What is the single strongest signal of an inauthentic audience?

There is not one, which is the point. Fake-follower estimates, engagement ratios, comment quality, follower growth shape and audience geography are each consistent with several explanations. Confidence comes from several signals pointing the same way plus a look at the content itself.

How far back should you review a creator’s content?

Far enough to see how they behave over time and how they handled anything that went wrong, which usually means more than the recent posts a profile surfaces. Old content is judged by different standards than new content, and a considered response to a past mistake is different evidence from the mistake alone.

Who should make the brand-safety decision?

Someone with the authority to say no and a written standard to apply, rather than whoever is running the campaign under a deadline. The standard matters more than the reviewer: applied case by case under launch pressure, the answer tracks urgency instead of risk.

Creator vetting and brand safety

How do you check a creator properly before signing them?

10 guides · Last verified September 2, 2026

What this part of the work is

Vetting exists to answer two different questions that are usually run together. The first is whether the audience is real, which is a data question. The second is whether the association is one the brand can carry, which is a judgement question about a person’s public conduct. They need different evidence and different people, and collapsing them produces a process that checks follower authenticity carefully and brand risk barely at all.

These guides cover reviewing a creator’s actual content rather than their profile, reading engagement for authenticity without over-reading it, interpreting the signals that suggest inauthentic activity, understanding what past sponsorships tell you, spotting conflicts, weighing the quality of the evidence you have, and deciding when the picture is incomplete.

The line held throughout is that a vetting finding is a probability, not a verdict. Most of the signals available are consistent with several explanations, and a process that treats one anomalous number as proof will reject good creators as confidently as it accepts bad ones.

Where it goes wrong

The failures that recur here

01
Anomalies have innocent explanations more often than not
A follower spike can be a purchased audience or a video that was recommended widely. A low engagement rate can be an inflated denominator or a platform that distributes differently at that scale. The diagnostic value is in the pattern across several signals and in the content itself, not in any single number crossing a threshold.
02
Brand safety review is mostly not automatable
Tools surface keywords and flag terms, and the things that actually damage a brand association are usually contextual: who someone associates with, how they handled a past controversy, what a joke was actually about. Somebody has to watch the content and read the comments, and the review has to be recorded so the decision can be explained later.
03
The incomplete case is the normal case
Contact routes are missing, audience data is thin for smaller creators, and some platforms expose almost nothing. Waiting for complete evidence means never deciding, so the practical requirement is a documented standard for what is enough, what is disqualifying and what gets escalated, applied consistently rather than per creator.

Every guide in this section