Answer Every Comment Without Typing Them

Step 2: write three versions of every reply, not one.

1. 1Write three different messages for each keyword and let the system rotate them.
2. 2This is the single biggest thing that separates a system that works from one that gets your account flagged. If forty people comment the same word and all forty get a identical message, word for word, your comment section reads like a bot convention.
3. 3Keep them short. Two lines and the link. Nobody reads a paragraph they did not ask for.
4. 4Write them so they make sense in public. More on that in step 3.

Step 3: answer in public AND in private.

1. 1Every keyword comment should get two things. A short public reply under the comment, and the actual delivery.
2. 2The public reply is the part most people skip and it is the one that compounds. Everyone else scrolling the comments sees that you answer. That is social proof you cannot buy.
3. 3Here is the trap. Instagram and Facebook let you send a private message. YouTube, Threads and LinkedIn do not have that door, so whatever you wrote as the private message posts publicly instead.
4. 4So write every message as if it will be public, because on half your channels it will be. Never write "sent to your inbox" or "check your messages". Write "here it is" and the link.

Step 4: make the replies sound like you.

1. 1If you let AI write the replies, it will get clocked. Perfect grammar, three neat clauses, and one word you have never said out loud in your life.
2. 2Fix it with a file. Write down how you actually talk. What you say, what you never say, where you break your sentences, the words you overuse.
3. 3Feed that file to the model before it drafts anything. I keep mine as VOICE.md and it goes into every prompt.
4. 4I wrote a full guide on this. It is linked at the bottom of this page.

Step 5: the four rules that keep it from going wrong.

1. 1Never reply to yourself. Match on the commenter account id, not on a list of handles. Handle lists break the day you rename, and then your automation has a conversation with itself in public forever.
2. 2Only auto answer fresh comments. The age of the COMMENT decides, not the age of the post. A 48 hour window is about right. Someone commenting on a two year old video still deserves an answer, but a bot answering a two year old thread at 3am looks exactly like what it is.
3. 3Answer each person once per post. Dedupe on the post, not globally, or a repeat viewer never gets an answer on your next video.
4. 4Have an off switch and test that it actually works on every path. A kill switch that only covers one of your three send paths is not a kill switch.

Get these three things first. None of them cost anything.

1. 1An Instagram professional account, either Business or Creator. You switch it in your settings and it takes about a minute. A personal account has no API at all, so nothing below this line works until you do that.
2. 2A developer app at developers.facebook.com. Free. You are making one for yourself, not shipping it to an app store, and it never leaves your account.
3. 3Somewhere for it to run on a timer. I use a Vercel cron job on the free plan. Any machine that stays on works.

Set up the Meta side. It is the boring part and you only do it once.

1. 1Go to developers.facebook.com, create an app, and connect your Instagram professional account to it.
2. 2Ask for two permissions. instagram_business_basic and instagram_business_manage_comments. Those two cover reading your comments AND sending the private reply, which surprised me, because the DM half sounds like it should need a messaging permission of its own.
3. 3Generate an access token. Put it in an environment variable and never paste it into a file you commit. If that sentence means nothing to you, tell Claude Code exactly that and it will walk you through it.
4. 4Grab your Instagram account id. Claude Code can pull it in one API call.

Here is what the thing actually does, in order.

1. 1Every minute, pull the comments on your recent posts.
2. 2Look for your keywords in each one. Ignore capitals, and match the word even when it is buried in a sentence, because plenty of people type "send me GUIDE please" instead of the word on its own.
3. 3Skip it if the commenter is you. Match on the account id, not on your handle.
4. 4Skip it if you already answered this person on this post.
5. 5Send the private reply, picking one of your three variants at random.
6. 6Post the public reply under the comment.
7. 7Write a row in a log. Who, which keyword, which post, which variant, and what came back.

Test it somewhere harmless before you put a keyword in a real caption.

1. 1Put up a throwaway post. Comment your keyword from a second account and watch the log.
2. 2Comment the same word again from that same account. Nothing should go out the second time. If it sends twice, your dedupe is broken and you just found it on a post nobody is looking at.
3. 3Comment from your own account. Nothing should happen at all.
4. 4Flip the kill switch on and comment again. If anything still sends, your kill switch is not covering every path, which is the entire reason you test it.
5. 5Now put a keyword in a real caption.