Let AI Answer Your DMs. Keep the Conversations That Matter.

"hi how much is it"

Sent at 2am, under the post that took off. Your DMs (direct messages) and comments are arriving faster than you can answer, and half of them ask the same thing. By the time you're awake to reply, some of those people have gone.

AI can send the first reply, answer what you've answered a hundred times and keep the door open until you're back.

The relationship stays yours, though. I won't hand you my bot (an automatic reply program). You'll build one from your own conversations, so the bot sounds like you and knows when to step aside.

Write the two lists

Start with a note, with no software yet.

The first list is the questions you answer again and again. Scroll back through your last 20 DMs and they'll jump out. The second list is the messages you'd want to read yourself, such as a complaint, a big order, or someone who sounds upset. Add anything where a wrong answer would cost you more than a slow one.

Here is my split, with my name in it.

Swap my name for yours when you write your own. Most setups skip the right-hand column, but that column is the one that protects your reputation. Everything after this follows the two lists.

Build the bot's rules from your own replies

One set of written instructions runs the assistant. If you borrow someone else's set, the bot sounds like them, sells their products and follows their rules.

Take your last twenty DM conversations, delete the names and paste them into ChatGPT or Claude along with your two lists.

These are twenty real conversations from my Instagram DMs, with names removed. I've also added two lists. One is the questions I get again and again. The other is the messages I always want to read myself. Work out from the conversations how I talk, what I get asked, and how I answer. Then write the instructions for an assistant that handles the repeat questions the way I do.

When you have enough, write the instructions for the assistant. Keep replies short. Only use facts I've given you. If the assistant doesn't know something, it should say I'll reply personally. Anything on my hand-back list gets one line saying I'll be in touch, and nothing else.

Before you finish, show me five replies the assistant would have sent to real messages from this batch. Put each one next to what I sent, and ask what I'd change.

Those five side-by-side replies are where you catch the bot out. Maybe the bot is chirpier than you, or it offers a discount you've never offered. Fix the instructions until the bot's replies could pass for yours. Allow two rounds of fixes.

If the bot still doesn't sound like you, let the AI interview you instead and ask it this. "Ask me one question at a time about how I talk to customers, what I'd never say, and where I draw the line. Keep going until you can write the instructions."

Your answers, in your words, usually fix what the examples couldn't.

Remember that the bot knows only what you've told it. If you change a price but not the instructions, the bot quotes the old price and sounds sure about it.

Plug the bot into Instagram

You need three pieces, and none of them involves code.

1. An Instagram Business or Creator account. Switching is free in your Instagram settings.
2. A DM tool that sits between Instagram and the AI. Manychat is the common choice and an official Meta partner. Its AI features need a paid plan, so check the current pricing before you start.
3. A way to plug the AI in. One option is your own ChatGPT or Claude API key, a private code that lets the tool use the AI. With your own key, you pay per reply. The other option is the tool's built-in AI add-on. On your own key, a short reply usually costs a fraction of a penny.

Then build one automation (a rule the tool follows by itself). When a message doesn't match your keyword replies, the tool sends it to the AI with your instructions and posts the answer back.

Menus change often, so take the exact clicks from Manychat's own help pages.

There's a monthly fee for the tool and a charge per reply for the AI, which are small sums but real ones.

Send the serious ones back to you

Your instructions already make the bot say you'll reply personally, so now make sure you do.

Add a check after the AI's reply. If the reply contains your hand-back line, the check stops the automation, tags the conversation and sends the conversation to your inbox. Most DM tools can assign a conversation to you and mark it as needing a human.

The hand-back matters more than any clever reply, because AI can sound sure and still be wrong.

Keep your keyword replies separate. These are the set replies sent when someone types a chosen word. Someone who types a word to get your free guide should get the guide, and no chat with a bot.

Message yourself and try to break the bot

Before you post anything that sends people your way, message your own account from a second one.

• Ask a normal question and read the answer. Is it correct, and is it in your voice?
• Ask about something you don't sell. Will the bot admit that it doesn't know?
• Send a complaint and see whether the bot hands back or tries to sell.
• Type your keyword and check that the guide arrives.

For the first week, read every conversation the bot has. Judge the bot on what the bot did, whatever you told the bot to do. If something went wrong, fix the instructions too.

Tell people they're talking to an assistant and that you'll pick up anything important. Most people don't mind a quick automatic answer. What they mind is finding out later.

The bot answers "how much is it". You answer the person who asked.