# Instagram comment-to-DM automation: the limits you design around

> Source: <https://dev.to/dosai/instagram-comment-to-dm-automation-the-limits-you-design-around-17k6>
> Published: 2026-10-10 19:04:07+00:00

"Comment PRICE and I will send you the list" looks like the simplest automation there is. A keyword, a direct message, done. I build DOS AI (dosai.pro), where businesses set up AI assistants for WhatsApp, Instagram and Telegram, and this feature ended up shaped almost entirely by what Instagram will not let you do.

Here are the limits and what we built around each.

A business cannot start a conversation in Instagram. Not by hand through a tool, not with a bot. The person starts it, with a direct message or with a comment.

So a comment is more valuable than it looks: it is permission to send one private reply. One. After that single message you are silent until the person answers.

In our dialog list such a conversation is marked "waiting for a message", and the reply box is closed. An open box that fails on send reads as a bug.

The practical rule for whoever writes the text: the first message must end with a question. "Here is our price list" ends the conversation. "Here are the prices, which day suits you?" starts one.

The private reply is a single message of up to 1000 bytes. In Russian or Kazakh that is about 500 letters, since Cyrillic takes two bytes per letter. The limit is in bytes, not characters, which is easy to miss if you only test in English.

Answering under the comment as well is good practice, but the same text under a hundred comments looks like spam to Instagram and can get the account limited. We require at least two variants for a public reply and rotate them.

We keep our own ceilings per project: 60 public replies and 200 direct messages an hour. When a post goes viral the rule keeps firing, and the replies above the ceiling wait.

A reply that cannot go now is never dropped. It stays in the feed marked "will go later" and is sent when the ceiling frees up, for as long as Instagram still accepts a reply to that comment, which is up to 7 days. A second comment by the same person does not queue a second reply.

The same queue covers the other reasons a send fails: a temporary error on Meta's side, or revoked access to the account.

Once the person writes to you, you may answer for 24 hours after their last message. The rule is the same for a bot, for a human operator in our dashboard and for any other third-party tool. After that the only way to reply is the Instagram app itself.

This is why Instagram is the one channel where our appointment reminders do not run: a reminder a day ahead cannot fit a 24 hour window that started at booking time.

A keyword matches from the start of a word. A rule on "price" fires on "pricelist". A rule on "book" does not fire on "facebook". Plain substring matching would fire on that second case.

One event is handled by one rule, and keyword rules win over the AI rule.

There is a rule that lets the assistant answer genuine questions under posts, such as "how do I book?", when the answer exists in the business's instructions or documents. It skips praise, emoji and spam, and it may not invent an answer.

Each checked comment costs the owner money, like any model call, so this rule has its own ceiling of 60 an hour and sits below the keyword rules.

Owners ask for "send it only to followers". It works: instead of the text the person gets a button, a follower gets the material, a non-follower gets a request to follow and a second button.

We describe the risk plainly in the guide. Instagram's anti-spam rules do not favour trading a follow for a reward, many accounts do it anyway, and after complaints reach can be cut. That is the owner's decision to make with the risk in front of them.

Most of the code here handles "not now": one reply only, not after 24 hours, not above the ceiling, not the same text twice. If you build on a messaging platform, list its refusals first and design the states for them, including what the user sees while a message waits.

The owner-facing description is in our guide: [rules and comments](https://dosai.pro/docs/guide/03-tools/automations).
