TikTok LIVE: A Nightly Routine

TikTok’s data export says I watched 12,607 livestreams in a year.

That number is true and it is useless. It counts doors, not rooms. TikTok’s LIVE feed advances by itself: a stream ends, or you swipe, and the next one is already playing. Every card you land on is written down. Most of them you never saw.

Here is how long I spent in a room before the next one opened:

The median is one second. Not one minute — one second. Two thirds of my year of watching TikTok LIVE was a thumb moving.

What counts as stopping

So I set a bar. A room counts as one I stopped in if I was still there five minutes later, or if I typed something in it. Anything else is a door I opened and closed.

By that measure: 1,815 rooms of 12,607. Fourteen percent.

The other 10,792 are the feed doing what the feed does.

Two completely different evenings

The 14% hides something, though, because it isn’t spread evenly. It comes from two habits that happen to use the same app.

49 nights account for 80% of every room I ever opened. On those nights I opened fifty rooms or more — once, 938 of them — and stopped in 7%. That is not watching. That is searching, or avoiding, or the thing you do with a phone when you do not want to be where you are.

The other 192 nights look nothing like it. A typical one: ten rooms opened, four stopped in. Forty-four percent. That is a person visiting.

Averaged together they make a number that describes neither.

The year narrows

Split it by month and the two habits turn out to be phases, not moods.

March and April are almost pure feed: 8,746 rooms opened in two months, and I stayed in about one in twenty. By the end of the year I was opening a couple of hundred rooms a month and staying in half of them.

Nothing about the app changed. What changed is that I had found the rooms I was looking for, and stopped looking. The scrolling wasn’t the habit. The scrolling was the search, and it ended.

The talking outlived the rest

If the looking narrowed, did the rest of it narrow too? The export and the gift ledger cover the same months, so they can be put on one timeline — three measures of very different size, so three panels rather than one chart with two scales.

Split the year down the middle and two of the three collapse together:

The scrolling went first, in April. The money followed it down from May. The talking did not go anywhere: July is the loudest month of the entire year, two months after the spending had already fallen off a cliff, and November is louder than April.

I had assumed — and, before I checked, had written here — that the spending stayed roughly flat while the attention narrowed. It didn’t. What actually happened is that I stopped looking for rooms, then stopped paying for them, and carried on talking in the four or five I had left. Whatever the year was for, by the end of it the part that survived was the cheapest one.

Who you stop for

Of the 1,815 rooms I stopped in, 1,386 resolve to a streamer I can name, and they are 337 different people. Two thirds of them I stopped for exactly once — a single evening, never again.

Five people account for 40% of every room I ever stopped in:

That is the shape of the year in one table. A few hundred strangers got a single evening each. Five people got a year.

When

The evening starts earlier than I would have guessed and runs later than I would like to admit. The busiest hour is 9pm; 39% of every room I opened was between 9pm and 3am.

The sessions themselves are mostly tiny. Breaking the year at every gap of thirty minutes or more gives 1,277 sessions, and the median one is two rooms and about a minute long — open the app, look, close it. The long ones are rare and enormous: the longest ran five and a half hours.

The unit isn’t an evening

I had been thinking in nights, because a night is how you remember it. The data doesn’t work that way. Break the year wherever there’s a gap of half an hour or more and it comes apart into 1,277 separate sittings — and 62% of them are one or two rooms, about a minute long. Open the app, look, put it down.

The rest are the opposite kind of thing entirely. The longest ran five hours and forty minutes; the biggest carried 660 rooms. There is almost nothing in between: the median sitting is a minute and the ninetieth percentile is fifty-one.

So “most evenings” is the wrong frame, and so is “an hour a night”. It’s dozens of glances a week, with a handful of evenings that swallow everything.

When

The busiest hour is 9pm, and 39% of every room I opened was between 9pm and 3am. The shape is what you’d guess — a slow climb through the afternoon, a peak after dinner, a tail that runs well past when it should. It is the one finding here that surprised nobody, including me.

Two things I expected that aren’t there

Worth saying, because the checks were the same and only the answers differed.

I thought an evening would have an arc — open the app, scroll to see who’s on, settle in once you’ve found someone. It doesn’t. Across the 194 sittings long enough to split into thirds, the share of rooms I stopped in runs 8%, 5%, 8% from start to finish. Flat. Whatever decides that I stay somewhere isn’t where I am in the evening.

I thought the day of the week would matter, and at first it looked spectacular: Monday 8%, Friday 25%. It’s an artifact. The feed-nights aren’t evenly spread across the week, and they’re so much bigger than ordinary nights that they drag whichever day they land on. Take those 49 nights out and every day of the week sits between 40% and 49% — Monday 46%, Friday 47%, Sunday 40%. There’s no weekday pattern. There’s only which kind of night it was, and that is the same finding as everything else here.

What the ledger doesn’t know

I have written a great deal about where the money went, because money leaves receipts. This is the first time I have been able to look at where the attention went, and the two records disagree in a way I find hard to put down.

The ledger knows I sent 8 million coins to 235 people. It has no idea that 10,792 of those rooms were a thumb moving, or that 49 nights hold 80% of the scrolling, or that the ratio of looking to staying went from one in twenty to two in three.

A year of gifting reads, in the ledger, like a straight line. In the watch history it reads like someone searching a building for a room he’d already found, and gradually stopping.