How the TikTok algorithm works, what changed, and how to reset it
TikTok ranks every video for every viewer by predicting how likely that person is to watch, finish, like, share or comment, using your past behavior, the video's details and, with less weight, your device and location. Follower count is not a direct factor. To reset it, go to Profile, the menu, Settings and privacy, Content preferences, and tap Refresh your For You feed.
Updated · 13 min read
The short answer: how does the TikTok algorithm work?
The For You feed is a recommendation system. It pulls a large pool of eligible videos, predicts how much you in particular will engage with each one, ranks them by that prediction, and then reshuffles the top of the list so you don't get the same thing over and over.[1][2]
TikTok's own explainer, first published in June 2020 and still the base of its public messaging, groups the inputs into three buckets: your interactions (likes, shares, follows, comments, videos you create), information about the video (captions, sounds, hashtags) and device and account settings (language, country, device type).[1] The last bucket gets less weight because you never chose it as a preference.[1]
Not every signal counts the same. TikTok gives one example: finishing a longer video is a strong indicator of interest and counts more than a weak signal like being in the same country as the creator.[1] TikTok has never published the actual weights, and it says they change.
What TikTok says the ranking uses
TikTok's Transparency Center page on the recommendation system goes further than the 2020 post. It describes a prediction score built from your own interactions and from other users' interactions, and lists signals in an appendix.[2] Here is what the official sources say, side by side.
| Signal | What TikTok has said | Source |
|---|---|---|
| Watching to the end | Finishing a longer video is a strong indicator of interest. | Newsroom 2020[1] |
| Finishes and skips | Recent videos you watched, finished, liked or skipped are listed signals; finishing and skipping are also predicted outcomes. | Transparency Center[2] |
| Time spent | The system predicts how long you will spend viewing a video. | Transparency Center[2] |
| Likes, shares, comments, follows | Core user-interaction signals. | Newsroom 2020[1] |
| Searches | Listed as an interaction behind “Why this video” explanations. | Newsroom 2022[3] |
| Video details | Caption, sound, hashtags, length, post time, region, author language. | Newsroom 2020[1], Transparency Center[2] |
| Device and account settings | Language, country, device type, operating system. Lower weight. | Newsroom 2020[1], Transparency Center[2] |
| Explicit feedback | Not interested, hidden creators or sounds, reports, interest picks at sign-up. | Newsroom 2020[1] |
| Follower count | Not a direct factor. | Newsroom 2020[1] |
Since December 2022 you can see a simplified version of this for any video. Open the share panel on a For You video and tap Why this video. TikTok groups the reasons into four types: your interactions (including searches), accounts you follow or were suggested, content recently posted in your region, and content popular in your region.[3]
There is also an old but still quoted rule about freshness. In 2020 TikTok told creators that For You recommendations generally pull from videos posted in the last 90 days, and that no single engagement metric is necessarily more important than another.[4] TikTok hasn't repeated the 90-day figure in recent material, so treat it as a rough guide, not a current spec.
Diversity: why your feed doesn't show the same thing twice
Ranking is only half the system. The other half breaks up the ranked list on purpose. TikTok says the feed generally avoids two videos in a row with the same sound or from the same creator, and filters out duplicates, videos you've already seen and spam.[1] The Transparency Center describes a similarity check that replaces top-ranked videos that are too similar to their neighbors.[2]
TikTok also inserts some videos outside your known interests. It says it does this to help you find new categories and creators, and to gauge how a video does with a wider audience.[1] In 2023 it added limits on how often it recommends certain themes in a row, such as sadness, extreme dieting or exercise, and sexually suggestive content, swapping in other topics when too many similar videos stack up.[5]
For creators this matters more than most tips. If you post ten near-identical videos with the same sound, the system is built to stop showing them back to back to the same person.
What reporting, leaks and research found
TikTok's public account is high level on purpose. Three outside sources fill in some detail. They don't fully agree with each other, which is useful in itself.
The “TikTok Algo 101” document (2021)
In December 2021 The New York Times reported on an internal document called “TikTok Algo 101”. According to Gizmodo's write-up, a TikTok spokesperson confirmed it was authentic. It described a simplified scoring formula that combines predicted likes, comments and playtime, plus whether a video is played at all, each multiplied by a weight. The document said the company optimizes for two goals: retention (whether you come back) and time spent.[12] Gizmodo noted it couldn't get a copy to check, and the real equation is more complex.[12]
The Wall Street Journal bot experiment (2021)
The Wall Street Journal created 100 fresh accounts run by bots, each with assigned interests. The bots lingered on or replayed videos that matched their interest and swiped past others. The system worked out a new account's interests in under two hours. One bot set up around sadness had a feed that was 93% about depression and mental health after 33 minutes and 224 videos.[13] TikTok said the experiment wasn't representative of real users, who have varied interests.[13]
The academic view (2024)
A peer-reviewed paper at the ACM Web Conference 2024 (Vombatkere et al.) used data donated by 347 real TikTok users, covering 4.9 million videos viewed 9.2 million times, plus bot accounts as a baseline.[14] It estimated that the feed exploits a user's known interests in 30% to 50% of recommended videos across the first 1,000. The rest is exploration.[14]
Its finding on drivers cuts against the popular “only watch time matters” line. Comparing the most and least personalized users, it found liking and following had a high impact, watch percentage a medium impact, and early skipping no significant effect.[14] The WSJ experiment, by contrast, steered feeds with lingering and replays alone.[13] Both can be true: watch behavior is enough to personalize a feed, and explicit likes and follows sharpen it further.
| Source | Method | Main finding |
|---|---|---|
| TikTok newsroom, 2020 | Company explainer | Ranks by predicted interest; completion is a strong signal; follower count isn't a direct factor.[1] |
| “TikTok Algo 101”, 2021 | Internal document reported by NYT | Optimizes for retention and time spent; scores predicted likes, comments, playtime and plays.[12] |
| WSJ, 2021 | 100 bot accounts | Lingering and replays alone narrowed a feed in under two hours.[13] |
| Vombatkere et al., 2024 | 347 users' donated data plus bots | 30–50% of videos exploit known interests; likes and follows drive personalization most.[14] |
What changed in 2025 and 2026
Search interest in “TikTok algorithm changes” jumped for a reason. The biggest change wasn't a feature. It was ownership.
The US joint venture and the retraining plan
On 23 January 2026 TikTok announced the TikTok USDS Joint Venture LLC, with Silver Lake, Oracle and MGX as managing investors at 15% each and ByteDance keeping 19.9%. Adam Presser was named CEO.[9] The announcement says the venture will retrain, test and update the content recommendation algorithm on US user data, and that the algorithm will be secured in Oracle's US cloud.[9]
Days later, on 25 January, a power outage at a US data center caused what the venture called a cascading systems failure. Creators saw zero views and missing earnings, which TikTok called a display error from server timeouts. Users also reported feeds full of generic or repeated videos.[10]
What hasn't been published is how far the retraining has gone or what it changed. As of April 2026, local reporting still described creators noticing repeated videos and drops in reach, with no statement from TikTok explaining the cause.[11] We found no official data or independent audit showing a measured change in the US For You feed. If your US reach changed in 2026, the honest answer is that nobody outside TikTok can yet say whether it was retraining, the outage, more competition, or your content.
More user control over topics, including AI content
On 3 June 2025 TikTok expanded Manage topics, which lets you turn up or down how often more than 10 broad topics (Creative Arts, Travel, Nature, Sports and others) appear in your For You feed. TikTok said it won't remove any topic entirely.[7] The same update added Smart Keyword Filters, which extend your keyword filters to similar terms, and promised to double the keyword limit to 200.[7]
On 18 November 2025 TikTok added an AI-generated content setting to Manage topics, so you can see more or less AI-made video. TikTok said it tailors the mix rather than removing content, and that it was also testing invisible watermarks on AI content made with its own tools.[8]
Rules on originality and labeling
TikTok's current Community Guidelines (released 25 August 2026, effective 24 September 2026) keep a separate tier between “allowed” and “removed”: content that stays up but is ineligible for the For You feed.[16] The integrity section says content is ineligible for the For You feed if it includes unoriginal or reused material without anything new.[15] It also requires creators to label AI-generated or significantly edited content that shows realistic scenes or people.[15]
Two more lines in that section matter if you post at scale. TikTok bans services that artificially boost engagement or trick the recommendation system, and it prohibits automation tools, scripts or other tricks designed to bypass its systems.[15]
Photo posts: more volume, mixed results
Photo carousels are a bigger share of TikTok than they were. A Metricool study of about 2.3 million posts from more than 92,000 accounts found users posted 72% more videos and about 140% more image and carousel posts between 2025 and 2026.[18] Coverage of the same study reported that average views per post fell 31% and reach fell 29% when comparing January and February of each year, and that videos got about 5.6 times the views of image posts.[19] Metricool's own summary puts it at about 5 times the views and 6 times the interactions.[18]
Another vendor study disagrees on engagement. Fanpage Karma looked at almost 700,000 posts from January to May 2025 and found TikTok carousels got about 3% more reach and 81% more engagement than videos, but about a third fewer shares.[20] Both are vendor studies with their own customer samples. The safe reading: video still wins on raw views, carousels can earn more likes per view, and TikTok has not said how it ranks the two formats differently.
Why your For You page gets stuck, and when a reset helps
Most people search for how to reset the TikTok algorithm because the feed has locked onto one thing. That's the system doing what it was built for. By the 2024 estimate, 30% to 50% of what you see is chosen because it matches interests the system is already sure about.[14] Every time you watch one of those videos to the end, you confirm the guess.
The WSJ test showed how short the loop can be: lingering and replaying alone were enough to turn a bot's feed almost entirely toward one theme within about half an hour.[13] TikTok has since added caps on how often it repeats some sensitive themes, and swaps in other topics when too many similar videos pile up.[5] Those caps cover a short list of themes, not your everyday niche. If your feed is stuck on cooking or car videos, nothing in the system will pull you out unless your behavior changes.
So pick the smallest tool that fits. One unwanted topic: use Not interested a few times and a keyword filter. A whole category you're tired of: turn it down in Manage topics. A feed that no longer feels like yours at all, or one built on a phase you've left behind: do the full refresh.
How to reset your TikTok algorithm, step by step
TikTok added an official reset in March 2023. It makes your For You feed behave as if you had just signed up, starting with popular videos and then learning from your new interactions.[5][6] It doesn't change your settings, the accounts you follow, your Following feed, your profile or your inbox.[5][6]
- Open TikTok and tap Profile in the bottom right.
- Tap the three-line menu in the top right.
- Tap Settings and privacy.
- Tap Content preferences.
- Tap Refresh your For You feed, then confirm. If you don't see it, TikTok said rollout could take weeks in some regions, so check back.[6]
A full refresh is a blunt tool. Most people want to fix one topic, not lose everything. Here are the smaller levers, from lightest to heaviest.
| Tool | What it does | Where to find it |
|---|---|---|
| Not interested | Tells the system to show less like this video; can also hide a creator or sound. | Long-press a video, or the share panel[1][5] |
| Why this video | Shows the broad reasons a video was recommended. | Share panel on a For You video[3] |
| Keyword filters | Hide videos with specific words or hashtags; Smart Keyword Filters add similar terms. | Settings and privacy, Content preferences[7] |
| Manage topics | Turn more than 10 topics up or down, plus an AI-generated content setting. | Settings and privacy, Content preferences, Manage topics[7][8] |
| Refresh your For You feed | Full reset to a new-account feed; follows and settings stay. | Settings and privacy, Content preferences[5][6] |
After a refresh, TikTok builds the new feed from your next interactions.[5] Watch what you want more of all the way through, like and follow deliberately, and swipe away from the rest quickly. Given the 2024 study, a few likes and follows should steer things faster than passive watching alone.[14]
What “first-hour testing” really means
Creator advice often says TikTok shows every new video to a small test group and then decides within an hour whether to push it. TikTok has not published a first-hour rule or a test-group size. What it has said is narrower: it deliberately mixes in videos to measure how popular they are with wider audiences, and it ranks everything by predicted interest per viewer.[1]
In practice that means a new video gets a first audience, and how those viewers behave decides whether it keeps getting shown. There's no fixed clock. Videos can pick up views days later, and TikTok's old guidance said For You draws from roughly the last 90 days.[4] Don't delete and repost a video because the first hour was slow. Duplicates are filtered out of the feed.[1]
Watch time, completion and rewatches
Completion is the one signal TikTok has singled out as strong.[1] Predicted time spent and the chance of finishing or skipping are both in the Transparency Center list.[2] When TikTok redesigned its creator payouts in March 2024, the Creator Rewards Program scored videos on four things: originality, play duration, search value and audience engagement.[17] Play duration covers viewing time and how often a video is watched to the end.[17]
Rewatches are less clear. TikTok's public lists don't name rewatches as a signal. The WSJ bots used replays as part of their behavior, and their feeds narrowed quickly.[13] It's reasonable to assume a replay adds watch time, but treat “rewatches are weighted extra” as unconfirmed.
- Cut dead air at the start and end. Finishing a video is a strong signal, and video length is itself one of the listed inputs, so a tight edit makes finishing easier.[1][2]
- Make length fit the idea. Creator Rewards only pays on original videos over a minute long, but that's a payout rule, not a ranking rule.[17]
- Write captions and on-screen text people would search for. Searches feed recommendations, and search value is one of the four Creator Rewards criteria.[3][17]
What it means for brands posting at scale
If you run many accounts or post a lot of UGC, the algorithm's design tells you what to avoid more than what to chase.
Volume doesn't buy reach by itself
TikTok told creators in 2020 that how many videos you post won't change how your content is recommended.[4] More posts give you more chances, not a better score per video. The Metricool data shows the other side: total posting volume rose sharply while per-post views and reach fell.[19] More supply means more competition for each slot.
Near-duplicates fight each other
The diversity rules work against a batch of near-identical videos. The feed avoids the same sound or creator twice in a row and swaps out videos too similar to their neighbors.[1][2] Unoriginal or reused material without anything new can be made ineligible for the For You feed altogether.[15] If you post twenty cuts of the same ad, vary the hook, the creator, the sound and the first frame, not just the caption.
Stay inside the automation rules
TikTok bans tools or scripts built to get around its systems and services that fake engagement, and it removes inflated metrics when it finds them.[15] Publishing real videos through approved posting tools is a different activity from bots that like, follow or inflate views, which fall squarely under the ban. Keep that line clear.
Label AI content
Realistic AI-generated people or scenes need a label, and unlabeled content can be removed, restricted or labeled by TikTok depending on harm.[15] Viewers can now also turn down AI content in their own feeds,[8] so a fully synthetic UGC strategy has a ceiling you don't control. Our guide to the TikTok AI-generated content label covers the details.
| Habit | Why it works against you | Better approach |
|---|---|---|
| Reposting the same clip across accounts | Duplicates are filtered; reused material can lose For You eligibility.[1][15] | Re-shoot or re-edit with a new hook and creator. |
| Same trending sound on every post | The feed avoids the same sound back to back.[1] | Mix sounds; use original audio where it fits. |
| Deleting and reposting slow videos | Duplicates are filtered; videos can still grow later.[1][4] | Leave it up and learn from it. |
| Engagement pods or bought views | Banned; fake metrics get removed.[15] | Spend that budget on more creators. |
| Unlabeled realistic AI video | Can be removed, restricted or labeled.[15] | Use the AI label. |
Running the test loop with an API
Because each video is judged on how its first viewers respond, the practical move is to post many genuinely different variants and keep what holds attention. UGC API lets you publish those variants across creator accounts from one call and compare results in one report. See how to A/B test UGC hooks and captions and our notes on UGC posting cadence.
What TikTok still hasn't made public
- The actual weights of any signal, and how often they change.
- Whether there's a set test audience for new videos, and how big it is.
- Whether rewatches count separately from watch time.
- How photo posts are ranked compared with video.
- What the US retraining changed, when, and how US and non-US feeds now differ.[9][11]
Anyone selling a precise formula for these is guessing. Stick to what TikTok has said and what your own analytics show. For the Instagram side, read how the Instagram algorithm works.
References
- How TikTok recommends videos #ForYou. TikTok Newsroom, 2020-06-18.
- Introduction to the TikTok recommendation system. TikTok Transparency Center.
- Learn why a video is recommended For You. TikTok Newsroom, 2022-12-20.
- 5 tips for TikTok creators. TikTok Newsroom, 2020-07-30.
- An update on our work to safeguard and diversify recommendations. TikTok Newsroom, 2023-03-16.
- TikTok rolling out a For You feed reset button. 9to5Mac, 2023-03-16.
- More ways to discover new content and creators you love. TikTok Newsroom, 2025-06-03.
- TikTok now lets you choose how much AI-generated content you want to see. TechCrunch, 2025-11-18.
- Announcement from the new TikTok USDS Joint Venture LLC. TikTok Newsroom, 2026-01-23.
- A TikTok US power outage caused a 'cascading systems failure' leading to multiple bugs. Engadget, 2026-01-27.
- TikTok users face uncertainty as platform shifts algorithm. Spectrum News, 2026-04-19.
- Leaked TikTok doc reveals its obvious secret to an addictive feed. Gizmodo, 2021-12-06.
- The Wall Street Journal's bot investigation into TikTok's recommendation algorithm. Tubefilter, 2021-07-21.
- TikTok and the Art of Personalization: Investigating Exploration and Exploitation on Social Media Feeds (Vombatkere et al., WWW '24). arXiv, 2024-03-19.
- Community Guidelines: Integrity and authenticity. TikTok.
- Community Guidelines: For You feed eligibility standards. TikTok, 2026-08-25.
- Introducing the new Creator Rewards Program. TikTok Newsroom, 2024-03-18.
- TikTok study (vendor study, 2.3M posts). Metricool, 2026-05-12.
- TikTok content volume up nearly 80 percent in early 2026 as per-post performance declined, report finds. Net Influencer, 2026-06-16.
- Carousel vs. video performance on TikTok and Instagram (vendor study, ~700K posts). Fanpage Karma.
FAQ
How do I reset my TikTok algorithm?
Go to Profile, tap the three-line menu, then Settings and privacy, Content preferences, and Refresh your For You feed. Your feed restarts as if you'd just signed up, and your follows and settings stay the same.[5][6]
How do I change my TikTok For You page without a full reset?
Use Manage topics to turn categories up or down, add keyword filters, and tap Not interested on videos you don't want.[7][1] These change one topic at a time and keep the rest of your feed.
Did the TikTok algorithm change in 2026?
The new US joint venture, announced 23 January 2026, said it will retrain, test and update the algorithm on US user data.[9] TikTok hasn't published what has changed so far, and reports of lower reach in early 2026 overlapped with a data center outage.[10][11]
Does follower count affect the For You page?
TikTok says follower count and past high-performing videos are not direct factors in For You recommendations.[1] Each video is judged on how viewers respond to it.
Does posting more often help on TikTok?
TikTok has said the number of videos you post doesn't change how your content is recommended.[4] More posts give you more chances, but each video is still ranked on its own.