Transparency

Methodology

What the numbers on this site mean, how they are collected, and where they fall short. A tracker is only as useful as its willingness to say what it cannot see.

How viewers are counted

Every tracked channel is polled on a fixed interval and its concurrent viewer count is stored as a timestamped sample. Nothing is estimated or modelled — a figure on this site is either a value a platform reported, or an arithmetic result derived from those values.

Average viewers is the mean of all samples taken while a channel was live, weighted by airtime — so a two-hour stream counts twice as much as a one-hour stream at the same size. Peak viewers is the single highest sample observed, never a smoothed maximum. Hours watched is the sum of (viewers × sample interval) across a session.

Sampling rates

Not every channel is sampled at the same rate. Polling every live channel every minute would be both rate-limit-infeasible and wasteful, since the long tail contributes a very small share of total watch time. Channels are tiered:

TierIntervalEffect
Top 1,000 by viewers60 secondsPeaks accurate to the minute
Next 4,0005 minutesShort spikes may be missed
Long tail15 minutesAverages reliable, peaks approximate

Per-platform coverage

Kick

High confidence
Source
Official public API v1
Sampling
Category poll every 60s
Caveat
Full category coverage.

Twitch

High confidence
Source
Official Helix API
Sampling
Top ~1,000 by category every 60s, plus every tracked channel by ID every 5 min
Caveat
Category pagination is capped, so the long tail is covered by a persistent creator roster rather than by paging the category.

YouTube

Medium confidence
Source
Official Data API v3
Sampling
Adaptive: largest streams every 60s, mid-tail every 5 min, tail every 15 min
Caveat
The API allots a fixed daily quota and search is by far its most expensive call, so finding new channels is rationed. Channels we already track are checked through YouTube's own public channel feeds, which cost no quota, and every viewer count is then confirmed against the official API.

Trovo

High confidence
Source
Official open platform API
Sampling
Category poll every 5 min
Caveat
Small absolute audience; included for completeness.

Known limitations

  • Collection gaps are recorded, not hidden.If a collector goes down, that window is stored as a gap and charts render a break rather than a run of zeros. A flat line at zero would misrepresent an outage as an audience collapse.
  • Language is the stream's, not the creator's.Language boards use the language tag set on the broadcast. A bilingual creator can appear under different languages on different days.
  • Country is the broadcaster's, not the audience's.Regional boards group by where the creator is based. They say nothing about where the viewers are.
  • Viewer counts are platform-reported.Platforms define and filter concurrent viewers differently, and none publish their exact methodology. Cross-platform totals are directionally useful, not exact.
  • Embeds are not re-hosted.No video is copied or cached. Players are the platforms' own embeds, and clicking through counts toward the creator on their own platform.