TikTok's own explainer names signals, not follower count

The test

On 18 June 2020, TikTok's newsroom published a post titled "How TikTok recommends videos #ForYou", its first detailed account of how the For You feed selects what to show. The post names three groups of signals: user interactions such as likes, shares, follows and comments; video information such as captions, sounds and hashtags; and device and account settings such as language, country and device type, carrying comparatively less weight. It states plainly that neither follower count nor prior high-performing videos are direct factors. Six years later, TikTok maintains a living version of the same explanation on its transparency page, "Introduction to the TikTok recommendation system", retrieved 16 September 2026, which describes the same idea procedurally: signals feed a prediction score, scores are ranked, and ranked videos pass a similarity check before reaching a feed.

What the evidence says

Both documents are TikTok's own account of its own system, not an independent audit. The 2020 post is a one-time announcement; the transparency page is current documentation, updated as the company chooses, and it adds detail the 2020 post lacked. It names five processing steps, from selecting eligible videos through prediction, ranking, a similarity check and a final layer of recommendation rules, and states that content from accounts under 16 is ineligible for recommendation. Both documents agree on the core 2020 claim: TikTok says the system does not treat follower count as a direct ranking input. Neither is a technical audit of the ranking model, and neither discloses training data or the actual weighting applied signal by signal.

The sample and the variance

Both sources describe TikTok globally, not a specific market, cohort or period, and both are self-description rather than a measured outcome. The transparency page calls its signal list non-exhaustive and says weighting can change over time, so neither document supports a claim about how any single video will perform. A platform's own account of what a system reportedly does is evidence of stated design intent, not of what any particular account will experience.

What to try next

This is an editorial reading of the two documents, not a testing method TikTok prescribes. A publisher comparing videos could track how closely their captions, sounds and hashtags describe the content, since the sources name those as direct video-information signals, separating that from account-level habits like posting time and language settings, which carry comparatively less weight. Distinguishing a stated signal from an assumed one beats chasing a single viral post.

• Which claimed ranking signal in your own results can you trace to something TikTok's documents actually name?
• Has TikTok's public description of a signal changed since your last review of it?
• Are you crediting follower count or past performance when the documents say neither is a direct factor?

TikTok's explainers describe intent and mechanism as the company presents them; they are not proof of outcome for any account, and reading the 2020 announcement against the current transparency page is a reminder that a platform's own description of its algorithm is itself a living document.