Coming Soon for Agent+ plan only
AI safeguards apply only to AI reply workflows (auto-send or pending approval). Manual replies, on-demand AI replies (via Generate), and workflows using saved replies are not affected by safeguards.
Why This Matters
Safeguards ensure AI replies never cross critical boundaries like tone, compliance, or safety. They give you confidence that only brand-safe drafts make it through to your queue.
How Safeguards Work
Safeguards run on two levels:
1. Input Analysis (Original Message)
Checks the incoming comment or message to decide if it’s safe and relevant to be replied to.
Relevance: Blocks off-topic queries.
Competitor mentions: Avoids replying when other brands are named.
Spam & promotions: Filters spammy or promotional content.
Profanity & Safety: Blocks explicit, threatening, or unsafe content.
2. Output Analysis (Reply Content)
Checks the drafted reply itself to ensure it’s compliant and on-brand.
Response tone: Enforces brand voice rules.
Brand voice compliance: Blocks drafts that don’t match your defined style.
Generic replies: Stops robotic or low-value responses.
Professional advice boundaries: Prevents legal, medical, or financial advice.
Promises & guarantees: Blocks drafts that commit refunds, discounts, or guarantees.
Biased content: Blocks hate speech, political, religious, or identity-targeted content.
What Happens When a Reply Is Blocked
Safeguards never stop a draft from being created. Instead, they control whether it can move forward:
Automated replies (auto-send)
The AI reply is drafted but held back.
The draft is flagged and does not publish.
It remains in a pending state for review or auto-archive, depending on your settings.
Pending approval replies (human review)
The AI draft appears in Ready for Send, flagged with safeguard warnings.
Reviewers can Dismiss, Edit, or Send anyway.
AAuto-Archiving Blocked Replies
If you don’t want flagged replies to ever appear in review:
Turn on Automatically archive blocked messages (for input safeguards).
Turn on Automatically archive blocked replies (for output safeguards).
When enabled, flagged drafts skip the review queue and go directly to Done > Archived.
⚠️ Note: Drafts are always generated first — auto-archive just moves them out of sight so your team doesn’t need to take action.
Best Practices
Start broad: Keep most safeguards on until you see where false positives occur.
Adjust by region: Different markets may need different thresholds.
Pair with prompts: Prompts shape the tone; safeguards enforce safety.
Decide on review vs auto-archive:
Review is useful if you want to learn from flagged drafts.
Auto-archive is best if you never want to see or act on them.
Safeguards Quick Reference
Safeguards always let AI draft a reply. What changes is what happens after drafting.
Type | Category | What It Blocks/Flags | Result (Default) | Result (If Auto-Archive On) |
Input (original message) | Relevance | Off-topic queries unrelated to brand | Draft flagged in queue | Draft flagged then auto-archived |
| Competitor mentions | Messages naming competing brands | Draft flagged in queue | Draft flagged then auto-archived |
| Spam & promotions | Excessive emojis, spam, promo content | Draft flagged in queue | Draft flagged then auto-archived |
| Profanity & Safety | Explicit, threatening, unsafe content | Draft flagged in queue | Draft flagged then auto-archived |
Output (drafted reply) | Response tone | Tone doesn’t match brand style | Draft flagged in queue | Draft auto-archived |
| Brand voice compliance | Non-compliant with voice/personality | Draft flagged in queue | Draft auto-archived |
| Generic replies | Robotic or low-value replies | Draft flagged in queue | Draft auto-archived |
| Professional advice | Legal, medical, financial guidance | Draft flagged in queue | Draft auto-archived |
| Promises & guarantees | Refunds, discounts, commitments | Draft flagged in queue | Draft auto-archived |
| Biased content | Hate speech, political, religious, or identity-based content | Draft flagged in queue | Draft auto-archived |
💡 Tip: Use safeguards to reduce noise and risk. Then decide whether your team should review flagged drafts or let them skip the queue with auto-archiving.