How the analysis works

Published in full, because a vetting tool that will not show its working is asking for the same blind trust it is meant to replace.

Where the data comes from

Everything is retrieved from official platform APIs. Instagram data comes through the Instagram Graph API, which returns public profile and post metrics for Business and Creator accounts. YouTube data comes through the YouTube Data API v3, which returns public channel and video statistics.

We do not scrape, we do not buy panel data, and we do not ask the creator to connect anything. If an Instagram account is set to Personal, no API exposes it and we cannot analyse it. No credit is charged in that case.

The five signals

No platform publishes a list of an account's followers, so nobody can directly count real versus fake. What is observable is behaviour. Real audiences watch, like and comment. Purchased ones do not. Each signal below is measured, scored, and weighted.

Engagement vs tier

35%

Engagement rate measured against the benchmark for accounts of that size.

A 2% engagement rate is strong for a 500K account and weak for a 5K one. Comparing against a flat number punishes large accounts and flatters small ones, so we compare each account to its own tier.

Comment quality

20%

Comments as a proportion of likes.

This is the cleanest fake-engagement signal available. Likes can be bought for pennies. Comments that read as human are far harder and more expensive to fake, so bought engagement shows up as a lopsided ratio. Below roughly 0.5% is a floor; a healthy band runs from about 1.5% to 25%.

Reach efficiency

25%

Typical views or impressions relative to follower count, smoothed against a platform baseline.

Bought followers do not watch. An account whose reach has collapsed relative to its size is showing the clearest symptom of an inflated follower count.

Posting consistency

10%

Average posting frequency across the recent sample.

A large account that has not posted in months is either dormant or never built a real audience habit. Weighted lightly, because plenty of good creators post infrequently on purpose.

Audience integrity

10%

Followers relative to accounts followed.

Follow-for-follow and mass-follow growth leave an obvious fingerprint. Capped, so a very high ratio confers no extra advantage.

Signals for YouTube are weighted differently, because the view-to-subscriber relationship carries far more information there than it does on Instagram.

Flags

Separately from the weighted signals, a handful of specific patterns are checked and reported as triggered or not. These are the combinations that rarely occur naturally.

Engagement collapseEngagement far below the benchmark for the account's size.
Zero comments at scaleA substantial account with effectively no comment activity.
Bought viewsReach far above what the follower count supports, with almost no engagement.
Mass-follow patternFollower-to-following ratio consistent with follow-for-follow growth.
Fresh-account anomalyVery few posts combined with engagement far above the benchmark.

The written analysis

The computed signals are passed to a large language model, which writes the plain-English read: what the combination indicates, which reading is more likely when two signals conflict, and what is worth checking by hand. Only the computed metrics are sent. Your account details are not.

What this cannot tell you

  • It cannot prove that a specific follower is fake, or that a creator bought anything. It reports patterns, and patterns have innocent explanations.
  • It does not produce a score, a rating or a grade, and it does not rank creators against each other on a single number.
  • It cannot see private data: follower demographics, story metrics, or anything the platform does not expose publicly.
  • It cannot tell you whether an audience will buy. Authenticity and commercial fit are different questions.

Ostryno gives you the lens. The call is still yours.