TikTok does not rank you. It ranks each video, one at a time, against the one signal it trusts most: did people watch to the end, and did they watch again. Everything else about the algorithm follows from that. If you understand two ideas, you understand the system. First, completion and rewatch are the currency. Second, TikTok is a search engine, so a strong video keeps earning long after the feed moves on.
This is the platform-level view that sits under how social media algorithms work in 2026. Here is how TikTok specifically decides what to push.
How does the TikTok For You algorithm decide what to show?
Every new video starts the same way, regardless of who posted it. TikTok shows it to a small cold-start seed batch, a few hundred or a few thousand viewers chosen to be a representative test. Then it watches what that batch does in a short early window.
If the video clears an engagement-velocity gate in that window, it scales out in expanding rings: a larger audience, then a larger one again, each ring a fresh test. If it fails the gate, the push stops and the video settles. That is the whole pipeline. Seed, gate, then rings of expansion.
The part most creators get wrong is what the gate measures. It is not raw views and it is not likes. It is the speed and quality of attention in that first window, and attention is measured mostly in watch time.
Does the TikTok algorithm favor completion rate?
Yes, and not as a minor factor. Watch time and completion percentage are the first-class terms in how a video scores.
A leaked internal document described as "Algo 101," reported by the New York Times around 2021, showed the recommendation score as a weighted sum of predicted actions. In shape it looked like this:
score = Plike × Vlike + Pcomment × Vcomment + Eplaytime × Vplaytime + Pplay × Vplay
Treat the exact coefficients as illustrative rather than literal. The point that holds is the structure: playtime appears as its own term in the score, alongside likes and comments, not as a tie-breaker applied after the fact. TikTok predicts how long you will watch and rewards the video that keeps you there.
That reorders everything. The direction of the signal priority is clear, even if the precise ranking is an estimate. From strongest to weakest, it runs roughly like this.
| Rank | Signal | What it tells TikTok |
|---|---|---|
| 1 | Watch time | Total seconds held. The core currency. |
| 2 | Completion percentage | Did they reach the end, not just start. |
| 3 | Rewatches and loops | Worth seeing more than once. |
| 4 | Shares and saves | Worth sending privately or keeping. |
| 5 | Comments | Worth responding to. |
| 6 | Likes | The weakest action. Cheap to give. |
| 7 | Follows | Intent to see more from you. |
| 8 | Text info | Caption, sound, hashtags. Context, not proof. |
A like costs a viewer nothing, which is why it sits near the bottom. Finishing a video, then watching it again, costs real attention, which is why it sits at the top. Design for the top of this list and the rest tends to follow.
Do followers matter on TikTok?
Not for reach. Follower count is not a direct ranking factor, and neither is your track record. TikTok tests every video fresh against a new seed batch, which is why a brand-new account can land a video on millions of feeds and a large account can post something that barely moves.
This is the sharpest difference between TikTok and the networks built on an audience graph. On TikTok, your last ten flops do not weigh down your next upload, and your best-ever video does not float the next one for free. Each post earns its distribution on its own merits, in its own early window.
Followers still matter for the business around the video: a place for people who liked one clip to find the rest, and a base that reliably sees you in their Following feed. They are an outcome of good videos, not an input to reach.
Is TikTok a search engine?
This is the half of TikTok most brands ignore, and it is where the compounding happens. An estimated 49% of US users and around 65% of Gen Z now use TikTok as a search engine, turning to it the way an earlier generation turned to Google.
To serve those searches, TikTok indexes the text of your video. Every word of the caption, the on-screen text, and the auto-transcribed audio becomes something a video can be found by. A keyword-rich video keeps surfacing in search results for weeks, long after the For You push has faded, which is the opposite of the pure feed networks where a post's life ends when the push ends.
The practical move is to write and speak the words people actually search. Put the keyword in the caption, say it out loud in the first few seconds so the transcript catches it, and show it as on-screen text in the opening frame. One video can then earn twice: a burst from the For You feed on day one, and a long tail from search for months after.
What a strong TikTok video actually does
The levers follow directly from the scoring. If watch time and completion win, build the video to earn them.
- A hook in under two seconds, because the early window decides everything.
- A seamless loop, so the end flows back into the start and completion ticks past 100%.
- A trending native sound, which the system reads as a timely, in-format signal.
- 9:16 full-frame video, with no letterboxing or borrowed aspect ratios.
- On-screen captions, which hold attention and also feed the text index.
- A tight length with no dead air. Every second that drags is a second someone swipes.
- A reason to share, save, and comment, the mid-weight signals that confirm the watch.
- Keywords in three places: the caption, the spoken words, and the first frame.
What gets a TikTok video shadowbanned or made ineligible?
Some mistakes do not just underperform. They take a video out of the For You running before it gets a fair test. These are the confirmed ones.
| Landmine | What happens |
|---|---|
| Foreign watermark | A logo from another app (Instagram or CapCut) makes the video ineligible for the For You feed. The single most common cross-posting mistake. |
| Unedited reposts and slideshows | Content posted without edits is downranked as low-effort. |
| On-screen QR codes | Treated as off-platform redirection and suppressed. |
| Unlabeled AI content | AI-generated content that is not disclosed gets downranked. |
| Engagement bait | "Comment YES" and similar prompts are penalized rather than rewarded. |
The watermark one is worth repeating because it quietly kills so many cross-posted videos. A clip that performed well on another app can go nowhere on TikTok purely because it still carries that app's mark. Export clean, and post native.
How TikTok compares to the others
TikTok is not alone in valuing attention, but it weighs the signals differently. Unlike Instagram, which leans on sends per reach as its headline currency (see how the Instagram algorithm works), TikTok puts raw watch time and completion first. YouTube ranks on a similar watch-time logic, rewarding videos that hold a session and keep viewers on the platform (see how the YouTube algorithm works). The through-line across all three: attention held is attention rewarded. TikTok just tests it faster and forgets your history harder.
Sources
The pipeline and the public signals are described in TikTok's own material, chiefly the TikTok Newsroom post "How TikTok recommends videos #ForYou" and TikTok's For You feed eligibility standards, which name the watermark, repost, and bait penalties. The scoring formula comes from the New York Times report on the internal "Algo 101" document, also covered by Popular Science and pxlnv. The search-usage figures are third-party estimates and are presented here as estimates.
TikTok ranks each video fresh on how long people watch and how often they rewatch, then keeps surfacing it in search for the keywords it heard and read. Earn the first second, hold to the end, and say the words people look for.
Keep up with the algorithm so you don't have to
Protegee runs a standing research function that tracks how every platform ranks content, from X's open-sourced code to the signals TikTok, Instagram, Facebook, and YouTube publish, and builds those learnings into every post it generates. The rule never changes: earn the first second, hold attention to the end, make it worth privately sending.
Sign up