How the For You feed actually works
Most of what gets said about TikTok's algorithm is invented. What is officially known is four steps, and knowing those four is enough to reject 90% of the myths yourself.
- Retrieval — the system gathers a pool of videos that passed initial moderation and are eligible for recommendation
- Prediction — for each video it estimates how likely you are to interact with it
- Ranking — it sorts them by that prediction score, highest first
- Diversity — if consecutive videos are too similar to each other (the same sound, for instance), some are swapped out
[OFFICIAL] This sequence is documented by TikTok itself.
[OFFICIAL] — the single most important sentence: TikTok states plainly that follower count and whether the account had previous high-performing videos are not direct factors in the recommendation system.
The practical consequence: a video from an account with 0 followers and one from an account with 500k followers land in the same pool. Your problem is not "I have few followers" — it is the reaction the video gets in its first small audience.
[PRACTICE] The "staircase" model creators observe. TikTok confirms this nowhere, but the behaviour of thousands of accounts produces this picture: a video is first shown to a small group; if the numbers there are good it moves to a larger group, and so on.
In your analytics it looks like this:
- A video stuck at 200-400 views — it did not clear the first step
- A video that jumps hours later — it cleared the step late (which is normal, and kills the "first hour" myth)
- A video that climbs again on day three — later steps can take days
Treat this as a working mental model, not a fact. Its value: the answer to "why did my video stop at 300" is not "shadowban" but "it did not clear the first step" — and what clears the first step is retention.
| What you think | What actually happens | What to do |
|---|---|---|
| "I get no views because I have few followers" | Follower count is not a direct factor — the video failed in its first group | Fix the first 3 seconds and the retention curve |
| "The algorithm is punishing me" | In most cases it is quality variance between videos, not a penalty | Line up your last 10 videos and sort them by retention |
| "Using the same sound will spread it further" | The diversity filter avoids showing same-sound videos back to back | Choose a sound for the hook, not as a distribution ticket |
| "Deleting the video will clean up my account" | Deleting does not undo past signals, and re-uploading can read as spam | Delete only when there is an actual guideline violation |
What the labels in this branch mean:
- [OFFICIAL] — written in TikTok's or the tool's own documentation; the source is listed under the topic
- [PRACTICE] — widely tested in the creator community, no official confirmation, you must verify it on your own account
- [MYTH] — widespread, but unevidenced or actively harmful
Keeping the distinction matters: breaking an official rule costs you the account; misapplying a practice tactic only costs you time.
🛠 Practice task
Open TikTok Studio → Analytics → Content and copy your last 10 videos into the table from the example above: views, the share still watching at 3 seconds, the share who watched it fully, and shares.
It is done when you can look at the table and write one sentence: "My weakest point is ___" — backed by the numbers. If every weak video sits below 50% at 3 seconds, the answer is the hook; if 3 seconds looks fine but full watches are rare, the answer is the middle of the video.
📚 Sources and documentation
- How TikTok recommends videos #ForYouofficialnewsroom.tiktok.com
The official explanation: which signals count, and that follower count is not one.
- Introduction to the recommendation system — Transparency Centerofficialtiktok.com
The official description of the retrieval → prediction → ranking → diversity steps.
- Understanding TikTok's Recommendation System — Creator Academyofficialtiktok.com
The same material in creator-facing language, with examples.