How to Spot Fake Followers on Instagram

“Spotting fake followers” suggests opening an account’s follower list and checking each one. We do not do that, and it is the first thing to say: we read no follower list.

Some tools go further and sort the followers themselves into categories. The figure they show is not a count: it is a statistical inference over a very large graph of collected accounts, and it needs the list to be readable. When it is not, there is no figure to give.

What our signals see is different, and useful anyway: a bought audience distorts the proportions of the account hosting it. An account that tripled overnight without its reach moving, or that has a hundred thousand followers and forty interactions per post, tells you so without meaning to.

The model starts at a hundred points and subtracts a penalty for each rule broken. Every rule is shown to the creator with its reason: a score without its "why" cannot be checked, so it is not useful.

The trigger thresholds and the weight of each rule are not published, and that is deliberate: publishing them would tell anyone who wants to game the score exactly what to avoid. Here is what we look at, with no exception:

• the ratio between accounts followed and followers,
• engagement abnormally low for its cell,
• engagement abnormally high for its cell,
• the ratio between comments and likes,
• a missing biography,
• too few published posts,
• a follower-growth spike.

All of them read from the outside, on public data. None assumes access to the follower list.

Very high engagement can signal a bought audience or an engagement pod. It can also signal an excellent creator. Telling the two apart from outside is hard, and being wrong costs asymmetrically: rejecting a very good creator is a silent loss, while letting a doubtful account through shows up in the first campaign.

So the threshold sits at several times the third quartile of the creator’s own cell — not the median, not the quartile. A merely remarkable creator never crosses it. The code comment says it in those words: a merely excellent creator is not a fraud.

Our growth-spike detection carries a clause we consider the best line in the model: when reach is unknown, it flags NOTHING.

The reason is simple and worth writing down: a bought follower spike and a viral day are indistinguishable if you cannot see views. The bought spike comes with no rise in reach; the viral day does. Without that data, flagging would be guessing.

The rule also requires at least a week of daily records. Less than that is not a trend, it is a point.

Past a small number of rules breached, the authenticity score is capped, whatever the penalty arithmetic says: the account can no longer make up its score by compensating elsewhere.

Why a cap rather than a sum: because the signals are not independent. An account tripping several rules does not deserve an “average” score obtained by compensation — it deserves a look. The cap forces that look instead of drowning it in an average.

• They are heuristics, not proof. A flagged creator is a creator to look at, not one to reject.
• The automated-account signal exists on Instagram only. On the other two platforms it stays UNKNOWN — and an unknown is never displayed as a zero.
• None of these signals reads the follower list: we score the account. Sorting followers into categories requires an inference over a graph we do not build.
• The thresholds are tuned to be conservative. They therefore let doubtful cases through, by construction — the accepted trade-off against the risk of accusing wrongly.
• An account can have fake followers without being responsible: unsolicited follower deliveries exist, and our signals do not distinguish purchase from sabotage.

BriefBench shows these rules with their reason on every profile, rather than a bare score — so you can judge, not merely believe us.