Why this category is exposed
Three things line up badly in iGaming streaming. Sponsorship fees are large relative to other verticals. They are usually priced against concurrent viewers. And concurrent viewers is a number the platform reports, the streamer influences, and the buyer cannot independently verify.
That does not mean the category is mostly fake — most of the channels in our top 20 have viewership patterns that look entirely normal. It means the incentive to inflate rises with deal size, so the check should be routine rather than reserved for suspicious cases.
Be equally sceptical of the other direction. Vendors selling detection have an incentive to publish alarming category-wide percentages. Any "X% of viewers are bots" claim without a published method is marketing, not measurement.
The six checks
1. Concurrency variance across a broadcast
The strongest single signal. A real audience drifts continuously — people arrive, leave, respond to what is on screen. Sample viewer count at five-minute intervals across a full broadcast and look at the shape.
- Normal: continuous drift, a visible ramp at the start, decay at the end, spikes tied to on-stream events.
- Flag: a flat plateau for hours, or step-changes that snap to a level and hold it. Human audiences do not form plateaus.
2. Chat-to-viewer ratio against comparable channels
Bots watch; they rarely chat convincingly. Compare unique chatters per 1,000 viewers against three to five channels on the same platform, in the same language, in the same concurrency band.
The comparison set is the whole method. Absolute thresholds are useless here: large channels have structurally lower participation than small ones, casino content is more passive than Just Chatting, and some language communities chat far less than others. A ratio far below its comparables is a flag; a low ratio on its own is not.
3. Follower growth versus stream days
Follower gain should track when a channel was actually live. Plot daily follower change against stream days.
- Normal: gains concentrated on stream days, larger gains after high-concurrency sessions.
- Flag: steady gains on dark days, or a step of thousands with no broadcast and no external event behind it.
4. Peak-to-average ratio
Not primarily a fraud check — a delivery-risk check that occasionally catches fraud. Divide peak concurrent viewers by airtime-weighted average concurrency.
- Under ~2×: a very flat channel. Predictable delivery. Worth checking against variance, because sustained flatness is also what inflation looks like.
- 2–5×: the normal range for this vertical.
- Over ~8×: event-driven. The headline peak describes one moment; average delivery is a fraction of it. Price against the average.
5. Airtime pattern
Look at when the channel streams, not just how much. Human schedules cluster around particular hours and days. A channel showing uniform 24-hour coverage with stable concurrency across every time zone is either a multi-host crew operation — legitimate and easy to verify from the video — or something else.
6. Cross-platform and cross-metric consistency
If a channel is active on more than one platform, its relative audience size should be broadly consistent across them. Sharply different concurrency on one platform, with similar content and schedule, deserves an explanation. Likewise check that hours watched, follower base and chat activity all tell the same story about channel size. Inflation usually shows up in one metric at a time.
What a clean channel looks like
- Concurrency drifts continuously; no plateaus.
- Chat-to-viewer ratio within the range of comparable channels.
- Follower gains concentrated on stream days.
- Peak-to-average between roughly 2× and 5×.
- A schedule that looks like a person's or a documented crew's.
- Hours watched, followers and chat activity all consistent about channel size.
Most of the channels in our dataset pass all six. That is the point of running the checks: they let you pay confidently, not just decline nervously.
What these checks cannot prove
Be honest about the ceiling on external measurement:
- They cannot establish intent. Anomalous numbers can come from third-party viewbotting aimed at a competitor, embeds on external sites, or platform counting quirks.
- They cannot see audience geography. Neither platform exposes reliable viewer-location data publicly. Broadcast language is a proxy, and only a proxy.
- They cannot measure conversion quality. A completely real audience can still be worthless for your product. Attribution is a separate exercise.
- They depend on the platform's own reporting. Every external measurement, ours included, samples numbers the platform publishes. We test whether they behave plausibly; we cannot audit the source.
Our methodology page states exactly which metrics we derive and which we do not claim to verify.
Protecting yourself in the contract
The cheapest protection is structural, not forensic. Four clauses do most of the work:
- Minimum delivery in watch-hours, with a make-good rather than a refund. This makes inflated concurrency the streamer's problem.
- A named third-party measurement source, agreed before the campaign.
- A pre-campaign baseline of the channel's normal performance. Without it there is nothing to compare the result against.
- Performance weighting — shift part of the fee to CPA or revenue share so fake viewers cost the seller, not the buyer.
Pricing mechanics are covered in casino streamer sponsorships. The category context is in iGaming in live streaming.