How the TikTok Algorithm Works: Official vs. Guesswork

Search for how the TikTok algorithm works and you will find confident posts with exact percentages, secret posting times and "the first hour decides everything." Almost none of it comes from TikTok. This article separates what the platform has said from what creators have guessed, then gives you a small testing routine you can run on your own videos.

Platform behavior changes, so treat everything here as a compass, not a rulebook.

What TikTok has officially said about the For You feed

TikTok published an explainer of its For You feed in 2020. In plain terms, it says recommendations draw on three groups of information:

• User interactions: likes, shares, comments, follows and videos watched through to the end.
• Video information: captions, sounds and hashtags.
• Device and account settings: these carry less weight than the other two, according to the explainer.

That is the whole official list. TikTok has not published weights, a ranking formula, an ideal video length or an ideal number of hashtags. Two practical points follow from it, and both are safe to rely on:

1. What viewers do matters. A video that people watch through, share, comment on or follow from gives the system a reason to show it to more people.
2. What your video says about itself matters. Captions, sounds and hashtags help the system work out what a video is about, so say the topic plainly.

What is creator guesswork, and how to spot it

Creators who post often do notice patterns, and some of those patterns are probably real. But "creator-observed, not confirmed by the platform" is a different category from "TikTok said." Here is how common claims sort out.

Claim	Status
Videos people watch through tend to do better	Watching through is on TikTok's official list. How much it weighs is unpublished.
Rewatches are a strong signal	Creator-observed, not confirmed by the platform
The first seconds decide whether people stay	Creator-observed, and consistent with common sense. Not a stated rule.
Search is an increasingly important way to be found	Creator-observed. Check how search works in your own app.
Old videos can resurface weeks later	Creator-observed
"A share is worth five likes" or any exact multiplier	Guessing. Platforms rarely publish weights.
There is a magic posting time	A hypothesis to test in your own analytics
You must post a set number of times per day	Not confirmed. Pick a rhythm you can keep.

A quick filter: if a claim contains a precise number and no source from TikTok itself, assume it is a guess. Guesses can still be useful as things to test. They are just not facts.

The feed is a matching system, not a judge

The next section is reasoning, not a TikTok statement, and it is a simplified teaching model rather than a diagram TikTok has published.

A recommendation feed has one job: show each person the few videos they are most likely to enjoy right now. It is not grading your video against an absolute standard. It is estimating whether a particular person will want a particular video.

In practice, that suggests a pattern many creators describe. A video is shown to a small group, the system watches what that group does, and reach grows or levels off depending on the response. If that is roughly right, three things matter:

• The opening has to tell the right viewer "this is for you." A smaller right audience beats a bigger wrong one.
• Your topic should be clear in more than one place: spoken words, on-screen text and the caption.
• Judge a video after days, not minutes. Many creators report that videos keep gaining views over time. Treat that as creator-observed.

A simple testing routine: first seconds, completion and shares

You cannot read TikTok's weights, but you can read your own results. This routine tests the three things most worth testing: whether people stay past the first seconds, whether they finish, and whether they share.

1. Pick one topic you can film several ways. For example, "three mistakes new gardeners make."
2. Change one thing at a time. Start with the opening: the first spoken line and the on-screen text in the first second. Keep length, format and topic the same.
3. Make at least three videos per version. One video is noise. A pattern across several is worth noticing.
4. Check at a consistent point. As a rule of thumb, look at the same checkpoint for every video, such as seven days after posting.
5. Read three things. Open each video's analytics (what is available depends on your account, so check the app) and note the retention graph or average watch time, shares compared with views, and follows from the video.
6. Keep what repeats and change one new thing.

Here is a log you can copy.

Test name: [Hook test 1]
Single change: [spoken hook A vs B / on-screen text / first frame]
Held constant: [topic, length, format]
Version A videos: [date, date, date]
Version B videos: [date, date, date]
Checked on: [same day count after posting for all]
Average watch time: A [ ]  B [ ]
Shares per 1,000 views: A [ ]  B [ ]
Follows per video: A [ ]  B [ ]
Pattern across all videos: [what repeated]
Decision: [keep / drop / test again]

How to read a retention graph

• A steep drop in the first seconds usually means the opening did not match what the viewer expected. Fix the first line or first frame, not the middle.
• A cliff at one specific moment means something happened there, such as a slow explanation or an unrelated cut. Watch that second as a viewer would.
• A flat stretch or a bump means people stayed or rewatched. Find what you did there and do more of it.

A question for shares

Shares are an official part of the list, and they are one of the few signals you can design for directly. Before you film, finish this sentence: "This would make someone think of ___." If you cannot name the person, the video may be harder to share. A line like "send this to the friend who keeps buying plants and forgetting to water them" gives the idea a destination.

What not to waste time on

These are common time sinks, and none is backed by anything TikTok has published.

• Hunting for the perfect posting time. Check when your own followers are active in your analytics, post at a few different times and compare.
• Copying a block of hashtags. Hashtags are listed as video information, so use a few that describe the video accurately. There is no published ideal number.
• Engagement pods, bought views and fake comments. They produce activity from accounts that do not care about your content, which can teach the system the wrong thing about who your video suits. This is reasoning rather than a TikTok statement, but it is also against the spirit of the platform's own goals.
• Chasing every trend regardless of fit. A trend that does not suit your topic brings the wrong viewers, who leave quickly.
• Deleting a video after a few hours. Give it several days before you judge it.
• Shadowban theories. A sudden drop is far more often a content-fit, topic or audience issue. If you suspect a real restriction, check the account status or content-eligibility pages in the app. For a step-by-step approach, see why your views dropped and how to diagnose it.

Where to go from here

Testing needs volume, and volume needs a system. If you want to understand how the major feeds work on one page, the opening chapter of The Content Creation Bible covers what each platform has said officially, labels everything else as creator-observed, and includes a TikTok chapter with templates and a 14-day plan.

If your main obstacle is making enough videos to compare, the Content Batch Planner is a simple three-table system (platforms, content calendar and captions) for planning and batch-recording content, so your tests have enough samples.

Two related reads: 40 video hooks you can use today gives you versions of an opening to test, and how Instagram ranks content shows what Instagram has said, in the same official-versus-guesswork format.

Frequently asked questions

Does TikTok prefer long or short videos?

TikTok has not published a preferred length, and its 2020 explainer does not list length as a factor. As a rule of thumb, make the video as short as it can be while still delivering what the opening promised.

How many hashtags should I use on TikTok?

TikTok lists hashtags as part of video information but has not published an ideal number. Many creators start with a few relevant ones, which is typical guidance rather than a rule, then adjust based on their own results.

Am I shadowbanned if my views drop?

A drop is more often explained by content fit, topic or audience than by a hidden penalty. Check the account status or content-eligibility pages in the app for any actual restriction before assuming the worst.

How long should I wait before judging a video?

Pick a consistent checkpoint, such as seven days after posting, and use the same one for every video. That rule of thumb keeps comparisons fair, since many videos keep gaining views for days.