Twitch shows you how many people are watching right now and nothing about last Tuesday. We sample Twitch's public live listing every 15 minutes and keep each reading, so for any channel we have caught broadcasting we can lay out a weekday-by-hour grid of real concurrent viewers - the strongest slots, the weakest ones, and the hours nobody on the channel has ever tested.
Twitch publishes no viewer history. The channel page shows a live number while a stream is running and nothing at all when it is not, and there is no endpoint anywhere that answers "how many people were watching this channel at 8pm last Friday". That is the gap this tool fills: our directory engine reads Twitch's public live listing every 15 minutes and appends each reading - channel, timestamp, concurrent viewers, category - to a time series of our own. As of 2 September 2026 that series covers 41,506 Twitch channels out of the 66,493 in our catalog.
A weekday-by-hour grid is what that series is for. Four readings an hour means an hour of broadcasting leaves four rows behind, so a cell in the grid is not an estimate of what a slot was like - it is the average of the readings actually taken inside it. The tool reports the airtime behind every cell alongside the viewer figure, so a slot backed by half an hour of evidence is never presented as though it were backed by six.
The second half of the answer is the platform, not the channel. For every hour of the clock we also hold the total viewers and the total live channels our sweep saw, and dividing one by the other gives viewers per live channel: the number that says whether an hour is worth entering. An hour where the audience outnumbers the channels serving it is an hour an unknown stream can be found in. That ratio is what ranks the hours a channel has never used, and it is why the recommendation is not simply "stream when everybody streams".
The window is seven days, and the page says so wherever it prints a number. Our viewer readings live in a table partitioned by day whose finished partitions are dropped rather than kept, which is what stops a million rows a day from becoming a maintenance problem. Seven days of readings is a week of evidence about a schedule. It is not a month, it is not a habit, and it is not presented as either.
A channel we have never caught live has no grid at all, and the tool says that rather than drawing an empty one. Sampling only sees a channel while it is broadcasting, so a channel that streams rarely, or streams between two sweeps, simply is not in the series. On 2 September 2026 that included kaicenat, ninja, xqc - large Twitch channels with no viewership on record. Zero is not the answer there, and neither is a schedule inferred from follower count.
Viewer readings are taken every 15 minutes and kept for seven days. Catalog figures on this page were measured on 2 September 2026 across 66,493 active Twitch channels.
A channel enters the viewer series only when it is broadcasting at the moment a sweep runs, so the measured set is always smaller than the roster. Both numbers are printed here because the gap between them is the honest answer to “why is my channel missing”.
| Measure | Twitch |
|---|---|
| Active channels in our catalog | 66,493 |
| Channels with a measured viewership | 41,506 |
| Categories seen | 2,018 |
| Languages seen | 37 |
| Readings per channel per hour | 4 |
| Days of readings retained | 7 |
| Median average concurrent viewers | 5 |
| 90th percentile average concurrent viewers | 104 |
Measured on 2 September 2026. On that date kaicenat, ninja, xqc all had no viewership on record despite large followings, which is the normal state for a channel that did not go live inside a sweep window.
Each cell is one weekday and one hour in the time zone you pick, holding the average and the peak of every reading taken in it over the last seven days, plus the airtime those readings represent. Cells with a single reading are shown but never ranked: fifteen minutes is not a slot.
The timing score compares where the channel actually streams against the platform's viewers-per-live-channel curve, and reports the result as a share of the best hour available. A channel that streams in the emptiest hours of the day scores near 100; one that streams only when the platform is at its most crowded scores low even if its raw viewer numbers are healthy. When the platform curve is unavailable the score falls back to comparing the channel against its own best hour, and the page says which basis was used.
Untried hours are ranked on the platform curve alone, because the channel by definition has no data for them. They are an opportunity ranking, not a prediction: nothing here knows whether your audience can move, only where the room is.
It cannot tell you what would have happened. The grid is a record of streams that ran, so a slot with strong numbers may be strong because the channel is good at that hour or because it happens to raid well then, and only running the experiment separates the two. It also cannot see a channel it never caught live, which on a 15-minute cadence means very short streams can be under-represented. And it is seven days: one holiday week, one collaboration or one week off is enough to move every number on the page, which is exactly why the airtime behind each cell is printed next to it.
It is ours. Our directory engine reads Twitch's public live listing every 15 minutes and stores each reading against the channel it belongs to. That is why this page can show an hour-by-hour history that Twitch itself does not offer, and it is also why the window is seven days rather than a year.
Because we have never caught it broadcasting. A channel only enters the series if it is live at the moment a sweep runs, so a channel that streams rarely or briefly can be missing entirely even with a large following. Go live once and the next sweep picks it up. We will never fill the gap with a zero or with a schedule guessed from your follower count.
Total viewers divided by live channels, at one hour of the clock, across the whole platform. It is the supply-and-demand ratio of a time slot. A high ratio means the audience outnumbers the people serving it, which is when an unfamiliar channel is findable; a low one means everybody is live at once and the same audience is split further.
It changes which hour a reading is filed under, so yes, it changes the grid, and it should. Readings are stored in UTC and converted when the page is built, so daylight saving is handled by the zone rather than by a fixed offset.
No. Your own dashboard sees every second you were live; we see one reading every fifteen minutes, from outside, and only while you are on air. Our figures are close on a long stream and will differ on a short one. The virtue of them is that they are computed the same way for every channel, so they compare.
Everything read here is the public live listing that anyone can load without an account, and PlayerSells is not affiliated with Twitch. We take no credential, use no logged-in session and apply no more load than a browser would.
the biggest audience pools, the least contested rooms and the fastest growing channels
the hours your channel actually draws viewers, and the hours you have never tried
What a channel earns from its concurrent viewers, at Twitch's own subscription split.
Where a follower count lands among the 66,493 Twitch channels we track. The median has 266.
The full Twitch directory: 66,493 channels ranked by measured viewership across 2,018 categories.
The same directory for Kick, where 44,847 channels carry a measured viewership figure.
PlayerSells is not affiliated with Twitch. Every viewer figure on this page is our own measurement, sampled from Twitch’s public live listing every 15 minutes. Twitch publishes no viewer history, so nothing here is a Twitch statistic and nothing here should be read as one.
