Method · v2 · 16 November 2026
How the score works
Followproof's audience quality score estimates how much of an account's audience is real and active. This page lists everything that goes into it, so you can check our reasoning and know when not to rely on it.
Every result is an estimate from public signals. It is not a finding about the account or the person behind it, and it can be wrong. Each result has a Report a wrong result button; a person reviews every report.
What we look at
- Instagram: the public profile (followers, following, posts, category, business country if listed) and the 12 most recent posts with likes, comments, Reel views, paid-partnership labels and captions.
- TikTok: the public profile (followers, following, lifetime likes, videos) and up to 20 recent videos with views, likes, comments, shares and duration.
- YouTube: the channel (subscribers, total views, videos, listed country) and the 15 latest long-form uploads with views and approximate publish dates.
Data is fetched when someone checks the account and cached for up to three days. Every result shows whether it was live or cached and the date.
What we can't see
Follower lists, reach and impressions, audience demographics, story views and saves are only visible to the account owner. So the score is an estimate from behavior, not a count of fake accounts. It cannot tell bought followers from people who stopped using the app, and it can be fooled by an account that buys likes and comments in proportion. When a decision matters, ask the creator for analytics screenshots.
The seven signals
The score starts at 100. Each signal can deduct up to its weight:
| Signal | Weight | Rule |
|---|---|---|
| Engagement vs. similar-size accounts | 30 | The geometric mean of average and median engagement per post (likes + comments, + shares on TikTok) ÷ followers, compared with the typical rate for the account's size tier (eight tiers, up to 50M+). YouTube uses views ÷ subscribers. Under 60% of typical starts losing points, under 30% loses 15–30. On Instagram and TikTok, a median more than 5× typical costs 6 points (pods, giveaways, bought likes). |
| Video reach vs. followers | 15 | Average Reel or TikTok views ÷ followers, with thresholds per size tier (for example under 3% loses 15 and under 8% loses 7 for Instagram accounts below 1M, under 0.3% and 1% for 50M+). |
| Following-to-follower ratio | 15 | Accounts over 5K followers: ratio above 1 loses 15, above 0.5 loses 8, above 0.1 loses 3. Smaller accounts only lose 6, above 1.5. |
| Comments per 100 likes | 15 | Under 0.3 loses 12 and under 0.7 loses 6 (lower thresholds for accounts over 1M and 5M). Over 20 loses 4: that is usually comment-for-the-link automations or giveaways, so it never counts as fake-follower evidence. Skipped when comments are turned off. |
| Variation between posts | 10 | Coefficient of variation of likes (Instagram) or views (TikTok, YouTube) across 6+ posts. Under 12% loses 10, under 20% loses 4. |
| Posting activity | 10 | Days since the last post: over 60 loses 10, over 30 loses 5. |
| Follower count vs. content | 5 | TikTok accounts over 10K with fewer lifetime likes than followers, or accounts over 50K followers with fewer than 15 posts, lose 5. |
The score is clamped between 5 and 99. If the account is private or we have fewer than three posts, we show "Not enough data" instead of a score.
When we estimate a fake share, and when we don't
Low engagement on its own is not evidence of fake followers. Big and old accounts, brands and institutions collect followers who stopped engaging years ago, and engagement per follower falls steeply with size. So the signals are grouped by what they measure:
| Group | Signals |
|---|---|
| Reach | Engagement vs. size, video reach |
| Like and comment pattern | Very few comments per like, suspiciously even engagement |
| Follower growth pattern | Following-to-follower ratio, followers vs. content |
A group raises a concern when it loses at least 40% of the points it can lose. Posting activity is shown but never counts as evidence of fake followers.
| What we find | Label | Estimated fake or inactive share |
|---|---|---|
| No group raises a concern | No red flags | Not estimated |
| Only one non-reach group raises a concern | One warning sign | Not estimated; score can't fall below 60 |
| Only the reach group raises a concern | Low engagement for its size | Not estimated; score can't fall below 60 |
| Two or more independent groups agree | Several signals agree | Score 60+: 10–25% · 45–59: 20–40% · below 45: 30–55% |
For accounts whose public category marks them as a company, organization, media outlet, government body or public figure (or whose website is a .gov, .mil or .edu address), reach signals can cost at most 15 points: creator benchmarks don't describe how their followers behave.
Confidence
Confidence describes what we sampled, and it is never "high": we see at most 20 recent posts and public counts, and never a single follower. It is medium with 10+ posts and 5+ signals for a creator account under 20M followers, otherwise low. Every result explains its own confidence level.
Public test panel
We run the scoring on 26 public accounts of every size and kind whose audiences are not in doubt: creators from 1.5K to 21M followers, brands, an institution and celebrities up to 680M followers. None of them may get a fake-share estimate; this is part of our automated tests and runs on every change to the method. Data fetched 16 November 2026; the "kind" column is our own label, not the score's input.