TikTok's For You feed is the most discussed recommendation system on the internet and the most widely misunderstood. The mechanism itself is not mysterious: videos are distributed in steps, tested at each step, and either promoted or stopped. What matters is knowing which measurements decide that.
Stepped distribution
A newly published video is shown to a small test audience — typically a few hundred people, drawn from users whose interests match the content and, to a lesser degree, from your followers.
TikTok measures how that group responds. If the rates clear a threshold, the video moves to a larger audience, perhaps a few thousand. The same measurement runs again. Clear it again and the audience becomes tens of thousands, then hundreds of thousands. At any step where the rates fall, distribution stops.
This is why follower count matters so little on TikTok relative to other platforms. A video from an account with no followers enters the same process as one from an account with a million. It also explains why a creator can have one video reach five million and the next reach two thousand — each video is judged independently.
Completion rate dominates
Among the signals measured, how much of the video people watch outweighs everything else. A video watched to the end, or restarted, tells TikTok the content held attention — which is precisely what the platform is optimising for.
The arithmetic strongly favours short videos. A nine-second clip needs nine seconds of attention to complete; a sixty-second one needs sixty. The short video will report a far higher completion rate almost regardless of content quality, and rewatches push it above one hundred percent.
This produces the counter-intuitive result that a video with fewer views can be performing better than one with more. If a video has 3,000 views and eighty percent completion, it is still being pushed. One with 30,000 views and twelve percent completion has already been shelved.
The signal hierarchy below completion
After completion, the remaining signals rank roughly by the effort they require.
Shares carry the most weight because sending something to another person puts the sender's own standing behind it. Shares also produce a second-order effect: the recipient opens the video in the app, generating a new view and possibly new engagement.
Comments are next, and they have a mechanism specific to TikTok. People read comments while the video loops, which directly raises watch time and rewatch count — so a busy comment section feeds the metric that matters most.
Likes are the base signal. Light, common, and necessary but not sufficient on their own.
Follows from a video are a strong but rarer signal, indicating someone valued the content enough to want more of it.
What the first second does
Since completion is measured from the start, the opening decides what the rest of the video is measured against. Most of the audience that leaves a TikTok leaves almost immediately.
The practical implication is that setup is expensive. A video that spends three seconds establishing context before the interesting part begins has already lost a share of its test audience, and the completion rate reflects that loss permanently.
Opening mid-action, leading with the result, or putting a short line of on-screen text that states the subject all work for the same reason: they give a scrolling viewer something immediately, before the decision to swipe is made.
Sound, text and accessibility
A substantial share of TikTok viewing happens with sound off, particularly in the first moments. Videos that require audio to make sense lose those viewers before sound is ever turned on.
Captions solve this and cost nothing. Audio choice serves a separate function: a trending sound places the video on that sound's page, where users are browsing for more of the same, and it can also cue TikTok about the content type.
Text on screen has one constraint worth remembering: the interface covers portions of the frame. Text behind the caption area, the username or the action buttons is partly hidden, and on a format decided in the first second that is a costly error.
What this means for using services
The clearest implication is that timing beats quantity. The test audience evaluation happens in the first hours, so engagement arriving then lands while the decision is being made. Engagement arriving on day three lifts counters and affects nothing.
The second is that ratios matter, because they are all visible. A video with 200,000 views and 300 likes sits far outside the three to fifteen percent band that genuine videos occupy, and that inconsistency reads poorly to viewers and in the measurement itself. Planning views, likes and shares together inside their natural bands is what keeps the picture coherent.
The third is the honest limit: no service produces completion rate. That number comes from the video, and it is the number that decides whether distribution continues.
How TikTok decides who is in the test audience
The composition of that first few hundred viewers is not random, and it explains a lot about why similar videos perform differently.
TikTok builds an understanding of what an account posts about, drawn from the audio used, on-screen text, the caption, hashtags and what the video actually contains. It then seeds the test audience with users whose viewing history suggests interest in that subject.
The practical implication is that topic consistency helps. An account posting about one clear subject accumulates a well-defined profile, and its test audiences get better targeted over time. An account posting about five unrelated things gives the system a muddled signal, and each video starts in front of a less relevant group — which depresses completion through no fault of the video.
This is also why a first video on a brand new account sometimes performs surprisingly well: with no history to go on, TikTok casts a broad net, and a broad net occasionally finds the right audience immediately.
What does not work
Several widely circulated tactics are worth dismissing directly, because following them costs effort for nothing.
Deleting underperforming videos does not improve the account. Distribution is per-video, and removing a video simply removes whatever it was still generating.
Posting at a universally optimal time is not a thing. What matters is when your own audience is active, which the analytics report, and that varies enormously by account.
Stuffing captions with unrelated hashtags confuses the topic signal rather than broadening reach. A few genuinely relevant tags help TikTok place the video; a wall of popular ones makes the subject harder to identify.
Frequently asked questions
Does follower count affect TikTok reach?
Very little. Videos enter a stepped distribution process and are judged on their own performance, which is why a small account can reach millions and a large one can stall.
Which signal matters most?
Completion rate — how much of the video people watch. It outweighs likes, comments and shares, and it is why short, loop-friendly videos systematically outperform long ones.
Why can a video with fewer views be performing better?
Because distribution depends on completion, not view count. A video with 3,000 views and high completion is still being pushed; one with 30,000 and low completion has been shelved.
What like-to-view ratio is normal?
Roughly three to fifteen percent on genuine videos. A video with 200,000 views and 300 likes sits well outside that band and reads as inconsistent.