TikTok does not publish an algorithm formula, and that is the most important thing to know before reading any article with “algorithm” in the title. The TikTok pages opened for this guide give no weights, no test-batch size, no watch-time percentage, no posting frequency, and no best time. What they do give is something more useful: a description of what the system measures, a long list of what it keeps out of the For You feed, and the controls viewers have over what they see. This guide sticks to those three things, using only TikTok’s own pages, and says plainly where the record stops.
If you want the viral angle, how a video earns its first push and what creators usually get wrong about it, read our companion guide on making a TikTok go viral. This one is about the system underneath: the scoring, the exclusions, and the controls.
What does TikTok say the algorithm actually measures?
TikTok’s recommendation-system explainer describes recommendation as prediction. When a video enters the pool, the system estimates how likely a particular viewer is to interact with it: to like it, share it, comment on it, follow the author, finish it, skip it, favorite it, or tap its soundtrack. Those predictions combine into an overall score, and higher scores mean a closer match to that viewer’s preferences. Videos are then ranked by prediction score, highest first, and the feed is built from the top-ranked videos that pass a similarity check.
The explainer page lists the signals feeding those predictions in three buckets. Viewer-side signals include videos watched, liked, or shared; how other people with overlapping taste interacted; region; device language and operating system; recently interacted-with videos, including ones watched, finished, liked, or skipped; and follow relationships. Video-side signals include time posted, region posted from, the author’s language setting, soundtrack, length, and hashtags. The third bucket is device and account settings, which TikTok says receive lower weight than the others because users do not actively express them as preferences.
Two of TikTok’s own examples show how the weighting works in practice. Finishing a longer video from beginning to end is a strong indicator of interest and receives greater weight. Whether the viewer and the creator are in the same country is a weak indicator. And one sentence on the newsroom page settles the most repeated myth in creator advice: neither follower count nor whether an account has had previous high-performing videos are direct factors in the recommendation system. A bigger account’s video is likely to receive more views, TikTok adds, by virtue of the larger follower base seeing it, not because the system boosts it.
Moderation gates the pool. The explainer page says the system retrieves videos that have passed initial content moderation and are eligible for recommendation; content violating the Community Guidelines is removed, and content not suitable for a broad audience is made ineligible for recommendation. Videos just uploaded or under review may also be ineligible, per TikTok’s newsroom post.
What gets a video kept out of the For You feed?
This is the longest confirmed section of TikTok’s own documentation, and it is the spine of this guide. Counting the categories on TikTok’s For You feed eligibility page gives 15 ineligibility categories (our count, from the page itself). Content in these categories is made ineligible for the For You feed, and TikTok says it may also make it harder to find in search. Ineligible content can still be found by searching or by following the account; low views alone do not mean a video was ruled ineligible, since TikTok says a video with few views may simply have lacked community engagement.
The 15 categories, condensed from the page: youth safety (any content created by an account holder under 16); hate speech; disordered eating and body image; dangerous activity and challenges; nudity and body exposure; sexually suggestive content (adults engaging in intimate kissing, sexualized framing or behavior, sex products); shocking and graphic content; misinformation; civic and election integrity; fake engagement (content tricking or manipulating others to increase gifts or engagement metrics, such as like-for-like promises); unoriginal content (reproduced material uploaded without new or creative edits, including visible watermarks from someone else, and extremely short or exclusively GIF-based clips); gambling; alcohol, tobacco, and drugs; undisclosed commercial content (marketing not disclosed using the content disclosure setting); and TikTok LIVE content primarily directing people off-platform.
Two entries on that list deserve attention because creators confuse them. Fake engagement ineligibility covers tricks like like-for-like promises and false incentives, which is the same family of behavior our TikTok AI label guide covers from the disclosure side: TikTok treats undisclosed manipulation and undisclosed marketing as feed problems, not just rulebook problems. And undisclosed commercial content is ineligible, which connects to the standing question in our viral guide about how strictly TikTok treats undisclosed promotion. The page itself does not soften that entry.
What happens when similar content repeats?
Not everything TikTok limits is banned from the feed. A separate cluster rule covers content that is fine occasionally but problematic in clusters: dieting, extreme fitness, sexual suggestiveness, sadness (statements of hopelessness, sad quotes), and overgeneralized mental health information (such as a quiz claiming to diagnose someone). This content may be eligible for the For You feed, but TikTok says it will interrupt repetitive patterns so it is not viewed too often.
That cluster list also resolves a distinction creators often blur. “Sexually suggestive content” in the ineligibility list means adults engaging in intimate kissing, sexualized framing or behavior, and sex products, and it is ineligible outright. “Sexual suggestiveness” in the cluster list is a milder theme that stays eligible but gets interrupted when it repeats. The page lists the two separately: the ineligible entry names intimate kissing, sexualized framing or behavior, and sex products, while the cluster entry names sexual suggestiveness as a theme.
The same variety instinct runs through the ranking itself. After ranking by prediction score, TikTok applies a similarity check: if top-ranked videos are too similar, for example because they use the same sound, they are replaced with others to ensure greater variety. The newsroom page states the feed generally will not show two videos in a row made with the same sound or by the same creator, and that TikTok does not recommend duplicated content, content already seen, or spam.
What controls do viewers actually have?
TikTok gives viewers feed-shaping tools that creators should understand because they are part of the system, not a workaround for it. Long-pressing a video and selecting “Not Interested” limits that type of content. Video keyword filters let a viewer filter out words or hashtags from the For You feed. For You refresh lets a viewer explore entirely new sides of TikTok. And viewers can hide videos from a given creator or made with a certain sound. Reporting a video that seems out of line with the guidelines is a separate safety action, listed on the newsroom page alongside these controls.
For creators, the practical reading is straightforward. A viewer who filters your topic out is not a viewer you lost to the algorithm; it is a viewer who used a control TikTok built for exactly that. Chasing that viewer with more of the same content runs against the system’s own variety push: the similarity check replaces videos that are too alike, and the cluster rule interrupts repetition.
What has changed in 2026, and what has not?
Here is where honesty requires labels. The TikTok pages opened for this guide show no version date on their fetched text: the recommendation explainer and the eligibility page carry no version date, and the newsroom post’s fetched text carries no date line either (third-party summaries date it June 2020). Third-party guides report the guidelines version as released 25 August 2026 and effective 24 September 2026, but that dating is third-party-reported, not page-verified, and this guide does not present it as TikTok’s own statement. No new official 2026 newsroom algorithm post was found in this research. Third-party summaries claim the June 2020 newsroom explainer has never been publicly revised, but that is a third-party claim, not TikTok’s statement, and this guide does not repeat it as fact.
What can be said from the pages reviewed is that the system TikTok describes is a prediction system with no published formula, a 15-category ineligibility list, a cluster rule, and viewer controls. Those pages do not establish whether every mechanism remains unchanged in 2026. Regulatory stories from 2026, such as preliminary findings about minors’ content in the EU and reports of a retrained US model, are third-party and regulatory claims, not TikTok statements about the recommendation system, and they are not presented here as algorithm changes. If TikTok publishes a new explainer, this guide’s frame changes with it. Until then, the absence of a formula is itself the finding.
The table below is our own synthesis of the pages above, not TikTok’s table. It maps each signal group to what TikTok says it is and what it does not mean.
| Signal group | What TikTok says it is | What it does not mean |
|---|---|---|
| Viewer interactions | Likes, shares, comments, follows, finishes, skips, favorites | Not a vote count: predictions estimate likelihood per viewer, not a global score |
| Video information | Captions, sounds, hashtags, length, language, time and region posted | Not a hashtag lottery: hashtags are one video signal among many |
| Device and account settings | Language preference, country setting, device type | Not a preference: lower weight because users do not actively express them |
| Follower count and past hits | Explicitly not direct factors | Not a boost: bigger accounts get more views through a bigger base, not system favoritism |
Does TikTok’s algorithm favor big accounts?
Not directly. TikTok states that neither follower count nor whether an account has had previous high-performing videos are direct factors in the recommendation system. A larger account’s video is likely to receive more views because more followers see it, not because the system boosts it.
Is there a best time to post on TikTok?
No best time, frequency, or length is published on the TikTok pages opened for this guide. Time posted is listed as a video signal, but no hour is recommended on those pages.
Does the 3-second hook rule come from TikTok?
Not on any TikTok page opened for this guide. Creator Academy advice, including the 3-second hook, was not confirmed on TikTok’s own pages in this research, so this guide does not present it as TikTok’s guidance.
Why would a video get few views if it was not ruled ineligible?
TikTok says low views do not necessarily mean ineligibility: a video with few views may be due to a lack of community engagement rather than being ineligible. Ineligible content can still be found via search or by following the account.
What is the difference between ineligible content and clustered content?
Ineligible content is kept out of the For You feed entirely, across 15 categories on TikTok’s eligibility page. Clustered content (dieting, extreme fitness, sexual suggestiveness, sadness, overgeneralized mental health information) stays eligible, but TikTok interrupts repetitive patterns so it is not viewed too often.
Has TikTok changed its algorithm in 2026?
No algorithm change was confirmed on TikTok’s own pages opened for this guide. Guidelines version dates reported by third parties (released 25 August 2026, effective 24 September 2026) are not visible on the fetched pages, and no new official 2026 algorithm post was found. The pages describe a prediction system with no published formula.
Sources (TikTok’s own pages, opened and read 28 October 2026): Recommendation system explainer; How TikTok recommends videos #ForYou (newsroom); For You feed Eligibility Standards. Note on dating: none of the three fetched pages shows a version or publication date; the guidelines version dates reported by third-party guides are not presented here as TikTok’s own statements. Internal reading: How to Make Your TikTok Go Viral, TikTok AI-Generated Content Label. Last verified: 28 October 2026.
