A seller I was watching had exactly one rule set up: if a DM contains the word price, send the price list. Clean idea. It fired about forty times a day and it made her feel organised.
Then somebody wrote: "hey how much is the beige one, do you have it in large?"
The bot sent the price list. All of it. Nine products in one image, no mention of beige, nothing about sizes. The customer read it, didn't reply, and went to look at somebody else's page.
Here's the part that bothers me. The seller never knew that conversation happened. On her dashboard it looked like a win: message received, message answered, reply time nine seconds. Green across the board.
That's the honest summary of most Instagram auto reply bot setups. They don't fail loudly. They fail politely, and they take the sale with them on the way out.
What a keyword bot is genuinely good at
I'm not going to tell you rule-based auto replies are useless. They're not. A keyword bot is a trigger and a canned response, and when the trigger is something you control, it's close to perfect.
- Comment-to-DM on a launch post. You told people to comment a word. They comment the word. The bot sends the link. Nothing ambiguous happened.
- Story replies with a code. Same logic — you set the vocabulary, so the vocabulary matches. This is the whole engine behind turning Instagram comments into paying customers.
- Off-hours acknowledgement. One honest line at 2am beats silence, as long as it doesn't pretend to be you.
- Repeat operational questions. "Where do you ship from?" has one answer and it never changes.
Notice what those four have in common: the customer is answering your prompt, not asking their own question. The second a buyer starts writing in their own words, the rule engine is out of its depth.
Where the rules quietly fall apart
Real DMs are messy. People type the way they talk, they ask two things at once, they send a voice note, they write "mojoode?" in half a language, they reply to your story with a single emoji and expect you to understand it.
Four failure modes come up over and over:
- The compound question. Price and colour and stock in one sentence. A keyword rule matches the first word it recognises and answers a third of the message. The buyer has to ask again, and a lot of them don't.
- The follow-up. "And the black one?" has no keyword in it at all. Rules have no memory of the last message, so the bot either says nothing or starts over with a greeting. Nothing kills a conversation faster than being greeted twice.
- Stale facts. Your canned reply says 890,000. You raised the price three weeks ago. Now you're arguing with a customer over a number your own bot gave them.
- Everything that isn't text. Voice notes, photos of a product from a competitor's page, a screenshot of a size chart. A keyword bot sees nothing at all.
Do the arithmetic on that seller's inbox. Forty DMs a day, and say one in five hits one of those four walls. That's eight conversations a day going nowhere — around 240 a month — and not one of them shows up as a failure anywhere. Her response-time chart stayed beautiful the entire time. I've written before about the automation mistakes that eat sales without leaving a trace, and this is the quietest one on the list.
What a good automated reply actually sounds like
Most auto replies are written like a notice board. "Thank you for contacting us. Our team will respond during business hours." That sentence has never sold anything to anybody.
A reply that works does three small things at once: it answers the part it can answer, it admits the part it can't, and it keeps the thread moving. Compare the two versions of the same beige jumper moment.
The notice board: "Hi! Thanks for your message. Please see our price list below."
The one that keeps the customer: "The beige one is 890,000 — large is in stock, small sold out yesterday. Want me to hold the large for you?"
Second version is shorter. It names the colour, gives the number, deals with the size question before it's asked twice, and ends with something the buyer can say yes to. That last part matters more than people expect: a reply that ends in a full stop ends the conversation, and a reply that ends in a question keeps it alive.
You can't write that by hand forty times a day, and you can't produce it with a keyword rule either, because the rule doesn't know which colour or which size was mentioned. It needs to read the sentence.
What Meta actually lets you do
This part gets skipped and then people are shocked when their page goes quiet. A few rules worth knowing before you automate anything:
- Use the official API. Anything that logs into your account like a person and clicks buttons on your behalf is against Instagram's terms and it's your account carrying the risk, not the tool's. Legitimate platforms connect through Meta's messaging API with your permission.
- The 24-hour window is real. After a customer's last message you can reply freely for 24 hours. Outside that window you can't just send whatever you like — the human agent tag extends it to seven days for genuine support follow-up, and marketing blasts are not what it's for.
- Comment-to-DM is fine. Cold DMs are not. Replying to someone who engaged with you is expected behaviour. Messaging people who never contacted you is how you collect reports.
- Blocks and reports are the actual penalty. Forget the policy document for a second: every annoyed person who mutes or reports you teaches the platform that your messages are unwanted, and reach follows that signal down.
The compliance summary is boring and short: answer people who spoke to you first, answer them quickly, and let them reach a human.
How I'd set this up if I were you
Skip the fifty-rule flowchart. Nobody maintains those past week three.
- Pull your last 200 DMs and read them. Not skim — read. You'll find your real FAQ in about twenty minutes, and it's never the list you assumed.
- Write the answers as facts, not scripts. Price, sizes, materials, shipping times, return window, payment methods. Facts can be reused in any wording; a script can only be sent whole.
- Keep rules only where you control the trigger. Comment codes, story keywords, welcome messages. That's it.
- Let something that reads full sentences handle the rest. This is the real split — rules for your prompts, an assistant that understands language for their questions. More on the difference in this walkthrough of Instagram auto replies.
- Connect stock and prices to the source. If the answer lives in a text box someone has to remember to update, it will be wrong by next month.
- Set your handoff rules before you turn it on. Not after the first complaint.
The line where the bot should stop talking
Every automated reply needs a point where it puts its hands up. Mine would be:
- Anything about money already paid — a refund, a wrong charge, a failed transfer.
- An angry message. Tone matters more than content here; a calm generic reply to a furious customer reads as contempt.
- A wholesale or bulk enquiry. That's a conversation you want to have yourself.
- The same question asked twice. If they're repeating themselves, the answer isn't landing.
And when it hands off, it should say so plainly — "I'm passing this to a colleague, you'll have an answer within the hour" — then actually deliver on that. A silent handoff is just a dropped conversation with extra steps.
Three numbers that tell you the truth
Response time is the vanity metric here. It measures how fast you sent something, not whether the something was useful. Track these instead:
- Fallback rate — what share of conversations got a generic or off-topic reply. Above 20% and your bot is mostly making noise.
- Silent drop rate — conversations where you answered and the customer never wrote again. These are your beige-in-a-large moments.
- Reply-to-order rate — of the DMs that got answered, how many turned into a paid order. This is the only number that pays rent.
Check them weekly for a month. The pattern shows up fast, and it's usually one or two question types doing most of the damage.
If you'd rather not spend your evenings maintaining keyword rules, that's exactly the gap Vardast fills — it reads the whole message, answers from your actual products and prices across Instagram, WhatsApp, Telegram and your site, and hands the conversation to you when it should. Same nine-second reply. Without the beige jumper walking away.