Both prompts, every tool with my discount codes, the send limits that keep your account alive, and the setup in order. Work top to bottom and you will have it running today.
Filter Instagram profiles by the words in their bio and their follower range, or scrape who an account follows and who comments on its posts.
Pull each profile, run it through the qualifier prompt, then research whoever passes so the opening line is about them and not about you.
The copywriter prompt writes one unique message per person in a fixed five line format, with a list of words it is not allowed to use.
Send in small batches on a human schedule, then answer replies, including threads from months ago that Instagram normally locks you out of.
These are the two I actually run. The first decides who is worth messaging. The second writes the message. Copy them exactly.
Feed it one scraped profile. It returns PASS or FAIL, a one line reason, and a cleaned up bio that will not break the next API call.
{{ $json.* }} lines are n8n expressions that pull fields off the scraped profile. If you are running this in Claude Code or another agent instead, replace each one with the actual value.Give it the qualified profile plus your research. It writes one message in a fixed five line format. The format is doing most of the work, so the AI only has to fill in the compliment and the one specific way this person could use the system.
You do not need all of these. The list, and what each one is actually for.
Replaces the whole scraping step. Filter hundreds of thousands of Instagram profiles by bio keyword, include and exclude terms, follower range and post count, then export a CSV. This is the fastest way to get a qualified list.
Open ConsultiRuns the two actors below. Sign up through this link and the discount code applies to your plan.
Open ApifySends the messages from your account using your logged in browser session. Several actors do this job. If this one is ever broken or overpriced, pick another with strong reviews that is actively maintained.
Open the DM senderPulls the bio, follower counts, category and privacy flag for each account, which is exactly what the qualifier prompt reads.
Open the profile scraperThe piece most people are missing. It rebuilds the private Instagram and LinkedIn messaging APIs, so your agent can read and answer conversations of ANY age, comment on posts, and reply to comments. The approved tools cut you off after 24 hours. It also keeps the connection logged in for you.
Open UnipileThe research pass. It finds the one true detail about the person that the first line of the DM is built on.
Open PerplexityOptional. It runs the steps on a schedule so you never think about it. You can hand the same prompts and rules to Claude Code or any agent instead and get the same result.
Open n8nThis is the part people skip, and it is the only part that decides whether you still have an account next month. None of the rest matters if you get this wrong.
This is the number. Do not test what happens above it on an account you care about.
Break the day into small batches at uneven times instead of one large send.
A random gap in that range reads as a person typing. A fixed delay reads as a script.
The copywriter prompt already handles this. Every message is written for one person.
Use a residential IP in the country you actually log in from. Mobile IPs are better. Apify does not sell mobile proxies, so bring your own if you want them.
These caps are for cold first messages. Answering someone who already wrote to you is far safer and does not need the same restraint.
The honest version. This is against Meta's terms of service. ManyChat and GoHighLevel automations run on endpoints Meta approves and hands out. This does not. It uses your own logged in session, which is why the limits above exist and why there is real risk to the account.
I have run this for over a year across multiple accounts without a problem, but I also pay for verification, spend real money on ads, and use the accounts normally every day. If your account is brand new, unverified, has never run an ad, and has no normal activity, you are the profile that gets suspended. Get the badge, run some ads, be a real user first.
Four stages. Open each one as you get to it.
The free AI Automation Insiders community is where people building this exact system compare notes on what is working, which actors are healthy, and what is getting accounts flagged.
Join The Free CommunityEverything above is the system explained. Inside AI Automation Insiders you download it instead.
Lifetime access, including everything added later.