Recommendation:
Use Zapier to catch Intercom conversation events, call ChatGPT for a concise first reply, and then log conversation metadata into Google Sheets. This keeps the ops side low-code, isolates PII before sending to the model, and preserves an easy audit trail.

Why this approach works
- Zapier glues Intercom -> AI -> Sheets without engineering overhead.
- ChatGPT gives a fast, friendly automated first touch while you keep humans for complex cases.

Decision criteria (pick based on your situation)
- Budget: Zapier and OpenAI/ChatGPT use typically need paid tiers for reliable webhooks, multi-step Zaps, or higher rate limits. If budget is tight, expect to pay for at least one mid-tier Zapier plan.
- Skill level: Non-technical ops can build this in Zapier with the steps below; you’ll need to test regex for redaction and a small prompt.
- Team size / SLA: If high volume or strict SLA, add throttling and human-hand-off rules to avoid missed responses.
- Output quality / safety: If customer messages may contain PII, add a redaction step and a human review path.

Checklist (step-by-step Zap outline you can copy)
1) Trigger: Intercom -> “New Conversation” or set Intercom webhook to Zapier “Catch Hook.” Map conversation id, sender id, message body, timestamp.
2) Formatter (Zapier) -> Text -> Replace: redact obvious PII (email regex, phone numbers, SSNs). Keep a copy of original in internal logs only if allowed.
3) Optional: Intent/keyword check (Filter or Formatter): detect words like refund, legal, cancel, password. If matched, route directly to human (create a ticket + log).
4) Action: ChatGPT (Zapier’s ChatGPT/OpenAI app) — send a prompt composed of:
- System: “You are a friendly customer support assistant. Keep the reply ≤2 short sentences, be polite, ask one clarifying question if needed, offer to escalate to a human. Don’t invent account data.”
- User: include redacted message + short metadata (product, locale if available).
Example prompt: “Customer message: {{redacted_message}}. Preferred language: {{locale}}. Keep answer short, friendly, ask one clarifying question if needed, and say ‘I can escalate this to a human.’”
5) Action: Google Sheets -> Create Spreadsheet Row. Log fields: timestamp, conversation_id, sender_id, original_message_excerpt, redacted_message, chatgpt_response, intent_flag, routed_to_human (yes/no), zap_run_id.
6) Action: Intercom -> Reply to Conversation (optional): post the ChatGPT response as the first automated reply or add a private note if you prefer a human to review first.
7) Monitoring: create a Zap step to notify Slack/Email on failures or when certain flags present (escalations, errors).

Best-for / Avoid-if
- Best for: teams wanting fast first-touch, lower engineering cost, and simple audit logs.
- Avoid if: you must handle regulated PII without vendor review, need very high customization or low-latency at massive scale (consider a direct API integration instead).

Operational tips
- Keep the automated reply conservative: short, ask to clarify, and offer escalation.
- Log the original (securely) and redacted copies separately.
- Test with a staging Intercom inbox and sample sensitive messages before going live.

If you want, I can draft the exact Zapier field mappings and a drop-in ChatGPT prompt for your Intercom field names.

Compare Zapier and Make