DM and email replies
The co-reply tool drafts your next message in a private thread by reading the transcript visible on your screen and continuing it in your voice. It is told outright that only the visible messages were readable, which stops it inventing an earlier conversation it never saw. It works in Messenger, Instagram DMs, LinkedIn messaging and Gmail, and it never sends — in bulk or otherwise.
Prefer the long version? Comment or DM: where your next client actually comes from walks through it in depth.
What it does
Reads the conversation on screen as a transcript — who said what, in order — and drafts your next message in it.
That is a different job from a comment reply, and it is built differently. A comment reply works from a post plus a set of parallel comments. A DM reply works from a back-and-forth where sequence matters and the last message is usually the one you are answering.
The visible-transcript rule
The prompt states outright that only the messages visible on your screen were readable.
That instruction exists for a specific failure. Without it, a model presented with a partial conversation fills in the earlier part — confidently inventing a history that never happened, which in a real customer thread is worse than an obviously incomplete draft.
The practical consequence: scroll up before you generate if earlier context matters. If a customer's original question is three screens up, scroll it into view or the draft will answer only the most recent message.
It drafts from the conversation you can see, and it knows that is all it can see. Scroll first if the earlier part of the thread matters.
Where it works
| Platform | Surface |
|---|---|
| Messenger threads | |
| Direct messages | |
| Messaging | |
| Gmail | Email replies |
What differs from a comment reply
| Comment reply | DM / email reply | |
|---|---|---|
| Material | A post plus visible comments | A transcript, in order |
| Match the room | Available | Not applicable — no comments to match |
| Audience | Everyone reading the thread | One person |
| Typical length | Short | Whatever the thread's rhythm is |
Match the room is ignored here, because there is no room to match — a private thread has no top comments to take a cue from. Your standing voice still applies.
Keeping the register consistent
A specific failure worth naming: the public comment sounds like a person and the private message sounds like a business.
It happens because the two get written in different modes — a comment is typed in the thread, a DM is composed in an empty box, which invites throat-clearing nobody would ever type into a comment.
Drafting from the visible conversation avoids most of it, because the draft inherits the register of the exchange rather than resetting to business English. If your DMs still read more formally than your comments, the fix is in your voice settings.
What it will not do
- It does not send. Ever, in any mode.
- There is no bulk send and no sequencing. One conversation at a time, by design — automated outbound messaging is what platforms restrict, and this product sits on the other side of that line.
- There is no auto-reply. Nothing fires when a message arrives.
- It does not read other conversations. Only the thread on your screen.
Limits
- Anything where being wrong is expensive — money owed, a complaint, a safety question — should be written by you. A draft is conciliatory by default, and conciliatory before you know what happened concedes a position.
- It does not know your history with a person beyond what is in the visible thread.
- Each generation costs credits. See Credits and usage.
More on this on the blog: DM Copilot articles