How the TikTok algorithm works (what TikTok has actually said)

How does the TikTok algorithm work?

TikTok's Transparency Center says: "Videos with the highest score are selected to be served to your For You feed." The score is an estimate of how likely you are to interact with each video. TikTok describes the system as several machine learning models that go through five broad steps, shown in the diagram.

The Transparency Center's own words for each step: selection retrieves "a large number of videos that have passed our initial content moderation processes"; prediction estimates "the probability of liking, sharing, commenting, or skipping a given video" and combines them into an overall score; ranking "sorts videos by their prediction scores, from high to low"; a similarity check replaces videos that are too similar, for example ones that use the same sound; and recommendation rules apply at different points. TikTok publishes no pool sizes, so unlike some other platforms there is no number to quote here.

One more official detail: the Help Center says the system also looks at others. In certain recommender systems, predictions "also consider interactions from others who appear to have similar interests." Its example: if user A liked videos 1, 2 and 3, and user B liked 1 to 5, the system may predict user A will like videos 4 and 5.

Is a video not shown because of eligibility or ranking?

They are two different questions, and TikTok answers them on different pages. Eligibility asks whether a video may be recommended at all. Ranking asks, among videos that may be recommended, which ones this viewer sees first. A video that is not eligible is never scored against the others; a video that is eligible can still sit low because its predicted score is lower than other videos' scores for that viewer.

TikTok's own order, from the Transparency Center: the first step retrieves videos "that have passed our initial content moderation processes", the later steps predict and rank them, and recommendation rules apply "at different points". So eligibility is mostly a gate in front of ranking, and some rules also act during ranking.

Source and date: TikTok Transparency Center and Community Guidelines pages, read October 24, 2026. The last two rows state what is missing from those pages, which is our reading of them.

Our suggestion, not a TikTok statement: if one video got far fewer views than your others, look in analytics for an ineligible mark first. If there is none, the cause is not an eligibility ruling you can see, and TikTok does not tell creators more than that.

Which signals does TikTok name for the For You feed?

The table lists only what TikTok's pages say. TikTok calls both lists examples ("non-exhaustive"), and says the inputs "are dynamic and constantly updated", so treat the table as a sample of what TikTok chose to publish, not the whole system.

Source and date: TikTok Help Center and Transparency Center pages, and the TikTok Newsroom post of June 18, 2020, all read October 24, 2026. The Help Center and Transparency Center pages carry no publication date. Method: we copied the wording from the saved page text and checked each quote by script.

Where this stops: TikTok does not say how much more interactions weigh, how a share compares with a like, or how many seconds of watching count. Anything with a number attached (a percentage of viewers, a point per like) is not from TikTok.

Do follower count or past hits matter?

TikTok's 2020 Newsroom post says: "neither follower count nor whether the account has had previous high-performing videos are direct factors in the recommendation system." It adds that a video is likely to get more views from an account with more followers "by virtue of that account having built up a larger follower base." So followers help reach through your own audience, but TikTok says they are not a ranking input for For You.

Two nuances from the Help Center keep this honest. For recommended accounts and the LIVE feed, TikTok does list follower counts as content information. And the post is from 2020; TikTok has not republished that sentence in the newer Help Center page we read, so we cannot say whether it was reconfirmed.

Does TikTok rank Following, Friends, LIVE and Search the same way?

No. The Help Center describes each feed separately with the same three groups of signals but a different example list and a different group that is generally weighted most. The ranked For You feed is only one of them.

Source and date: TikTok Help Center, read October 24, 2026 (the page is undated). The "Examples" column is a summary of TikTok's own lists, not a complete list.

For Search, TikTok says match to the query is generally weighted more than interaction history. That is the only feed on this list where TikTok states that content information leads. Our practical reading: captions and spoken words that match what people search can matter for Search, but TikTok publishes no ranking recipe for it, so we do not give a keyword formula.

What does TikTok say it does not recommend?

Eligibility is an official, written filter that runs before ranking. TikTok says only videos that pass moderation are in the pool that can be recommended; content that is not suitable for a broad audience can be made ineligible for the For You feed. The Transparency Center states two hard rules: videos should include content from creators in your region, and "content created by accounts aged under 16 is ineligible for recommendation."

The current For You feed Eligibility Standards were released August 25, 2026 and took effect September 24, 2026 (TikTok Community Guidelines page, read October 24, 2026). The page says: "If a video isn't getting many views, it also doesn't necessarily mean it broke a rule," and that creators can check analytics "including if there were any made ineligible for recommendation."

Other official limits on repeated content: the system aims not to repeat a video you have already seen ("We also typically don't recommend content you've already seen"), and the similarity check swaps out videos that are too alike, for example ones using the same sound. TikTok also aims to interrupt frequent recommendations of some topics, such as extreme fitness or dieting.

How do I reset the TikTok algorithm?

TikTok calls it refreshing your For You feed. Per the Help Center, "When you select the option to refresh your feed, we'll show you some popular content. Your interactions with this content, such as liking and commenting, will help reshape your For You feed recommendations." It is not a wipe of your account: "only your For You feed is affected," not Following, profile, inbox or ads.

1. Open TikTok and tap Profile at the bottom.
2. Tap the Menu button at the top.
3. Tap Settings and privacy, then Content Preferences.
4. Tap Refresh your For You feed and follow the on-screen steps. TikTok warns: "Once you refresh the feed, it can't be undone."
5. Alternative path "in select regions": press and hold a video (or tap Share), tap Why this video, tap Adjust your For You, then tap Refresh your For You feed.
6. Interact with the popular videos you are shown: like and comment on what you want more of. TikTok names liking and commenting as the interactions that reshape the feed.

Not tested by us: we have no TikTok account, so we did not run these steps in the app and cannot confirm what the screens look like today or how long the feed takes to change. The steps are copied from the Help Center page read in headless Chromium on October 24, 2026. The page is undated. Third-party guides disagree about it: one says you cannot completely reset the algorithm (while also describing refresh), another says refreshing "clears past signals". Neither matches TikTok's own wording, so we use TikTok's.

Gentler options before you refresh, all from TikTok's Help Center: long-press a video and choose Not interested ("we'll show you less content like that"); Manage topics to adjust your preferences for different topics (the Help Center gives no more detail, and the Newsroom says it changes how often topics are recommended); Filter keywords to remove words and hashtags from For You, Following, Friends, LIVE and STEM feeds (with an AI option for synonyms). Switching personalization off exists only in the EU and Kazakhstan, per TikTok's note; there the For You feed becomes a Popular feed.

Has the TikTok algorithm changed?

TikTok publishes no ranking change log, so there is no official date for "the algorithm changed" on any given day. What TikTok has published are dated changes to controls, policy and ownership. The table lists each one, most recent first. Posts that claim a specific hidden change on a specific day are not confirmed.

Sources and dates: the TikTok Newsroom posts and the Community Guidelines page named in the Source column, all read October 24, 2026. Dates in the first column are the dates TikTok gives; "Undated" rows are Help Center or Transparency Center pages that show no date.

If your feed or reach suddenly feels different, TikTok's own pages give three non-secret reasons it can happen: input signals change, "how important each factor is can change over time," and your own recent interactions reshape what you see. TikTok does not publish a status page for ranking changes.

What has TikTok not confirmed about the algorithm?

Claims made in the top three guides for this query (Sprout Social, Hootsuite, Buffer, read October 24, 2026), or common in creator advice, that we did not find on any TikTok page we read. Treat each as not confirmed, not as false:

• A point system. TikTok's pages list signals and predictions, never points per action.
• That watch time, or the first few seconds, is "the strongest ranking signal." One guide states it; TikTok only says interactions, which may include time spent watching, are generally weighted more heavily for most users. The first-seconds claim is not on the pages we read.
• A "qualified view" that must exceed 5 seconds, or a test batch that decides whether a video stops. One guide (third-party) states both; no TikTok page we read does.
• A fixed first-hour window, or a fixed number of first views (often given as 1,000), that decides whether a video keeps being shown. The TikTok pages we read name no such window or number. They say only that the system keeps learning from reactions and that its inputs change over time, so we cannot confirm or rule out a test stage.
• A best time or day to post. TikTok lists "the time the video was posted" and "time zone and day" as signals, but gives no hour that helps. See our best time to post guide for data, which is not a TikTok statement.
• A hashtag count, caption length or ideal video length. TikTok lists hashtags and video length as signals, which is different from saying a number ranks better.
• That posting more often, deleting a video or editing a caption changes reach.
• "Shadowbanning" as a named mechanism. TikTok's eligibility page describes videos made ineligible for recommendation, with analytics showing which ones; it does not describe a hidden account penalty.

What should I do with this?

Pick the situation that matches yours. Official means a TikTok page we read states it; Our suggestion is our reading and not a TikTok statement. TikTok publishes no weights and no targets, so none of these promise a result.

What TikTok says, what we suggest.

Sources and dates: the Official column is taken from TikTok pages read October 24, 2026 (Help Center, Transparency Center, For You feed Eligibility Standards effective September 24, 2026). The Our suggestion column is our reading of those pages; we did not test it on a TikTok account.

Where this stops: TikTok does not say which change will move your reach or how long it takes. For how Instagram describes its own ranking, see our Instagram algorithm guide; the two platforms publish different things, so do not transfer numbers between them.

Limits and how we checked

We read every TikTok page listed under sources on October 24, 2026. The Help Center, Transparency Center and Community Guidelines pages are client-rendered, so we read them in headless Chromium; the Newsroom pages are server-rendered. We then ran a script that checks each excerpt we quote against the saved page text (we keep the log). Help Center and Transparency Center pages show no publication date, so for those we give the date we read them.

What we could not do: we did not test the reset steps in the app (no account), the Manage topics Help Center article would not load for us, so topic controls are sourced from the Newsroom post only, and we found no TikTok post describing the AI content control as fully launched. We did not measure any reach ourselves, and we repeat no third-party performance numbers as fact.

Koldflux is not affiliated with or endorsed by TikTok. TikTok is a trademark of its owner. This page explains public statements; it is not advice about your account.