From @rajeevkugan
My outlier finder.
A small tool I had Claude build. It tracks the creators in my niche, scores every reel against that creator's own average, flags the ones that were probably paid, and transcribes the real winners. Here's how to build yours.
The idea
Views alone don't tell you much
01
Score it against their usual
Score = views ÷ that creator's median views. A small account hitting 40× its normal found something. A huge account hitting 2× is just having a good day.
02
Flag the likely paid ones
Ad views land on the post, but ad likes and comments mostly don't. If a reel scores 3× or more and its engagement is under 25% of that creator's usual, it gets marked likely paid and hidden.
03
Study the real winners
Transcribe the outliers that are left, work out why they worked, and make your own version. Your idea, your words, your angle.
Setup
Build yours in Claude Code
- Use your own Instagram account. It has to be a Business or Creator account connected to a Facebook Page. The tool can only look up other Business and Creator accounts.
- Get your own Meta access token. Create an app at developers.facebook.com and generate a token with instagram_basic and pages_read_engagement, plus your Instagram account ID. Check Meta's current docs for the exact permissions.
- Paste the prompt below into Claude Code. It builds the tool and walks you through adding your token and ID.
- List the creators you want to track in channels.txt, then run fetch and outliers.
Build me a small Python tool called ig_outliers.py that finds viral outlier reels from Instagram accounts I track. It is read-only. Rules - Use the Instagram Graph API's Business Discovery endpoint with MY OWN Meta access token and MY OWN Instagram Business or Creator account ID. - Read them from the environment variables IG_TOKEN and IG_USER_ID. Never hard-code them, never print them, never commit them, and add .env to .gitignore if you use one. - Never post, comment or change anything on Instagram. - Store everything in a local SQLite database: creators, reels, and a snapshot of views, likes and comments on each fetch. - Read the usernames to track from channels.txt, one per line. Commands - fetch [--pages N]: pull recent reels for every username (default 2 pages of 50). Keep going if one account fails, and report it. - outliers [--min 3] [--limit 25] [--include-paid]: print the top reels by score. - report [--min 2]: write a sortable, filterable report.html. - channels: list tracked creators with followers and their usual views. - transcript <url>: download one reel's audio with yt-dlp and transcribe it with a local Whisper. Scoring - Score = a reel's views divided by the median views of that creator's reels older than 3 days. Skip creators with fewer than 5 such reels. - Show reels younger than 3 days but mark them "fresh", because their views are still climbing. - Engagement = (likes + comments) / views. - Mark a reel "likely paid" when its score is 3 or more and its engagement is under 25% of that creator's usual (median) engagement. Hide those from `outliers` unless --include-paid is passed, and always label them "likely", because it is only an estimate. Then walk me through setting IG_TOKEN and IG_USER_ID, and run fetch on a short channels.txt to prove it works.
python3 ig_outliers.py fetch python3 ig_outliers.py outliers --min 3 python3 ig_outliers.py report
Want it done for you?
If you want scroll-stopping video for your brand, that's what we do at FROZN Media. DM me @rajeevkugan.
DM on Instagram ↗API permissions, commands and URLs can change. Check each tool's own docs if a step doesn't match.
