Skip to main content

​
Inbox terminology FAQ — replies, consent, and AI drafts

Three vocabularies mix across the Inbox surfaces, and they are easy to conflate while you scan a screen: replies (the conversation and its approval queue), consent (receipts and the posture they form), and AI drafts (the co-pilot’s suggested-turn records). This page defines each term once and answers the collisions operators keep asking about, so the support team can cite a stable link instead of re-explaining.
  • Section 1 — term conventions: what counts as a reply, a consent record, or an AI draft turn.
  • Section 2 — the specific collisions operators ask about: the pending-replies count, approving a draft versus consent posture, a draft versus a queued reply, and more.
  • Section 3 — the three guides this page leans on and what each one owns.
  • Section 4 — how to cite this page from support replies.

​
1. Term conventions

These are the conventions used across the platform’s Inbox screens. Use the same vocabulary in tickets and operator hand-offs.
  • A reply — in the Inbox, a reply is a message drafted to send into a conversation. It exists in one of two places: dispatched directly when it is sent, or held for review when the workspace gates its author.
  • A pending reply — a reply held by the reply-approval gate, waiting for a supervisor decision. “Pending” describes that queue position, not a conversation state, a message-delivery state, or customer consent.
  • An AI draft turn — when the Inbox AI co-pilot proposes a response, each proposal carries a model-confidence score and becomes an auditable turn record. High-confidence drafts can qualify for one-click approve-and-send; escalate or end suggestions never qualify, at any score.
  • A consent receipt — the contact-level record of opt-in or opt-out (often created alongside a conversation when a channel captures consent). Receipts feed the compliance-posture surface, where per-scope details roll up into a posture block.
  • Consent posture — the screen that aggregates consent receipts into a posture summary for the workspace. It reports state; it is not part of the reply workflow and is never an approve/reject decision.
Conflating these four ideas is where most terminology errors come from, so the distinctions above are the ones to settle first.

​
2. Collisions operators ask about

Each question below names the two vocabularies that collide and the distinction that resolves it.

​
What exactly does the pending-replies count include?

It counts replies held by the reply-approval gate — nothing else. A queue that shows N pending rows means N gated replies are waiting for a supervisor to approve or reject; it is not a count of unread conversations and it is not a count of blocked sends. The queue renders up to 50 pending rows and mounts the first 10, growing with Load more, so a short visible list plus a larger remainder is normal. No. Approving a pending reply is a send decision on one queued body, made by an owner, admin, or supervisor — the queue row records the decision and, for approvals, points at the message that went out. Consent posture is the workspace’s aggregated consent state, report-only. The gate that holds the reply is a tenant-owned control you turn on per agent (off by default, with a workspace default); posture never participates in that approve/reject step.

​
Is an AI draft the same as a pending reply?

No. A draft is what the AI co-pilot proposes; a pending reply is what a gated teammate submitted. Both can require review before they send, but the requirements are separate per-agent/per-draft settings: the gate is about who wrote the reply, the AI-confidence rule is about what the co-pilot suggested. A draft always carries require_human_review = true, and only send-shaped suggestions can be eligible for one-click approve-and-send. No. Rejecting simply holds the queued reply out of the customer and records a reason for the submitter. Nothing changes on the consent side. (Any consent effect — for example cleaning up an opt-in captured alongside the message — would be a separate contact-level action a supervisor takes outside the queue.)

​
Posture versus receipts: why are there two pages?

Receipts are the individual opt-in and opt-out records; posture is the aggregate. You inspect receipts when you need a particular contact’s history, and you read posture when you need the workspace’s overall state. The FAQ keeps both because the deep-dive page for receipts and the FAQ page for posture answer different questions.

​
3. The three guides this page leans on

Open the guide that owns the surface rather than re-deriving terminology in every ticket:
  • Pending replies — the supervisor approval queue — how the queue works and who can act: when the gate applies, reading each card, approving, rejecting with a note, queue hygiene, and the privilege model (owners, admins, supervisors decide; agents submit).
  • Reply approvals — API — the endpoints the queue mirrors, with curl samples for listing, approving, rejecting, and reading the gate defaults.
  • Compliance Posture FAQ — the consent/gate vocabulary: GET /compliance/health readback, consent receipts, posture blocks, defaults, and the sole platform-wide TCPA carve-out.

​
4. Cite this page in support replies

This page is stable enough to cite verbatim in tickets. The route is /guides/inbox-terminology-faq, and the sections above are named in the H2s so a link plus a section number lands the reader on the exact answer. When a new collision shows up in support, add the collision to Section 2 and keep the three see-also links pointed at the owning guides.

​
See also