Auto DM is open. A rule attaches to one post on a connected Instagram or Facebook account, and when somebody writes the keyword you chose a private message goes out carrying a short link minted for that one send. Asking for a follow, asking for an email and sending a reminder afterwards are all optional steps.
Where an automation actually lives
The first thing worth understanding is where the rule sits, because almost every later answer comes out of it. An automation is not an account setting: it is a row pointing at one specific connected account, one specific post and one specific trigger. That trio is unique, so two rules can never fight over the same comment.
The post is picked from a list, with its thumbnail and its first line of caption in front of you. Nothing has to be copied from an identifier field and no URL gets pasted anywhere, which is where the most common mistake in this category comes from: binding the rule to the post next to the one you meant, and finding out a week later.
Comment rules also carry three scopes, and the third one is the one that changes a routine. This post, the one you just picked. Any post on the account, for a keyword you use all the time. Or the next thing you publish, which binds itself to a piece that does not exist yet, and which is how a launch going out on Friday morning gets set up on Thursday without anyone opening the app that day.
What each network opens
Instagram and Facebook do not carry the same things, and the difference is not an implementation detail but a question of what each platform exposes. It is worth seeing before you design a campaign on the wrong network.
| Comment carrying the keyword | Yes | Yes |
| Reply to a story | Yes | The event does not exist |
| Direct message carrying the keyword | Yes | Yes |
| Follow gate before the link | Yes | No |
| Reaction to the story before answering it | Yes | No |
The second row is the one most people assume works on both. Messenger has no story reply event, so a rule of that shape on a Facebook account is not half saved: it is refused at creation. A clear no is better than a rule that never fires and that you spend a month waiting on.
The first row carries a limit worth knowing before you build anything. Meta allows one single private reply per comment, and only for seven days, and a failed call spends it all the same. That is why the public reply under the comment and the private message travel together rather than as two loose steps: a public "sent you a DM" hanging off a thread with no message behind it is worse than saying nothing.
The keyword is the most underrated decision
An ordinary word turns the automation into noise. Pick "info" and anybody writing "can you send me info?" under a three week old post gets a commercial message they never asked for. A word of your own, easy to type and written into the caption with instructions, filters by itself and makes nobody guess.
Three settings around that word get used more than you would expect. Typo tolerance is on by default, because a misspelling is the normal case in a comment thread rather than the exception. Exclusions always win, even over the switch that answers every comment, so a word you never want to touch is written once and stops being a problem. And the variants of the public reply are picked from the send itself rather than at random, so a platform retry writes exactly the same words instead of leaving two different answers hanging off one comment.
The follow gate asks, and reads the answer
This is where most of the category overclaims, so it is worth being precise. With the gate on, the message asks the person to follow the account before the link goes out, and they confirm by tapping the button the message carries. That is their own statement, not a check run behind the scenes against a follower list.
Two practical consequences. The gate exists on Instagram and not on Facebook, where there is no relationship to read and switching it on would leave people waiting for an answer nobody can give. And the rule insists very little on purpose: it asks three times at most, counting the first, and then goes quiet forever in that conversation. Somebody who does not want to follow you gets the ask, two reminders and nothing else, which is the difference between a gate and a bot that argues.
The email step is equally restrained: it asks twice at most and then sends the link anyway. A contact is worth more than an email address, and somebody who has dodged the question twice has already answered it their own way.
The short link, and the message that arrives later
Every message carries its own link, shaped as gofeedapp.com/r/ plus a ten character code.
It is not one per rule and not one per destination: it is one per delivery. Because that
link was handed to exactly one person, saying "this send was opened" needs no cookie, no IP
address, no browser fingerprint and nothing that follows anybody outside the link itself.
The code uses an alphabet with no zeros, Os, ones, Ls or Is, because a short link ends up read aloud or copied by hand more often than anyone expects. The redirect is temporary rather than permanent, on purpose: the destination stays an editable field, so if you change the page you send people to tomorrow, the links already out in the world keep working and keep counting.
And there is an optional second message a few hours later, inside the 24 hour window Meta imposes, which can go only to the people who did not tap the first one and can carry its own link to somewhere else. It is the step most people add in the second month, once they have seen how many conversations stall halfway.
Everything in this article happens in one place in GoFeed.
Try it freeWhat it does not do, said early
It does not write for you. The message, the public reply, the gate and the reminder are all your own words, and there is no text generation anywhere in this product. It is also not a conversation tree with branches and conditions: it is a short sequence of steps that are switched on or off, and that limit is deliberate. The contrast with a tool built around the tree is in the head to head comparison.
It does not reach further than those two networks either, and it will not turn a TikTok or LinkedIn comment into a private message, because those platforms open no such road to anybody. And it does not replace a person answering: the thread lands in the inbox next to the comments and reviews from every other network, with its 24 hour clock on top, which is the same Meta window. The craft of replying without sounding like a machine is in managing your DMs.
What to set up in the first week
One rule, on a post that already works, with a keyword of your own and no gate and no email step. That shows you the floor of the funnel before you add the steps that narrow it, and the rule's own panel counts how many times it fired, how many messages went out, how many people passed the gate and how many tapped the link.
With that number in front of you, every step you add is a decision rather than a hunch. The gate raises followers and lowers taps. The email step raises contacts and lowers taps. The reminder recovers a slice of the people who never tapped. Which trade is worth it depends on what each of those is worth in your business, and only you can do that arithmetic.
The screen by screen setup is on the Auto DM page, and the detail of how each piece behaves underneath is in the message automation write up.
The contact ceiling stops nothing
The automation's contacts are the only soft quota in the product, and that means something different from what most of the market does. The workspace's other four quotas are hard ceilings: you reach them and that is the end. This one is not. Sends carry on, nobody is left without their link, and whatever goes past the ceiling shows up as a line on the invoice, counted per contact.
The ceiling is not on every tier either: the entry plan of each profile carries one and the rest do not, so for most workspaces the right sentence is not "careful not to overshoot" but that there is nothing to be careful about. The current figures live on the pricing page, which reads the live catalogue, and that is why none of them are printed here.