What percentage of your followers watch your stream?

About 0.6%.

That is the median across 41 of the hundred largest Grand Theft Auto channels in the world, measured over the thirty days to 04:04 UTC on 19 October 2026 — the 41 where the platform publishes a follower count and we hold an audience figure. Half of them sat between 0.25% and 1.63%. The best converted 4.31% of its followers into a live audience. The worst converted 0.055%.

Every streamer has seen some version of this number in their own dashboard and wondered whether it is normal. As far as I can find, nobody publishes a measured answer. Search the question and page one is platform-statistics round-ups — Twitch’s total concurrent viewers, its channel count, revenue estimates — none of which divides one by the other. That is not an oversight. The ratio needs a follower count and an audience figure for the same channel, and the sites that hold a lot of one usually do not publish the other, which leaves it as something you can only normally see for yourself, in your own analytics, with nothing to compare it against.

We hold both, for every channel we track. So here is the ratio.

It collapses as the channel grows

The single clearest thing in the data is that the share falls, hard, with size.

followers	channels	median share watching	median followers	median live audience
Under 1M	11	2.17%	430,901	6,693
1M – 5M	21	0.59%	1,677,543	10,073
Over 5M	9	0.19%	12,641,622	27,488

The bands barely overlap. Every channel above five million followers landed between 0.06% and 0.27%; every channel under a million landed between 0.96% and 4.31%. Across all 41, the correlation between follower count and the share watching is −0.886 on logged values, which is about as clean a relationship as audience data ever gives you.

Put it in channels instead of percentages and it is stark. Chatterbox holds 135,388 Twitch followers and averaged 5,833 concurrent viewers on GTA. Typical Gamer holds 16.1 million on YouTube and averaged 8,883. That is 119 times the followers for 1.5 times the audience.

It runs all the way up the board. The seventh-largest GTA channel by average audience in this window, Tumblurr, has 2.0 million followers and averaged 37,243 — more than Windah Basudara’s 29,837 off 20.6 million. Follower count and live audience are correlated, but weakly: on rank order across these 41, the figure is 0.489. Followers tell you less than half of what you want to know.

The obvious objection, tested

The first thing anyone says to this is that it is not a size effect at all, it is a specialisation effect: a channel with 400,000 followers earned them from GTA, so of course they turn up for GTA, whereas a 16-million-follower generalist earned them across a decade of other games and the GTA audience is a sliver of the whole.

That is a good objection and the data does not support it. If it were right, channels that stream a lot of GTA would convert better than channels that dip in. They do not. Split the 41 into terciles by how many hours of GTA they streamed in the window:

GTA airtime tercile	channels	median airtime	median share watching
Low	13	2.3 h	0.51%
Middle	14	10.9 h	0.60%
High	14	91.8 h	0.69%

Flat. The correlation between airtime and the share watching is +0.099 — nothing, against −0.886 for follower count. And the two extreme bands are not separated by airtime either: channels under a million followers streamed a median 13.1 hours of GTA, channels over five million a median 8.0, with both ranges running from under two hours to over 260.

So within this sample, the thing that predicts what share of your followers show up is how many followers you have, not how committed you are to the game. xQc streamed 266.9 hours of GTA in thirty days — far more than anyone else here — and converted 0.23% of his 12.6 million. Chatterbox streamed 256.1 hours and converted 4.31% of 135,388.

I would not call that cause. The honest reading is that a follower is a once-ever click that never expires, so a ten-year-old channel’s count is a historical total with a decade of lapsed interest inside it, while a live audience is a single moment that only currently-interested people can be in. The bigger the number, the more of it is archaeology. That mechanism fits the gradient, and this data cannot prove it over the alternatives.

What a streamer should take from it

If you are small and your dashboard says 2% of your followers are watching, you are not underperforming — you are doing roughly twice as well as the median channel on this board, and several times better than the household names on it. The ratio is not a quality score. It is mostly a measure of how long you have been accumulating followers relative to how many people care today.

The practical version: a follower target is a weak proxy for an audience target, and it gets weaker the bigger you get. Going from 135,000 followers to 16 million bought Typical Gamer 1.5 times the live audience. Whatever the next thousand followers are worth to you, do not price them off the first thousand.

What this does not show

Kick is missing, and Kick is the larger half of this category. 59 of the hundred channels here are Kick channels and not one of them publishes a follower count to us, so all 59 are excluded — including sxb, the single largest GTA channel in the world by average audience in this window at 131,328 concurrent. At the census behind this piece, Kick held 84,897 of 162,053 live GTA viewers against Twitch’s 67,256, so the platform carrying the majority of the audience contributes nothing to the ratio above. Read this as a Twitch-and-YouTube finding.

n is 41. The bands are 11, 21 and 9 channels. The gradient is strong and monotonic, but nine channels is nine channels.

The two numbers are not measured over the same thing. The follower count is the whole channel, every game, all time. The audience figure is airtime-weighted mean concurrent viewers across the GTA streams we sampled in the window. We only see GTA, so we cannot compute what share of anyone’s streaming it is — which is exactly why the tercile test above rules out one version of the specialisation objection and not all of them.

It says nothing reliable about small channels. I tried to extend this down the distribution: a systematic sample of 297 Twitch channels taken every fifth place by rank through today’s live board and its tail. Only 38 of the 297 carried a follower count, and the missingness is not random — the 38 had a median 279.8 hours of 30-day airtime against 41.2 for the 259 without, and a median audience of 275 against 3. Follower coverage tracks how closely we track a channel, so any percentile I computed from it would describe our own sampling and not streaming. The finding stops at the top hundred on purpose.

One window, one game. Thirty days to 19 October 2026, Grand Theft Auto only, on Kick, Twitch and YouTube. A channel that spent those thirty days on other games appears here with whatever GTA it did stream, and that is a thin slice of its year.

Cite this

The figures above come from our own sampling and are free to quote. The population is the hundred channels on the 30-day leaderboard as it stood at 2026-10-19 04:04 UTC, of which the 41 carrying a published follower count are analysed; the live category figures are a census at 04:09:45 UTC the same morning, 162,053 viewers across 1,803 channels, reconciled against a second census nine minutes later. Permalink: https://luciastream.com/blog/what-percentage-of-followers-watch-your-stream/. How the collector samples, and what it does and does not see, is written up at /methodology/.