Any if/else logic needed
- Pipeline updates: Stage changes that should happen
What the AI Builder Does Well
- Creates basic workflow structure
- Sets up trigger conditions
- Adds action nodes in sequence
- Asks clarifying questions (improved behavior)
- Handles common patterns (SMS, email, wait, pipeline move)
What the AI Builder Gets Wrong
- Sometimes inserts wrong custom value tags — always verify
- May not respect exact formatting (bold, line breaks)
- Can create overly complex flows when simple ones suffice
- Custom webhooks vs normal webhooks — it defaults to premium actions unnecessarily
Reusable Skill Approach
Create a Claude skill called "GHL AI Automation Workflow Prompt" that:
- Takes your plain English description of what the workflow should do
- References GHL documentation for accurate trigger/action names
- Outputs the most detailed prompt possible for the GHL AI builder
- Iterates based on what the builder produces
Key Learning
- "Keep in your skill that custom webhooks are premium actions — we only need normal webhooks"
- Reference GHL documentation skill to pull in correct context
- The GHL AI builder has gotten significantly better recently — worth using for 80/20
- Always verify custom value names match exactly what's configured in your sub-account