Helpin
Dark mode
AI response modes & follow-up limits

AI response modes & follow-up limits

Two settings define the AI's day-to-day behavior: what it does with a message (response mode) and how long it keeps trying (follow-up limits). Both live in Settings → AI Assistant.

Response modes

When a message comes in offers two modes:

Mode

Behavior

Best for

Reply directly

The AI posts a customer-visible answer in the thread.

Deflection — let AI resolve routine questions end-to-end.

Add internal note

The AI drafts an answer as a private internal note for agents; nothing goes to the customer.

Review-first teams — agents send (or edit and send) the draft.

The mode is workspace-wide and applies to all AI-handled conversations. Switching modes changes behavior for new messages; existing conversations continue with their current state.

With Reply directly, agents can still take over any conversation at any moment — replying as a human immediately sets takeover.

Confidence threshold

The Confidence select (10%–100%) decides when the AI is allowed to answer:

  • AI answers only when its confidence in a grounded answer is at or above the threshold.

  • Higher confidence → fewer AI answers and more handoffs (safer, slower).

  • Lower confidence → more AI answers (more deflection, higher risk of a wrong answer).

The tooltip guidance: "AI only answers when its confidence is at or above this value. Higher confidence means fewer AI answers and more human handoffs."

Start around 70% and adjust from observed outcomes: wrong answers → raise; too many unnecessary handoffs → lower.

Follow-up limits ("Handoff after")

The Handoff after select (1–10) bounds AI persistence on the same problem:

  • Counts AI follow-up replies on the same problem in a conversation.

  • Resets when the visitor brings up a new problem.

  • Reaching the limit forces handoff to a human (see Stuck detection & same-issue handoff).

Talk to a human

The Show "Talk to Human" button toggle controls whether visitors can request a person at any time. On means the customer always has an exit from AI; off, escalation happens only via AI judgment. Recommended: keep it on — a customer-driven escape hatch builds trust and produces cleaner escalations.

Escalation messages

Customer-facing copy shown when the AI hands off, per scenario:

  • Default — normal handoff, e.g. "Let me connect you with a team member — they typically reply in {reply_time}."

  • Team busy — handoff while the team is busy, using {reply_time} for the expected reply time.

  • After hours — handoff outside business hours, using {next_open} for when the team returns (accurate only when business hours are enabled).

Each field shows a live preview. These templates support the placeholders shown below the field.

Where this fits with routing

Response mode and limits decide how AI behaves; routing decides where handoffs land. Configure both — see Inboxes / routing settings for the destination and assignment behavior.

Was this article helpful?