There is a special kind of chaos that happens when your Instagram page starts growing and suddenly the same questions arrive all day long.

"How do I join?"

"Send the link."

"How do I get in the group?"

"Can I get the WhatsApp?"

At first, it feels manageable. Then one day you realize you are spending real human energy answering the same join question in three different places:

DMs
story replies
post comments

That is when the fun starts.

I wanted one automation that could handle the whole thing:

if someone sends a DM with a join keyword, reply
if someone replies to a story with a join keyword, reply
if someone comments join on a post, send them a private DM

Simple in theory. Extremely Meta in practice.

What I Was Actually Building

This project was for Lifestyle Hikers, a growing hiking community that needed a cleaner way to guide people into the group without making the admin team type the same welcome message forever.

The mission was:

catch the right Instagram events
only respond to join-type questions
avoid spam loops
keep the setup reliable enough that it would still make sense two months from now

The stack looked like this:

Meta app dashboard
Instagram API with Instagram Login
n8n
DigitalOcean

And the live workflow listened on a webhook that Meta could call whenever something happened on Instagram.

The User Experience I Wanted

I did not want a bot that answered everything like it had just discovered caffeine.

I only wanted replies when the person clearly meant:

join
link
group
whatsapp
how to join
signup
member
community

If someone says "beautiful hike" or "nice post," the bot should politely stay out of it.

That one rule matters more than people think. A lot of bad automation is not broken. It is simply too eager.

The First Big Lesson: Not All Tokens Are the Same

This is where the project stopped being "set up a webhook" and turned into "earn your stripes."

At one point I had a token that looked valid, felt valid, and absolutely refused to do the thing I needed it to do.

The workflow was receiving the messages. The keyword logic was matching. The send step was still failing.

The problem was not the workflow. It was the token type.

The working solution ended up being an Instagram Login token from:

Instagram
API setup with Instagram login
Generate access tokens

Not the first token Meta puts in front of you. Not the Page token from a different flow. Not the "close enough" token.

The exact token type matters.

That was the unlock.

The Second Big Lesson: DMs, Story Replies, and Comments Are Cousins, Not Twins

On the surface, they all feel like "someone contacted us on Instagram."

Technically, they are not the same.

DMs and story replies

These come in through the messages side of the webhook flow.

In practice, a story reply behaves like a message event with story context attached to it.

So if your DM automation works and your story reply is wired correctly, you can usually treat both with the same reply logic.

Post comments

Comments are different.

They arrive as comments events, and if you want to private-message the commenter, you have to reply using the comment ID, not just the sender ID.

That sounds like a small detail. Meta thinks otherwise.

The Third Big Lesson: Testing with Your Own Account Can Lie to You

This one was sneaky.

Some tests looked broken when they were not actually broken. They were just being run from the wrong type of account.

The most reliable test path was:

use a different Instagram account than the business/admin one
comment join on a post
check that account's DMs

That is how the comment-to-DM path finally proved itself.

It is a very Meta kind of sentence:

"The workflow was fine. The test user was the problem."

The Fourth Big Lesson: Bots Love Talking to Themselves

When I say I had a version that spammed replies, I mean it.

Meta sends back some outbound activity as events too. If you do not filter those correctly, your bot can see its own messages and decide, with full confidence, that it should answer them.

That is how you create a robot that gets stuck in a conversation with itself.

The fix was to ignore:

is_echo: true
non-message events like edits and reads
anything that is not a real inbound user message

Once those filters were in place, the workflow finally behaved like an assistant instead of a toddler with a megaphone.

The Final Working Setup

The finished system does this:

comment join on a post -> private DM
reply join to a story -> welcome message in DM
send join in a normal DM -> welcome message in DM

And the welcome message gives people the exact path they need:

follow the page
check the website
fill out the community form
show up to hikes
understand that WhatsApp access comes after real participation

Which is what good automation should do.

Not replace community. Support it.

Why I Like This Kind of Project

This is the type of work I love as a Creative Technician.

It is not just "make a thing work."

It is:

understand the real communication problem
design a system that feels human
survive the weird edge cases
document the fix so nobody has to relearn it the hard way

A lot of useful automation lives in that space.

Not glamorous enough to become a conference talk. Not trivial enough to hand-wave. Just real, practical, high-leverage work.

If You Want One of These

If you run a brand, community, event page, or membership-based business, you probably have at least one repeated conversation that should not still be manual.

That might be:

"How do I book?"
"Where is the price list?"
"How do I join?"
"Send me the link."

Those are perfect automation candidates.

The trick is not building a louder bot. The trick is building a calmer system.

And if Meta decides to make you earn it along the way, welcome to the club.