The TikTok algorithm is a set of recommender systems that choose from eligible videos and rank them for each viewer by predicting how likely that person is to watch, like, share, comment on or skip each one. TikTok's Help Center page on recommendations sorts the inputs into user interactions, content information such as sounds and hashtags, and user information such as language and location, and says interactions, including time spent watching, generally carry the most weight for most users. Videos that fall short of the For You feed eligibility standards are left out of that feed however well they might rank.
TikTok has published a fair amount about this, but it is scattered across a 2020 Newsroom post, a Help Center article, a Creator Academy explainer and the Community Guidelines. This guide pulls those sources together, flags where they differ, matches every signal TikTok names to a number you can read in TikTok Studio, and ends with a weekly review sheet. Search optimisation and diagnosing a sudden drop in views are separate jobs, so they only get a mention here.
The six stages TikTok describes
TikTok's Creator Academy explainer on the recommendation system breaks a For You recommendation into stages. In plain terms:
- Reading preferences: the system reviews what a viewer has recently watched, finished, liked, skipped, commented on or saved.
- Choosing candidates: it pulls videos connected to those interests, taking into account the viewer's location and the region where each video was posted.
- Predicting: it estimates how likely the viewer is to like, share, comment on or skip each candidate.
- Ranking: candidates are sorted by those predictions, and the top-ranked ones move on.
- Checking similarity: videos too similar to each other are swapped out so the feed has variety.
- Applying rules: a last layer of rules shapes the feed, such as making sure it includes creators from the viewer's region.
The Help Center adds a detail many creators miss. Some of TikTok's systems also learn from people with similar tastes: if two viewers have liked the same run of videos, a video one of them liked can be predicted to suit the other. Your audience is partly defined by what else its members watch, and TikTok Studio's Viewers tab shows the creators and posts your audience also watched recently, which is the nearest view you get of that neighbourhood.
The signals TikTok names, matched to TikTok Studio
Each row lists a signal TikTok has published, where it says so, and the closest metric in TikTok Studio. Metric names follow TikTok's Creator Academy guide to analytics. TikTok doesn't publish weights, so the last column is a reading aid, not a score.
| Signal TikTok names | Where TikTok says it | Closest TikTok Studio metric | How to read it |
|---|---|---|---|
| Watching to the end, and time spent watching | The Newsroom post calls finishing a longer video a strong indicator of interest; the Help Center says watch time is generally weighted heavily | Average watch time, watched full video and retention rate on each post | Compare with your own posts of similar length; the drop-off point shows which moment lost people |
| Skips | Listed among interactions by the Help Center and the Creator Academy explainer | The opening of the retention curve | A steep early fall means viewers decided fast, which makes the opening the first thing to test |
| Likes, comments, shares and saves | Help Center and Creator Academy explainer | Likes, comments and shares in Overview and on each post | Judge them relative to views and against your recent posts, not as raw totals |
| Follows, including follow-backs | Newsroom post and Help Center | New followers, and Your top posts ranked by new followers gained | Shows which posts turned viewers into an audience rather than one-off views |
| Captions, sounds and hashtags | Newsroom post (video information) and Help Center (content information) | Traffic source and Search queries in Overview | Tells you whether people arrived from For You, search or elsewhere, and which terms found you |
| Views, and the country where the video was published | Help Center (content information) and the Creator Academy rule on regional creators | Post views and viewer locations | If viewer locations don't match the audience you want, look there before anything else |
| Language, location, time zone and day, device type | Help Center (user information); the Newsroom post says device and account settings get lower weight | Viewer insights and most active times in the Viewers tab | Context for who sees you, not a lever you control |
| Variety rules: no duplicates, no repeats, spacing by sound and creator | Newsroom post | New and returning viewers in the Viewers tab | Re-uploading a near-copy of your own video doesn't earn a second chance, because duplicated content isn't recommended |
| For You eligibility | Community Guidelines eligibility standards and enforcement section | Each post's analytics, which TikTok says show when a video was made ineligible | Check this first when one post stalls, before reading any other number |
The Newsroom post, How TikTok recommends videos #ForYou, also explains why the weights differ. It contrasts a strong indicator of interest, such as a viewer finishing a longer video, with a weak one, such as viewer and creator being in the same country, and says signals are weighted by their value to the viewer. The Help Center goes further and says the importance of each factor can change over time.
What changes by feed: For You, Following, LIVE and search
TikTok runs several recommender systems, and the Help Center gives each its own mix of inputs. Those differences explain a lot of the contradictory advice online.
- For You: likes, shares, comments, full watches and skips, plus follow-backs; sounds, hashtags, views and country of publication; and viewer settings such as language and time zone. Interactions are generally weighted most for most users.
- Following: much the same content and user information, with visits to the profiles a viewer follows counted as an interaction.
- LIVE: Gifts sent, LIVE views, likes and the creator's follower count all appear among the inputs, alongside watch time.
- Search: content information, including how well a video matches the search term, is generally weighted most, ahead of interactions.
- Account recommendations: follower and view counts, account location and mutual connections play a part.
That split is where the follower-count argument comes from. The 2020 Newsroom post says neither follower count nor a history of high-performing videos is a direct factor in the For You recommendation system, while noting that bigger accounts naturally collect more views from their followers. The current Help Center page doesn't list follower count for For You either, but it does list follower numbers for LIVE and for account recommendations. Both statements can stand, because they describe different systems.
For You eligibility: the filter before the ranking
Ranking only applies to content that has passed TikTok's safety checks. The For You feed eligibility standards, released in August 2026 and in force since late September 2026, say some content stays on TikTok but isn't recommended in the For You feed because it may not suit a broad audience. Each section of the Community Guidelines marks which content falls into that category.
For creators, these are the items most likely to matter, taken from the current guidelines and TikTok's recommendation pages:
- Reused or unoriginal material posted without creative edits, such as clips carrying someone else's watermark, and low-quality or minimally edited content, under the integrity and authenticity rules.
- Engagement bait, which the guidelines describe as tricking people into engaging: promising likes in return for likes, offering fake rewards for gifts or follows, or making misleading claims to inflate views.
- Significant adult body exposure and sexually suggestive behaviour involving adults, which the guidelines on sensitive and mature themes treat as age-restricted as well as ineligible.
- Commercial content that should carry the content disclosure setting but doesn't, which TikTok's commercial disclosure rules say may have its visibility reduced; see our walkthrough of the TikTok paid partnership setting for where it lives.
- Content created by users under 16, which the Creator Academy explainer says is not eligible for recommendation.
- Videos that have just been uploaded or are still under review, which the Newsroom post says may be ineligible while that lasts.
The standards also make three points worth remembering: few views don't by themselves mean a rule was broken; a video that misses the For You feed remains reachable via search and your profile; and TikTok's analytics report when a video was made ineligible. When a whole account rather than a single video is affected, that is an account status question, and account-level limits, warning strikes and appeals are covered in our guide to TikTok shadowbans.
One more rule shapes reach in some niches. TikTok says it tries not to recommend certain themes over and over, naming extreme fitness or dieting, sexual suggestiveness, sadness and overgeneralised mental health advice. A single video on one of those themes can be eligible; a viewer's feed filling up with them is what the system is built to interrupt.
Claims TikTok hasn't published
Plenty of algorithm advice goes further than TikTok's own documents. That doesn't make it wrong, but it does make it unverified, and a creator business shouldn't be planned around it. Check any rule you hear against this table.
| Common claim | What TikTok's pages actually say |
|---|---|
| Every video is shown to a small test batch, then pushed wider in stages | TikTok describes candidate selection, prediction and ranking, but none of the pages cited here gives batch sizes, expansion stages or time windows |
| Follower count decides For You reach | The Newsroom post says it isn't a direct For You factor; follower numbers appear only in the LIVE and account recommendation lists |
| The algorithm rewards one best posting time | The viewer's time zone and day are listed as user information, which ranks below interactions for most users; posting time itself isn't named as a factor |
| Videos past a certain length get pushed further | The recommendation pages don't say so; the length rule creators remember belongs to the rewards program, and our Creator Rewards eligibility checklist sets it out |
| Hashtags are the main lever | Hashtags are one piece of content information; interactions generally weigh more in For You, and in search the heavier category is content information such as query match |
| A verified badge boosts reach | Verification appears in none of the factor lists on the pages cited here; see how to get verified on TikTok for what the badge does |
| The algorithm works the same in every country | US recommendations now sit with a separate joint venture, and EU viewers can switch personalisation off, as the next section explains |
Region differences: the US joint venture and EU feeds
Most of TikTok's explanations are global, but the product isn't run identically everywhere. US accounts sit with a different operator, TikTok USDS Joint Venture LLC, a majority American-owned company. Its launch announcement lists securing US user data, apps and the algorithm as part of its mandate and gives it decision-making power over US content moderation and trust and safety rules. Reuters reported TikTok saying the venture would retrain, test and update the recommendation algorithm on US user data. The global pages quoted in this guide may therefore not describe US recommendation and moderation exactly, and US creators should treat US-specific notices in the app as the current rule.
In the EU and Kazakhstan, the Help Center says viewers can turn off personalised feeds. The For You feed then becomes a Popular feed serving content that is popular in the viewer's region and internationally, so some EU viewers reach your videos without any personal signals involved. Separately, the sensitive and mature themes section says TikTok may apply those rules differently depending on the region, to respect local norms, so a video on a borderline theme won't necessarily be treated the same way in every market.
A weekly review sheet
Run this once a week on posts that are old enough for their numbers to have settled. Compare each post with your own recent posts of a similar length and format, never with someone else's account.
| Question | Where to look in TikTok Studio | What to write down |
|---|---|---|
| Was any post made ineligible for recommendation? | Each post's analytics and your notifications | The post, the notice wording and whether you appealed |
| Where did the views come from? | Traffic source on each post | The For You share against profile, Following and search |
| Did people stay? | Average watch time, watched full video, retention rate | The moment the retention curve drops and what was on screen then |
| Did people react? | Likes, comments and shares on each post | Each figure against your recent average for similar posts |
| Did the post build an audience? | New followers, and Your top posts sorted by new followers | Which posts brought followers, not just views |
| Who watched? | Viewers tab: viewer insights, new and returning viewers | Whether location and language match the audience you want |
| What else does your audience watch? | Your audience also follows and Your audience also watches | Formats and topics worth testing next |
| What will you change next week? | Your own test log | One variable to change, and the metric that will tell you if it worked |
Change one thing at a time and log it, or you won't know what moved the result. Our TikTok content testing framework sets out how to structure those tests, and the TikTok hook test log is built for the opening-seconds problem the retention curve exposes.
Limitations of this guide
TikTok publishes the categories of signals it uses, not the weights, thresholds or model details, and it says the importance of each factor can shift. Everything here reflects TikTok's public documentation on 1 October 2026, including a Newsroom post from 2020 that may not match today's system in every detail. The US system now runs separately and may drift from the global description. Use the signal table to read your own data, not as a promise that moving any one metric will change your reach.