AI replies
Suggested drafts above the compose box and the Reviews approval queue for Instagram.
Scrollmark's AI reply system has two surfaces. Both sit inside the Inbox. Both keep a human in the loop.
- Inbox suggestions — up to three AI-drafted replies appear above your compose box whenever the last message came from a follower. Click one, edit if you want, send.
- Reviews — a separate tab in the classic Inbox where journey-triggered replies wait for approval before they go out.
Reviews is Instagram-only today
The Reviews approval queue only runs on Instagram. Other platforms get inbox suggestions but not the approval workflow.
Inbox suggestions
When you open a conversation where a follower sent the last message, Scrollmark fetches three draft replies from your tenant's tuned model. Each draft carries a tone label and a confidence score under the hood.
How it behaves
- One follower message in, three suggestions out. The draft is built from the most recent follower message plus prior conversation context.
- Similar-conversation lookup. Scrollmark searches past conversations that matched this one and borrows patterns from replies your team actually sent.
- Cached for an hour. The same message on the same org hits cache, so opening the conversation again is instant.
- Click to insert. The suggestion drops into the compose textarea. Edit freely — nothing sends until you hit enter.
When you won't see suggestions
- The last message in the conversation was from you, not the follower
- Messages are still loading
- The conversation is empty (a fresh comment with no reply yet)
The Reviews tab (Instagram)
When a Scrollmark automation — a journey — wants to reply on your behalf, it can be configured to draft the message and hold it for human approval. Those drafts land in Inbox → Reviews.
Working the queue
Open Inbox → Reviews
Classic Inbox only. The tab shows a count badge with the number of pending approvals.
Read the draft
Each draft is shown in a chat bubble with a Draft badge. The journey that produced it is linked at the bottom — click to see the full Journey History.
Edit if needed
Click the edit button in the bubble to tweak wording, insert an emoji, and rewrite. An Edited tag shows next to the Draft badge so you know the reply no longer matches the original generation.
Approve or reject
Approve sends the message. Reject discards it and removes it from the queue. Approving also saves your edit — the model uses those edits as signal for future drafts.
Use the Review only toggle
From the main Inbox, flip the Review only pill in the filter bar to hide everything except conversations waiting on human review. This is the fastest way to clear your approval backlog without switching tabs.
What the AI learns from
Scrollmark tunes replies per tenant. The things that feed the model:
- Your past approved replies — both manual and edited drafts
- Rejected drafts (negative signal)
- Conversation metadata — platform, follower tags, engagement type
- Similar past conversations with the same intent
You don't train the model by uploading examples. You train it by using it — approve the drafts you'd send, edit the ones that are close, reject the ones that miss. A week of normal inbox work is usually enough to shift tone noticeably.
Route sensitive replies through Reviews, never direct-send
Scrollmark's Reviews queue exists because some messages should never go out without eyes on them. If a journey could produce a reply to a refund request, a legal threat, a medical question, or a harassment report, that journey must route through Reviews. Always.
Troubleshooting
| Symptom | Likely cause |
|---|---|
| No suggestions appear | Last message is from you, not the follower |
| Same three suggestions every time | Cache hit — suggestions refresh hourly per message |
| Drafts stuck in Reviews | Approval queue only runs on Instagram accounts |
| Drafts feel off-brand | Edit and approve instead of rejecting — edits are the strongest training signal |