Key takeaways
  • TikTok's own 2020 post names three signal groups (your interactions, video information, device and account settings), says a strong signal like finishing a video outweighs a weak one, and says follower count is not a direct factor. Nothing published since replaces it.
  • The small-test-then-wider model is a fair inference from how views arrive, but TikTok has not published batch sizes, time windows or pass marks for creators; one seller education page mentions about 500 people for sellers. Numbers like "the first 200 viewers" are guesses.
  • In a September 2026 sample of 58 creators, the typical creator's best post got 46 times the plays of their worst over ten posts, and 50 of 58 had a gap of at least 10 times. One weak post is the normal shape, not a penalty.
  • Shadowban, posting time, hashtags, links, deleting videos and warming up an account are each answered with what TikTok has said and what the data shows. Most reduce to the same mistake: treating one video's result as a verdict on the account.
  • To retrain your own feed: Settings and privacy, Content preferences, Refresh your For You feed, then Not interested, Manage topics and Keyword filtering.

The TikTok algorithm is the system that picks which videos land on your For You feed, and it works one video at a time: for each video it guesses how likely you are to watch to the end, like, share or save, based on what you did with earlier videos, and shows the ones it rates highest. In a September 2026 sample of 572 posts from 58 creators, the typical creator's best recent post got about 46 times the plays of their worst, which is what judging every video on its own looks like.

That much TikTok has said itself. Most of the rest of what people repeat about the TikTok algorithm is guesswork. This page sorts the claims into three piles, what TikTok has published, what real posting data shows, and what creators report, and labels each one.

How the TikTok algorithm works, one video at a time: 46x gap between a typical creator's best and worst post; 50 of 58 creators had a gap of 10x or more; 40% of a creator's plays came from one post (median); 267K 7-day searches for "algorithms for tiktok"
What a September 2026 sample of 58 creators and 572 posts shows about the For You feed, plus TikTok search demand in the 7 days to 5 October 2026.
Share or embed this infographic

Free to use on your site, in a deck or a newsletter with a credit link to TokConnect. Copy the code below.

Download PNG

What has TikTok published about the algorithm?

The fullest description is still TikTok's own newsroom post from June 2020. It names three groups of signals. First, your interactions: the videos you like and share, the accounts you follow, the comments you post and the videos you make. Second, video information: captions, sounds and hashtags. Third, device and account settings: your language, country and device type. TikTok says the third group carries less weight, because you never chose those as preferences.

Two more points from the same post settle most arguments. A strong signal, such as finishing a long video, counts for more than a weak one, such as the viewer and the creator being in the same country. And neither follower count nor a creator's earlier hits is a direct factor in whether a video gets recommended. Both are still quoted as current by Sprout Social (February 2026) and Hootsuite (June 2026), and TikTok has published nothing that replaces them.

Since 2020 TikTok has published controls rather than mechanics.

WhenWhat TikTok publishedWhat it does
June 2020How TikTok recommends videos #ForYouNames the three signal groups, strong versus weak signals, and says follower count is not a direct factor
March 2023Refresh your For You feedResets your recommendations so the feed behaves as if you had just signed up; followers, likes and posts are kept
April 2024For You feed Eligibility Standards and Account CheckSets out which videos are kept off the For You feed, warns that repeat offenders lose For You eligibility as an account, and adds a tool that audits your last 30 posts
August 2024 to June 2025Manage Topics and Smart Keyword FiltersSliders for more or less of about ten topics, and a keyword filter that also catches synonyms, with the limit rising from 100 to 200 words
August 2025Five-hashtag cap per postCaps hashtags at five; TikTok gave no reason when asked
January 2026US joint venture closesThe US business gets a licensed copy of the algorithm, to be retrained on US data only; TikTok has not said what this changes for creators

TikTok's own pages: the 2020 recommendation post, the 2023 refresh post, the For You feed Eligibility Standards and the Making your feed For You guide. Dates and details confirmed through TechCrunch (March 2023), Variety (April 2024), TechCrunch (June 2025), Social Media Today (August 2025) and Music Business Worldwide (December 2025).

In 2026 the US deal closed on 22 January, and a ByteDance memo reported in December 2025 said the US copy would be retrained on US data only. Guides describing a "new 2026 algorithm" with new rules for creators cite no TikTok document, and we could not find one. Treat them as creator reports.

How does the TikTok algorithm decide who sees a new video?

The popular version: TikTok shows your video to a small test group, watches how they respond, then pushes it wider. Parts of that are confirmed, parts are inferred, and the numbers people attach to it are made up.

ClaimStatusWhere it comes from
Each video is judged on its own signals, not on your follower count or your past hitsPublished by TikTok2020 newsroom post
The score for a video predicts whether you will like it, comment on it and how long you will play it, and the goal is keeping you in the appLeaked internal document, authenticity confirmed by TikTokNew York Times, December 2021
Watch time alone is enough to steer what the feed shows youOutside testWall Street Journal bot study, July 2021
A new video goes to a small group first, then wider if they watch, share, save and commentInferred from how views arrive in analyticsCreator reports; no TikTok statement
The first group is "a few hundred" people, or your own followers firstUnconfirmedMarketing blogs; SocialPilot (August 2026) labels it "not officially confirmed"

The 2021 document was reported by the New York Times and summarised by DeepLearning.AI (December 2021). The Wall Street Journal's bot investigation is described by the Online Journalism Awards. The 2026 "followers first" claim is in SocialPilot's guide, which says it rests on community testing.

The staged model is a fair reading of what TikTok has published and what analytics show. TikTok has not published batch sizes, time windows or pass marks for creators. One seller education page mentions a "basic traffic pool" of about 500 people, but it is guidance for shop sellers, not a rule TikTok has said applies to creator videos, so anyone quoting "the first 200 viewers" is guessing.

See which of these claims your own account backs up: ask TokConnect to compare your top and bottom posts

What the data shows: the same account swings 46 times over

If each video is judged on its own, one account's videos should get very different view counts. They do. We took a September 2026 sample of 58 creators surfaced by TikTok search in six niches, ten recent posts each, and compared each creator's best and worst.

The typical creator's best post got about 46 times the plays of their worst. Fifty of the 58 had a gap of at least 10 times, and 10 had a gap of at least 100 times. The single best post took about 4 in 10 of all the plays across a creator's ten posts. One comedy account with 6.2 million followers ranged from 696 plays to 742,800 across ten posts. A personal finance account with about 117,000 followers ranged from 2,915 to 3 million.

Follower tierCreatorsMedian best-to-worst ratioCreators with a gap of 10x or moreMedian plays per post
Under 10,00052.8x1215
10,000 to 100,000758.3x712,130
100,000 to 1 million1740.7x1521,800
Over 1 million2956.5x27102,450
All 585845.5x5022,075

Data: TikTok profile and post data, 572 posts with at least one play. Ratio is each creator's highest play count divided by their lowest over their last ten posts; the tier figure is the median creator. Re-cut, full sample.

Bar chart of the median creator's best post divided by worst post over the last 10 posts: 2.8 times under 10,000 followers, 58.3 times for 10,000 to 100,000, 40.7 times for 100,000 to 1 million, 56.5 times over 1 million, 45.5 times across all 58 creators.
Best-to-worst ratio for the median creator in each tier. September 2026 sample of 58 creators, 572 posts. Re-cut.
Horizontal bar chart of 58 TikTok creators by the gap between their best and worst post: 54 had a gap of 5 times or more, 50 of 10 times or more, 10 of 100 times or more, and 4 under 3 times.
How many of 58 creators swung by 5, 10 and 100 times between their own posts. Same sample.

The only flat accounts were the smallest. Four creators had a gap under 3 times, and three of them had under 2,000 followers and no post over 700 plays. Once an account has any reach the swing appears, and it does not shrink as followers grow. Followers are not a floor under your views, which is what TikTok's "not a direct factor" line predicts.

So a post that gets a tenth of your usual views is not evidence of a penalty. It is the normal shape of an account. What separates your best posts from your worst is the thing worth studying. Traffic sources sit in your own TikTok Studio analytics, but the public counts are enough for a first pass.

Run the same check on your own account: TokConnect pulls your last 30 posts, ranks them by plays and shows what your top five share.

Try this question

Ask me for my TikTok handle, then pull the last 30 public posts from that account and rank them by plays. For the top five and bottom five, list the plays, likes, comments, shares, saves, video length, the topic in the caption and the posting day. Tell me what the top five have in common that the bottom five do not.

The search half

The For You feed is not the only way videos get found. TikTok said in September 2024 that 57% of its users use the search function. Adobe's February 2026 survey of 807 US consumers found 49% had used TikTok as a search engine, up from 41% in 2024, although only 4% of Gen Z said they were more likely to rely on TikTok than Google, down from 8%. TikTok has not published what share of all views comes from search.

You can see your own share. In TikTok Studio each video's detail page lists traffic sources: For You, Following, Personal profile, Search and Sound. The analytics guide explains each line. Creators report that search views keep arriving for weeks after the For You push stops, so a video with a high Search share is worth more than its first-day count suggests.

Search ranks differently: it matches your words against the caption, on-screen text, speech and sound, then orders the matches by predicted interest. That is its own subject, covered in how TikTok search works, with the practical side in TikTok SEO.

What people ask TikTok about the algorithm

Typed searches show the scale. On 5 October 2026 TikTok's topics for "tiktok algorithm" held two typed searches: "algorithms for tiktok" (266,520 searches in 7 days) and "tiktok changed algorithm" (153,953). The other 18 of 20 were TikTok topic groups that pool many searches, so we do not quote their figures.

Type "tiktok algorithm" into TikTok's search bar and the suggestions show what people worry about. Three of ten are about an update or change, one says it is broken, one asks how to reset it.

PositionTikTok search suggestion
1tiktok algorithm
2tiktok algorithm post time
3tiktok algorithm 2026
4tiktok algorithm is broken
5tiktok algorithm views
6tiktok algorithm 2026 update
7tiktok algorithm change
8tiktok algorithm explained
9Tiktok algorithm update
10tiktok algorithm reset

Data: TikTok search suggestions for "tiktok algorithm". Snapshot.

Six myths, answered

The biggest myth: the algorithm judges your account

It judges videos. TikTok's published position is that follower count and earlier performance are not direct factors, and the 58-creator sample shows the same account swinging from a few hundred plays to hundreds of thousands within ten posts. Almost every myth below comes from treating one weak video as a verdict on the account.

The algorithm judges your account

It does not. See the box above: it judges each video, and one weak video is not a verdict on the account.

Shadowbans

TikTok does not use the word, but it does keep videos off the For You feed under its eligibility standards, tells you on the video when that happens, and since April 2024 shows account-level restrictions under Account status, with an Account Check that audits your last 30 posts. Creators report drops with no notice, and some are real restrictions, but a drop alone proves nothing when a 46-fold swing is normal. The guide to getting more views covers how to check, and the free shadowban checker reads an account for you.

Posting time

TikTok's list of signals does not include the clock, and the 572 posts in our sample went up in every three-hour block of the day. Post when your own followers are online, which TikTok Studio shows you, and see the best time to post for the numbers.

Hashtags

Hashtags are one part of "video information", the group TikTok lists second, and since August 2025 you can only add five per post. The words in your caption, on screen and in your speech do the same job with no cap; hashtags versus keywords shows the difference in numbers.

TikTok has published nothing that says a link lowers reach. Links in captions are not clickable in the first place, and a clickable bio link needs 1,000 followers on a personal account, so the reports of "punished" link posts are about videos that told people to leave, which viewers skip, not about a rule.

Deleting videos hurts you

TikTok has published nothing on this, and its own position that each video is judged on its own signals leaves no room for a deleted video to drag down the next one. If you are nervous, set the video to private instead; the effect on your other posts is the same, which is none that anyone has measured.

You need to warm up a new account

TikTok has published nothing about a warm-up period, and its 2020 post says a new account is not held back by having few followers. Creators are split, and the people most certain that warm-up is required tend to sell a guide to doing it.

How do you reset or retrain your For You feed?

Four tools, all from TikTok. They change what you see, not how your own posts are distributed.

  1. Refresh the feed. Open your profile, tap the menu in the top right, then Settings and privacy, then Content preferences, then Refresh your For You feed, and confirm. TikTok says the feed then behaves as if you had just signed up. Your followers, likes and posts stay.
  2. Tell it what you do not want. Press and hold any video and tap Not interested. Do it for a run of videos, not once.
  3. Set topic sliders. Settings and privacy, Content preferences, Manage topics. Move each topic up or down. TikTok says this changes how often a topic appears, not whether it appears at all.
  4. Filter words. Settings and privacy, Content preferences, Keyword filtering. Add words, phrases or hashtags you do not want to see; the smart filter also catches synonyms. TechCrunch reported the limit rising from 100 to 200 words in June 2025.

Then watch what you want more of, all the way through, for a few sessions. Finishing a video is the strongest signal TikTok has named. Steps confirmed against Metricool's guide (June 2026) and TechCrunch (June 2025); menu labels can differ slightly by region and app version.

How we got these numbers

Statements about how the feed works come only from TikTok's own published pages, labelled as such; we have no access to TikTok's code. The account data is a September 2026 sample of 58 creators surfaced by TikTok search in six niches (fitness, cooking, personal finance, small business, comedy, skincare), each creator's last 10 public posts, 572 posts with at least one play, pulled through TokConnect on 21 September 2026. The swing is each creator's highest play count divided by their lowest; tier figures are the median creator. Ten posts is a short window, and the sample reflects accounts TikTok search surfaces, not all accounts.

Search figures come from one search-count pull for "tiktok algorithm" on 5 October 2026: TikTok's 7-day worldwide search counts, typed searches only. A record counts as a TikTok topic group if its 7-day figure is 5,000,000 or more, or 1,000,000 or more with the week's first daily reading under 5% of its last, or TikTok files it as Featured Content. Data: swing re-cut, full sample, autocomplete, search topics, 5 October 2026.

Key statistics

  1. The typical TikTok creator's best post got 45.5 times the plays of their worst over their last 10 posts, in a September 2026 sample of 58 creators and 572 posts.

    TikTok post data, 21 Sep 2026#
  2. 50 of 58 TikTok creators in a September 2026 sample had a best post at least 10 times their worst over their last 10 posts, and 10 had a gap of at least 100 times.

    TikTok post data, 21 Sep 2026#
  3. A TikTok creator's single best post took a median 40% of all plays across their last 10 posts, in a September 2026 sample of 58 creators.

    TikTok post data, 21 Sep 2026#
All 5 statistics
  1. TikTok's June 2020 explanation of the For You feed names three groups of ranking signals (user interactions, video information, device and account settings) and says follower count is not a direct factor; TikTok has published nothing that replaces it.

    TikTok Newsroom, June 2020#

Frequently asked questions

How does the TikTok algorithm work?

TikTok scores each video for each viewer using three groups of signals it published in 2020: what you have done on the app (watched, liked, shared, followed, commented, posted), what the video is about (caption, sound, hashtags, and now speech and on-screen text), and your device and account settings, which count for least. Strong signals such as finishing a long video weigh more than weak ones. Follower count is not a direct factor. Videos then appear to go out in stages, wider when the first viewers watch, share and save, though TikTok has not published how those stages work.

How do I reset my TikTok algorithm?

Open your profile, tap the menu, then Settings and privacy, Content preferences, Refresh your For You feed, and confirm. TikTok launched this in March 2023 and says the feed then behaves as if you had just signed up, while your followers, likes and posts stay. Follow it by pressing and holding videos you do not want and tapping Not interested, setting Manage topics sliders, and adding words to Keyword filtering. Then watch what you do want all the way through for a few sessions.

Does the TikTok algorithm favour new accounts?

TikTok has not said so. Its published position is that follower count and a creator's past performance are not direct factors, which means a new account is not held back by being new, but it is not boosted either. Creators report that early posts on a new account sometimes get an unusual number of views. In a September 2026 sample, the five accounts under 10,000 followers had a median of 215 plays a post, so "new account boost" is not something you can plan around.

Do hashtags help the TikTok algorithm?

A little. TikTok lists hashtags as part of "video information", the second of its three signal groups, alongside captions and sounds. Since August 2025 you can add at most five per post, and TikTok gave no reason for the cap. Words in your caption, on screen and spoken aloud do the same job of telling TikTok what the video is about, with no limit, and they also help the video appear in search. Use three to five relevant hashtags and spend your effort on the words.

Is there a TikTok shadowban?

Not under that name. TikTok publishes For You feed Eligibility Standards that keep some videos off the feed, tells you on the video when that happens, and since April 2024 shows account-level restrictions under Account status, with an Account Check that reviews your last 30 posts. So restrictions exist and are visible. A sudden drop in views on one post is not one of them: in a September 2026 sample of 58 creators, the typical creator's best post got 46 times the plays of their worst.

Does posting time affect the algorithm?

TikTok's published list of signals does not include when you post. Posting time matters only in the ordinary way: if your followers are awake, the first viewers arrive sooner. In a sample of 572 posts from 58 creators, videos went up in every three-hour block of the day. Post when TikTok Studio shows your own followers are online, hold the slot for two weeks, and judge by first-hour views.

Sources and data