A 6-step setup for an AI-powered Instagram DM setter using ManyChat and Claude — responds in 3 seconds, qualifies leads, handles objections, and books calls 24/7.
This guide is reviewed for clarity, service accuracy, and AI-search readability. The next quarterly content review is tracked internally before unsupported metrics or client proof are added.
What an AI setter actually does
A human setter responds when they feel like it, works a fixed schedule, and costs $300–$600 a month plus taxes and management time. An AI setter responds in under 3 seconds, works every hour of every day, and costs roughly $10–$20 a month even at high volume.
The stack is simple: ManyChat handles the automation trigger (someone comments a keyword on your post or reel), Claude handles the actual conversation — qualifying the lead, navigating objections, and pushing toward the next step. At Hypd, the teams running this system are cutting setter costs by over 80% while booking more calls.
What you need to get started
- Instagram Business or Creator Account — must be connected to a Facebook Page for API access
- ManyChat account — free to start; handles automation triggers and message routing
- Claude API key — from console.anthropic.com, pay-per-use (roughly $0.05 per full qualification conversation)
- 15 minutes to configure everything end to end
6-step setup
These steps go in order. Each one is a prerequisite for the next.
- Step 1 — Create a ManyChat account at manychat.com and connect your Instagram Business profile. Grant permissions for Instagram messages and media.
- Step 2 — Get your Claude API key at console.anthropic.com. Generate a key, copy it immediately — it's only shown once.
- Step 3 — Connect Claude inside ManyChat: Settings → Integrations → find Claude AI → paste your API key → Connect.
- Step 4 — Build the automation: Automation tab → New Automation → Start From Scratch. Set trigger to 'User comments on your Post or Reel' with a keyword. Add a public reply (3–4 variations). Add a Send Message block as the private reply, then an Action block set to Claude.
- Step 5 — Write the prompt (see template below).
- Step 6 — Click Set Live, then comment on your own post with the trigger keyword. Verify the DM arrives and flows naturally. Have 3–4 people test it and refine the prompt.
The AI setter prompt — 5-section template
This is the most important part of the setup. Fill in each section with your real information — the more specific you are, the better it performs. Paste your best real DM conversations into the prompt, especially how you handle 'I can't afford it' or 'I need to think about it.' Claude absorbs your exact patterns and replicates them.
promptVOICE & TONE You are [YOUR NAME]'s AI assistant responding to Instagram DMs. Tone: [describe your energy — e.g. direct and warm, casual but professional, high-energy] Sentence length: [short and punchy / conversational / detailed] Slang or phrases you always use: [list them] Phrases you never use: [list them] Always sound like a real person texting — not a chatbot. YOUR OFFER Programs available: - [Program name]: [what it is, who it's for, price or price range] - [Program name]: [what it is, who it's for, price or price range] Payment options: [e.g. pay in full, 3-month plan, deposit + balance] Current promos: [any active offers or urgency] QUALIFICATION FLOW Ask these questions to separate serious buyers from browsers: 1. [First qualifying question — e.g. 'What's your main goal right now?'] 2. [Second question — e.g. 'How long have you been working on this?'] 3. [Third question — e.g. 'What's held you back from solving this so far?'] If they answer positively to all three, move toward booking a call. If they seem unqualified, be polite and direct — don't waste their time. OBJECTION HANDLING When they say 'I can't afford it': [your exact response] When they say 'I need to think about it': [your exact response] When they say 'I'm not sure it'll work for me': [your exact response] [Add any other common objections and your responses] CALL TO ACTION The goal of every conversation is to get them to: [book a call / click a link / fill out an application] Calendly / booking link: [URL] If they're ready: [exact CTA message] If they're hesitant: [softer CTA] INSTRUCTIONS - Ask one question at a time — never send a wall of text - Match their energy and response length - Never mention you are an AI unless sincerely asked - Never make promises about specific results - If a conversation is too complex, say: 'Let me get [YOUR NAME] to jump in here — one sec'
Pricing breakdown
ManyChat is free up to 25 contacts and 3 keywords. Pro plan starts at $15/month for 500+ contacts. Claude API runs roughly $0.05 per full qualification conversation — at 200 conversations a month, that's $10. A human setter at minimum costs $300/month plus management overhead. The AI version is 95% cheaper at any meaningful volume.
Advanced tips
- Paste your highest-converting real DM threads into the prompt — Claude absorbs your patterns with precision
- Add 3+ comment reply variations in ManyChat — rotating responses makes the automation feel human
- Watch real conversations for the first two weeks — identify where Claude misses, then refine the prompt
- When volume exceeds 25 contacts, upgrade to ManyChat Pro at $15/month
Frequently Asked Questions
Does Claude have to disclose it's an AI? Claude will not proactively tell leads it's an AI. However, if someone sincerely asks, include a fallback in your prompt that deflects gracefully — 'I'm an assistant for [your name] — want me to get them on here?'
What happens if a lead asks something the prompt doesn't cover? Include an escalation instruction: 'If a question is outside what you know, say: let me check with [your name] and get back to you.' This keeps conversations moving without Claude guessing.
How do I know if the prompt is working well? Monitor conversations in ManyChat for the first two weeks. Look for: where leads drop off, which objections Claude handles poorly, and which messages get the most responses. Refine the relevant section of the prompt each time you spot a pattern.
