How accurate is the verdict?

Accuracy comes from breadth: roughly thirty independent tells, weighted, judged against the live category baseline. A channel has to fail many at once to score red, and any sample too thin for a confident read returns UNCLEAR rather than a false red. Every measurement is shown.

Can a real channel be wrongly flagged as fake?

It is what the method is built to avoid. No single signal condemns a channel, thresholds track the category rather than a fixed line, and an inconclusive sample is called UNCLEAR, not FAKE. A genuinely engaged room passes comfortably.

Does the streamer know they are being audited?

No. The sampler joins the public chatroom the same way any viewer’s browser does. Nothing is installed and no cooperation is required.

What if I disagree with a verdict?

Read the reasons and measurements on the certificate and re-run the check while the channel is live. A verdict is a point-in-time, evidence-weighted opinion, and the evidence trail lets you judge the call rather than take it on faith.

ProvenAudience / How it works

Methodology · What we measure and how we stay honest

How the audit works — and how we keep the verdict honest.

A verdict is only worth as much as the method behind it. This is the whole method: how the sample is taken, what the roughly thirty signals actually measure, how a score becomes a green, yellow or red verdict, and — the part most tools skip — how we guard against calling a real channel fake.

Run a free check The signals in detail →

The method — three movements

i.
Resolve the channel
You paste a kick.com channel, or the desk audits the top of the Slots & Casino category on its own. We resolve the live chatroom over the same public connection any viewer’s browser uses — nothing installed, no cooperation from the streamer, no notice that an audit is running.
ii.
Sit in the room for ten minutes
A sampler joins the live chat and records every message — who sent it, the exact millisecond, the account id and badges — while polling the viewer count in parallel. Ten minutes is long enough to read the room’s behaviour and short enough to stay quiet. A free public check runs a five-minute sample.
iii.
Score the evidence
The sample is scored across roughly thirty forensic signals, each one cheap for a real audience to pass and expensive for a farm to fake. No single signal condemns a channel; the score is the weight of the evidence, and it comes back as a verdict with the reasons written in plain language and the measurements attached.

The signals — five families, ~30 measurements

•
Engagement
Chatter-to-viewer ratio judged against the live category baseline, message rate, and whether the chat actually reacts when the viewer count moves.
•
Account forensics
Kick ids are sequential, so batch-registered “fans” cluster at the top of the id space and inside narrow signup windows. We measure the fresh-account share and the cluster tightness.
•
Behaviour over time
Per-chatter inter-message timing fingerprints, burstiness, and shared or repeated script banks — including phrases that match another channel’s transcript, which exposes one script running two rooms.
•
Money & history
Subscriptions and paid-badge share, follower-spike-then-flat analysis, the VOD cross-check (bots don’t watch replays), and consistency against months of the channel’s own recorded baseline.
•
Coherence
Off-platform footprint, language and geo mix, and sponsor-title history — a channel showing thousands of viewers with no trace anywhere else is incoherent.

The full signal list and the reasoning behind each one lives on the viewbot-detection page. Thresholds are tuned against the live Slots & Casino category, not a fixed rulebook a seller can reverse-engineer.

The verdict — three bands and a confidence

Green, yellow, red — with the reasons attached.

The score maps to three bands. PROVEN (green) is a room that behaves like a real audience across the signals. UNCLEAR (yellow) is a channel the sample could not settle — too few viewers to judge, a short or interrupted broadcast, or genuinely mixed evidence — and it says so rather than guessing. FAKE (red) is a channel that fails the tests money cannot buy its way out of. Every verdict carries a confidence and the plain-language reasons, so you are reading evidence, not trusting a black-box number.

ProvenFake audience

Staying honest — the part most tools skip

How we avoid calling a real channel fake.

A fraud tool that cries wolf is worse than none, because it teaches you to ignore it. Three things guard against false positives. First, no single signal condemns a channel — a quiet chat or a batch of new accounts is a data point, not a verdict; the score is the weight of many independent tells, so one noisy reading cannot sink an honest room. Second, we judge against the live category baseline, not a fixed line, so a channel is measured against its actual peers rather than an arbitrary cut-off a small real streamer might trip. Third, a sample that cannot support a confident read returns UNCLEAR, not a red — low viewer counts, short broadcasts and interrupted samples are called out as unresolved rather than punished.

A verdict is an evidence-weighted opinion formed at a point in time, not a statement of legal fact — and we say so, because a tool that pretends otherwise is one you should trust less, not more. What we can stand behind is the evidence: every measurement that produced the verdict is on the certificate, so you can read it and judge for yourself.

The boundary

An audit, not an exposé.

The certificate carries verdict-level evidence only: the score, the live viewer and chatter figures, the ratios, the account-age and language mix, and the reasons. Individual chatter identities, raw transcripts, and any reasoning that names other channels stay behind a privacy boundary by construction. It is a document for the person about to wire money, not a naming-and-shaming board — which is exactly why a compliance team can file it.

See a real verdict — free

FAQ

How accurate is the verdict?: Accuracy comes from breadth, not any one signal: roughly thirty independent tells, weighted, judged against the live category baseline. A channel has to fail many of them at once to score red, and any sample too thin to support a confident read returns UNCLEAR rather than a false red. Every measurement is shown, so you are never asked to trust the number alone.
Can a real channel be wrongly flagged as fake?: It is what the method is built to avoid. No single signal condemns a channel, thresholds track the category rather than a fixed line, and an inconclusive sample is called UNCLEAR, not FAKE. A genuinely engaged room — real chat ratio, bursty timing, mixed account ages, proportional VOD views — passes comfortably.
Does the streamer know they are being audited?: No. The sampler joins the public chatroom the same way any viewer’s browser does. Nothing is installed and no cooperation is required, so the behaviour it measures is the behaviour that was already there.
What if I disagree with a verdict?: Read the reasons and the measurements on the certificate — they are all there — and re-run the check while the channel is live. A verdict is a point-in-time, evidence-weighted opinion; a channel that was quiet during one broadcast can read differently in another, and the evidence trail lets you judge the call rather than take it on faith.

Also — the rest of the desk

Audience verificationThe ten-minute in-chat probe and the stamped verdict.Viewbot detectionHow fake viewers are caught, signal by signal.Market intelligenceThe always-on ledger of the whole category.Deal pricingWhat a real audience is worth, and what you paid.For affiliate managersVerify a streamer before the wire goes out.RatesFree, Scout, Pro and Agency — crypto billing.The journalField notes on Kick casino audiences and fraud.

Read the evidence yourself

Point it at any live channel and watch it work.

The method is only convincing once you have seen it run. A free check takes ten minutes.

Run a free check