Do 24/7 live streams work? 4.9 years live, 13 viewers

Not in the biggest game category we can measure. At 04:13–04:28 UTC this morning there were eleven Grand Theft Auto channels that had been continuously live for a week or more across Kick, Twitch and YouTube. Between them they held 93 concurrent viewers — 0.030% of the 307,213 people watching GTA at the time. The channel at the top of that list has been live, without a break, for 42,560 hours. That is 4.86 years. It has thirteen viewers.

Stack all eleven into one imaginary channel and it would rank about 283rd out of 2,003 live channels. The single biggest GTA stream running at the same moment had 420 times their combined audience.

The pitch

Round-the-clock streaming is an industry. Search for it and page one is almost entirely companies selling the service — Upstream, OneStream, YoloLiv, LiveReacting, StreamHouse — and the pitch is consistent: a channel that is always live accumulates watch time, gets surfaced in the Live tab, catches every time zone, and earns ad revenue while you sleep. You upload a few hours of video, a cloud service loops it as a live broadcast, and the algorithm is supposed to do the rest.

It is a plausible story and none of those pages puts a number on it. So here is one, from the only angle we can actually see: how many people are watching.

What is actually live

Our collector samples every live GTA stream on the three platforms once a minute and records the start time each platform reports for it. Grouping this morning’s fourteen censuses by how long each stream had been running gives a very lopsided picture.

Continuously live for	Channels	Share of channels	Viewers	Share of viewers	Median viewers
Under 24 hours	1,960.1	97.85%	302,669	98.521%	6.0
24–48 hours	30.4	1.52%	4,337	1.412%	5.1
48 hours – 1 week	2.0	0.10%	114	0.037%	57.2
A week or more	10.7	0.53%	93	0.030%	2.0

The middle bands are not loops. Read the titles in the 24-to-48-hour group and they are subathons and charity marathons run by people who are genuinely awake — SUBATHON DAY 11, MARATON DIA 848, 24H DE LIVE, a Brazilian DIA DE SUBATHON. Twenty-seven of the forty-three channels past a day were on Twitch, where the subathon format lives. Those streams do fine: the band holds 1.41% of viewers on 1.52% of channels, which is roughly its fair share. Ignore the 48-hour-to-a-week row — it is two channels, and a median drawn from two numbers is not a measurement.

The bottom row is the one that behaves differently. A week is long past the point where a human could still be at the keyboard, and the audience collapses: 0.53% of channels holding 0.030% of viewers, a median of two people each.

The eleven

Channel	Platform	Continuously live	Concurrent viewers
Weiss Network TV	YouTube	42,560 h (4.86 yr)	13
SunDry	YouTube	24,851 h (2.83 yr)	1
SourceSpy91	YouTube	15,970 h (1.82 yr)	21
RTX IS LIVE	YouTube	5,244 h (0.60 yr)	1
Airz	YouTube	3,410 h (0.39 yr)	2
Aodiak	YouTube	2,209 h (0.25 yr)	2
xenos888	Kick	1,220 h (0.14 yr)	1
izzytalez	Kick	1,218 h (0.14 yr)	2
Wondersgta Gaming	YouTube	881 h (0.10 yr)	10
Classic Hugo One GTA Speedruns	YouTube	640 h	41
logo blocks	YouTube	212 h	2
All eleven			96

Nine of the eleven are on YouTube, and the tooling explains most of that: nearly every guide and service on that first page of results is written specifically for YouTube, down to the titles. Within YouTube’s own slice of GTA the format is not marginal at all: 8.9% of YouTube’s live GTA channels had been live for over a week, against 0.53% of the category overall. They held 3.5% of YouTube’s GTA viewers.

Read the titles and these are not really one thing. Three are countdown timers to GTA VI rather than gameplay at all — a clock ticking down to 19 November, running for months. Several are archive loops of the older games: GTA III, Vice City, San Andreas, GTA IV, played on repeat. One is a speedrun compilation, and at 41 viewers it is the most watched of the eleven and sits well above the population median of six.

That last point deserves more weight than a contrarian framing would give it. The spread inside the eleven runs from one viewer to forty-one, so “it never works” is not what the data says. What it says is narrower and more useful: being permanently live is not, by itself, worth anything. The channel with 4.86 years of unbroken uptime has thirteen viewers, and the one with 640 hours has 41. If continuous presence compounded the way the sales pages imply, that ordering would be the other way around.

There is also an honest case for the format that our data cannot test. We measure concurrent viewers; we cannot see ad impressions, watch time, memberships or what any of this costs to run. A loop that holds three viewers around the clock still accumulates roughly 72 viewer-hours a day for the price of a cheap server, and for a countdown clock or an archive channel three simultaneous viewers may be exactly the product. None of that is growth, which is what the pitch is selling, but it is not nothing either.

What this does to our own numbers

Eleven channels out of two thousand should not be able to move a population statistic. These do, badly, and it is worth showing where.

The mean time a live GTA stream has been running is 53.8 hours. Drop the eleven and it is 4.8 hours — an eleven-fold difference from 0.53% of the sample. The median, 3.6 hours, barely notices them. Anyone who quotes an average session length off a feed like this one, ours included, is quoting these eleven channels.

It is worse on our own milestones page, and we would rather say so than wait to be asked. That page currently reports a record of 2,314 hours of airtime in 30 days. Thirty days contain 720 hours. The number is impossible, and the reason is that our airtime figure sums each broadcast separately while the record holder runs several 24/7 loops at the same time — its own profile shows concurrent sessions of 372, 462 and 243 hours. Three simultaneous streams, 3.3 times the wall clock. The same page’s “longest session” record is capped at 48 hours by a deliberate sanity check in the pipeline; airtime has no equivalent guard, and it should. That is a bug in how we compute a record, it has been public, and it is now on the list to fix properly rather than clamp.

What this does not show

The measurement is fifteen minutes long, taken at 04:13–04:28 UTC, which is near the bottom of the daily curve. It counts concurrent viewers, not people, revenue or watch time — the three things the 24/7 pitch is actually about.

Start times are whatever each platform reports, taken at face value. A channel that dropped and reconnected without the platform resetting its start would look continuously live to us when it was not, and we have no independent way to check a four-year uptime claim beyond YouTube’s own metadata.

Our content classifier reads stream titles, and it is wrong about at least one of the eleven: the second-longest-running channel is broadcasting a Blox Fruits private server, not GTA, and a title keyword swept it into our category. We have left it in the table rather than quietly dropping it, because it is evidence about the method. Exclude it and the count is ten channels and 92 viewers, which changes nothing about the argument and is the honest version of the number. One further wrinkle in the same direction: the channel running concurrent loops appears in only ten of the fourteen censuses, because our exporter picks one of its simultaneous sessions and does not always pick the same one.

Finally, nothing here is causal. These are low-effort rebroadcasts, and it would be circular to conclude that looping caused the small audiences when the content would likely draw small audiences either way. The claim is the weaker one: uptime on its own does not buy an audience. Our earlier findings on session length and on total hours streamed point the same way, and the three together are about as close as we can get to saying that time-on-air is a much weaker lever than the advice industry treats it as.

Check us

Every figure above comes from fourteen complete censuses of every live GTA stream on Kick, Twitch and YouTube between 04:13:34 and 04:28:35 UTC on 13 September 2026, taken from data.luciastream.com. In all fourteen the top-400 board and the tail file carried matching timestamps, the per-stream counts reconciled exactly to the published totals, no channel appeared twice, and no platform reported degraded coverage. The mean across the fourteen was 307,213 viewers on 2,003 channels, with individual samples ranging from 302,059 to 315,074 and the board cut-off drifting between 41 and 44 viewers. Per-channel figures in the table are each channel’s own mean across the samples in which it appeared, which is why they sum to 96 rather than the group mean of 93.

The live board is at /live/, the full creator leaderboard at /leaderboard/, collection is described on our methodology page, and the permalink for citation is https://luciastream.com/blog/do-24-7-live-streams-work/.