Wrapped for TikTok · Research

State of the Feed

"Wrapped for TikTok" gives people a personal recap of their time on TikTok. At the end, users can choose to donate their scrolling data to science. 16,981 people shared their data, and this report is what we learned from it.

Each of them donated their complete, anonymized TikTok data export to our research program: 568 million logged events from 83 countries, some reaching back to 2015.

464,272,908
videos watched between them — 2.9 million hours of feed time, about 335 years
16,981 consent-based donationsfully anonymizedaggregates only, no individualsmethodology at the end

01 · 2024 → 2026

Two years, one direction

Two waves of donations, eighteen months apart, measured with the same instruments. Before we go deeper, the headline: everything about how we watch moved the same way — faster, fresher, more often, later into the night, and on more days of the year.

The swipe · 2024 → 2026

40% less patience per video

10s10s
0s
15s

Videos dismissed within three seconds: 23% → 33%.

The culture cycle · 2024 → 2026

Content is 41% younger

9.4d9.4d
upload day14 days old

Watching that is less than a day old: 17% → 24%. A shared reference now expires in days, not weeks.

The sitting · 2024 → 2026

21% more sittings — and longer ones

3.9×3.9×
sittings per day6

The median sitting grew too: 19 min → 26 min. Consumption is rising from both ends at once.

The night · 2024 → 2026

The day's last scroll ends 60 minutes later

10:00 pm10:00 pm
midnight
8:00 pm
2:00 am

Days whose scrolling runs past midnight: 30% → 36%.

The habit · 2024 → 2026

The feed claims 24 more days a year

305 days305 days
one dot = one daya full year

Days with any scrolling at all: 84% → 90% of the year. A day without the feed is now the exception — about 7 in 10 panelists have held an unbroken 30-day streak.

All readings · 2024 → 2026

Every needle, one direction

Late-2024 panel2026 panel
Median time before a video is swiped away
10s6s−40%
Videos dismissed within 3 seconds
23%33%+10 pp
Median age of a video when watched
9.4d5.5d−41%
Watching that is less than a day old
17%24%+7 pp
Hit video's views arriving in its first 14 days
64%68%+4 pp
Sessions per active day
3.9×4.7×+21%
Median session length
19 min26 min+33%
Median end of the day's last scroll
10:00 pm11:00 pm+60 min
Days whose scrolling ends past midnight
30%36%+7 pp
Days a year that include the feed at all
305d329d+24 days

Bars are drawn to each metric's full scale. Two different panels, measured identically from their watch history — panel-to-panel, not within-person.

02 · Attention

The 9-second audition

Every video gets a heartbeat to prove itself before a thumb passes judgement. Across 464 million views, the verdict is remarkably consistent — and remarkably fast. Scroll on to watch a minute of feed time play out.

A video starts playing…
0.0s
0%of viewers swiped away

Keep scrolling to watch a minute tick by — and the panel's verdicts come in.

0%25%50%75%100%0s15s30s45s60s
How long a video holds us
under 1s
6%
1–2s
13%
2–3s
8%
3–5s
10%
5–8s
10%
8–15s
16%
15–30s
15%
30–60s
11%
over 60s
12%
Share of all watched videos by inferred watch time. Highlighted: the 26% dismissed within three seconds.
The swipe is getting faster
0s5s10s15slow panel coveragelate-2024 donations2026 donations
Median seconds before a video is swiped away, per month. The line is split where the panel composition flips from the late-2024 donation wave to the 2026 wave — the two segments are different people, so they are deliberately not joined. The greyed stretch between the waves has too few contributing panelists to be reliable.
Late-2024 panel
10s
median swipe · 23% of videos gone within 3s
2026 panel
6s
median swipe · 33% of videos gone within 3s

The two waves are different people, so this is a panel-to-panel comparison, not a within-person change — see methodology.

03 · Culture

The feed has no memory

The feed lives in an eternal present. Half of everything watched was less than 9 days old — and the past barely exists.

8.1 days
the median video's age at the moment it crossed someone's screen
19%
of everything watched was uploaded within the previous 24 hours
5%
was older than a year — the archive barely exists
How old the feed is
under 1 hour
3%
1–24 hours
16%
1–7 days
29%
1–4 weeks
21%
1–12 months
26%
over a year
5%
Share of all watched videos by their age at watch time. Upload times are decoded from TikTok's own video IDs — the high bits of every ID are the upload's unix timestamp.
Late-2024 panel
9.4 days
median content age · 17% fresh within 24h
2026 panel
5.5 days
median content age · 24% fresh within 24h

Different people in each wave — a panel-to-panel comparison, not a within-person change. Both are measured identically from watch history.

Trends burn faster
Hits of 2024
64%
of a hit's views arrive in its first 14 days · 1.6M videos
Hits of 2026
68%
of a hit's views arrive in its first 14 days · 349K videos
For widely-seen videos (30+ panel watchers), the share of their lifetime panel views collected within two weeks of first appearing. The higher it climbs, the shorter a trend's whole life — a shared reference now expires in days.

04 · Rhythm

The world scrolls on one clock

Aligned to each person's own local time, the panel's 83 countries collapse into a single daily heartbeat: quiet at 4 am, building all day, peaking after dinner — and never actually stopping.

Peak hour: 8 pm12 am6 am12 pm6 pmpeak8 pm
11%
of all watching happens between midnight and 6 am — the feed never really sleeps
27%
of people have their single busiest hour between 10 pm and 4 am

Watching climbs all day and peaks at 8 pm. The quietest moment of the feed's day is 4 am.

A week on the feed
12am
6am
12pm
6pm
Mon
Tue
Wed
Thu
Fri
Sat
Sun
Share of all watching by local weekday and hour (brighter = more). Weekend days carry 9% more watching than the average weekday. Times are each user's own inferred local time.
The day ends later now
Late-2024 panel
10:00 pm
median last scroll · 30% of days end past midnight
2026 panel
11:00 pm
median last scroll · 36% of days end past midnight

Deep-night watching (midnight–3 am): 9% of the late-2024 panel's watching vs 8% for the 2026 panel.

When each panelist's last session of the day ends (median, local time), and how often a day's scrolling runs past midnight. Days are counted 4 am to 4 am, so a 1 am session belongs to the evening before.

05 · Sessions

“One more video”

The feed is consumed in sittings. Most are a coffee-break long — but the outliers do the heavy lifting: the 20% of sessions that run past an hour swallow 60% of all time spent watching.

30
videos in the median session
20 min
length of the median session
3.9
sessions on a typical active day
3.8 h
the 99th-percentile session
How long a sitting lasts
under 5 min
24%
5–15 min
19%
15–30 min
18%
30–60 min
20%
1–2 h
14%
over 2 h
6%
Share of the panel's 7,191,842 sessions by length. A session ends after 30 minutes without a new video.
Where the time actually goes
Sessions over an hour make up 20% of sessions
Sessions over an hour make up 60% of all feed time
Sessions of an hour or more are a minority of sittings — but they absorb most of all time spent on the feed.
More often — and longer
Late-2024 panel
3.9×
sessions per active day · median 19 min · 15% under 2 min
2026 panel
4.7×
sessions per active day · median 26 min · 14% under 2 min
Session frequency and shape by donation wave. The 2026 panel doesn't just open the feed more times a day; the median sitting is also longer. Consumption is growing from both ends at once.
A daily habit, measured
Late-2024 panel
84%
of days on the feed · 74% kept a 30-day streak
2026 panel
90%
of days on the feed · 70% kept a 30-day streak
Of the days each panelist's history covers, how many included the feed at all — and how common a 30-day unbroken daily streak is. Only panelists with at least two months of watch history are counted.

06 · Engagement

Mostly, we just watch

The like button tells a quieter story than the culture around it suggests. For the overwhelming majority of videos, the only trace a viewer leaves is the swipe itself — and the reactions that do exist pool around astonishingly few people.

Out of every 1,000 videos watched…
liked88 in 1,000 · 8.8%
shared14 in 1,000 · 1.4%
favorited7 in 1,000 · 0.7%
watched live5 in 1,000 · 0.5%
commented on3 in 1,000 · 0.3%
led to a follow3 in 1,000 · 0.3%
Each grid is exactly 1,000 watched videos; lit cells are the ones that got the reaction. Everything else scrolled past in silence.
91%
of videos are watched without so much as a like
17 chars
median length of a comment — about three or four words
34%
of comments are 10 characters or fewer — emoji and one-word reactions
Where the likes land
75%
of all likes received go to the top 1% of panelists
26
likes received by the median panelist — in their entire time on TikTok
34%
have never received a single like
The 159 million likes these panelists have received in their lifetimes are distributed the way attention economies always are: almost all of it flows to almost none of them.

07 · Sharing

Sharing is whispering

When something is good enough to pass on, it almost never goes on a story or a wall. It goes to one specific person, in a private chat — 90% of the time without ever leaving TikTok.

90%
of all shares go straight into TikTok's own DMs
7.5M
shares observed across the panel
0.8%
end up on a public story — WhatsApp status, Instagram or Snapchat stories combined
Where a shared video goes
TikTok DMs
90.1%
Copied link
4.0%
WhatsApp
2.5%
Other
1.3%
Messenger & SMS
0.8%
Instagram
0.4%
Telegram
0.4%
Snapchat
0.3%
Discord
0.1%
Share of all shared videos by destination. Even counting every external app together, the private message wins by an order of magnitude — on TikTok, sharing is a one-to-one gesture, not a broadcast.

08 · The algorithm

A feed of one

We talk about TikTok as shared culture, but the data shows the opposite: personalization is so deep that most of what you saw this year, statistically, no one else in a 16,981-person panel saw at all.

seen by one person onlyseen by two or more
71%
of all videos in the dataset were seen by exactly one panelist — nobody else's feed ever showed them

Of the 132M distinct videos the panel encountered, the overwhelming majority appeared in a single person's feed and no one else's. The dimmed cells on the left are that private majority; the lit cells are the shared culture.

Few videos, most of the views
share of all videosshare of all views
seen by 1 person
70.6%
22.9%
seen by 2
12.0%
7.8%
seen by 3–5
9.5%
11.3%
seen by 6–20
6.0%
19.1%
seen by 21–100
1.7%
21.4%
seen by 100+
0.3%
17.5%
Videos grouped by how many panelists saw them. The long tail dominates the catalog (gold) while the tiny shared head absorbs an outsized slice of actual watching (blue).
29%
of the panel saw the single most widely-seen video — even the biggest hit missed most feeds
11%
of all watches are rewatches of a video that person had seen before
2.5M
times someone returned to the same video at least five times
But the shared sliver is global
1 country
1%
2–4 countries
24%
5–9 countries
17%
10+ countries
58%

The median hit spread across 12 countries, with its single biggest country holding only 41% of its audience.

The few videos that do break out of a single feed break out of countries too: of 1.7M hit videos (30+ panel watchers), 99% were watched in more than one country. Personalization runs deep — but the shared layer on top is one worldwide pop culture.

09 · Methodology

How to read this

Everything above is computed from voluntarily donated, anonymized TikTok data exports. Here is exactly where the numbers come from — and where their limits are.

Watch-history coverage by month
20242026
Panelists contributing watch history per month. The two donation waves are clearly visible; statistics lean on the densest regions.

Who is in the panel

16,981 people who used Wrapped for TikTok and explicitly donated their anonymized TikTok data export for research. The panel skews toward our largest audiences (BR, US, SG) and toward people curious about their own screen time — it is not a representative sample of TikTok.

Watch time is inferred

TikTok exports record when each video started, not how long it played. We measure the gap to the next video start and treat gaps up to five minutes as watch time; longer gaps end a session. True watch time can differ when a video is left running.

Two donation waves

Most data arrived in two waves: December 2024 and March–July 2026. Any comparison across the waves compares different people, and every export only reaches as far back as TikTok's own retention for each data category. Where that matters, the report says so on the chart.

Local time is inferred

Exports carry UTC timestamps. Each panelist's timezone is estimated from all data in their dataset.

Only aggregates are published

Every number on this page is an aggregate over thousands of people. No individual viewing histories, videos, creators or sounds are identified, and monthly statistics require at least 50 contributing panelists.

The numbers

568,385,037 events, of which 464,272,908 are video views, spanning 2015-06-25 to 2026-07-16. Dataset release e317db0d, report generated 2026-07-30.

Are you a researcher?

Dig deeper than this page can

We are already working with academic researchers investigating this dataset well beyond what a report can show. If you would like to use it in your own scientific work, we'd love to hear from you — anytime.

[email protected]

The dataset is available for scientific research only — commercial applications are not permitted. Access is granted exclusively under a data usage agreement that restricts misuse, and everything in it is anonymized and donated with explicit consent.

Wrapped for TikTok is free and funded by sponsors. Want your brand in front of millions of TikTok fans?

Sponsor the project →

Curious what your feed says about you?

These are everyone's numbers. Get your own — free, in your browser, from your TikTok data export.

A screenside project, by vantezzen
Silly little projects for when you need a break from reality
SimpleAnalytics Logo

TikTok is a registered trademark of Bytedance Ltd. This website is not affiliated with or endorsed by TikTok or Bytedance Ltd.

Spotify is a registered trademark of Spotify AB. TrueShuffle is not associated with Spotify.