TikTok Recommendations: Documented Signals and Testable Hypotheses

TikTok describes a personalized recommendation system, not a fixed ladder that every video climbs. This guide separates platform documentation from creative hypotheses. Editorial review: 22 September 2026.

What TikTok documents

TikTok lists viewer interactions (including watching, skipping, likes, comments, and shares), content information (such as sounds and hashtags), and user information among For You recommendation inputs. It says interactions, including time spent watching, generally carry more weight for most users. It also describes recommendation eligibility limits. This is a description of input categories, not a published scoring formula. Source: TikTok Help Center.

TikTok’s earlier explanation describes ranking for an individual viewer and gives finishing a longer video as an example of a stronger interest signal than sharing a country with the creator. Treat that as explanatory context, not a newly announced 2026 algorithm change. Source: TikTok Newsroom.

Rules we cannot substantiate

We do not have evidence for universal 200-view test pools, expansion at 1,000 or 10,000 views, a decisive first 30 minutes, a two-hour expansion deadline, or a 24-hour ceiling. The linked documentation does not establish those thresholds. Earlier versions of this article presented them too confidently.

We also cannot substantiate a fixed ranking of DM, Story, and external shares; a guaranteed new-creator boost; a universal optimal hashtag count; or a claim that a looping edit multiplies watch time by a particular amount. Do not use these as platform rules or spend decisions.

What we have observed

No TikTok post-level dataset accompanies this article. Our Instagram account experiment is a separate platform and a limited first-party account report. Its results cannot validate TikTok ranking weights or thresholds.

Practical hypotheses to test

Try opening with the result, reducing unnecessary setup, or making the promised payoff clearer. These are creative recommendations to test with your audience, not guarantees of distribution. Choose relevant captions and hashtags that describe the content; avoid treating a tag as a reach shortcut.

Keep a log with the original post URL, publication timestamp and timezone, duration, opening hook, format, paid promotion, and reporting time. Compare views, average watch time, completion, shares, and follows at consistent post ages where analytics provides them. A 24-hour and seven-day check is a measurement choice, not an algorithm deadline.

Include low-performing posts. Report the number of posts, the median as well as the standout, and factors you could not control: topic, audience, seasonality, and changing platform behavior. Change one major creative variable at a time; small organic samples are exploratory rather than causal experiments.

Turn findings into the next brief

Write the next creative brief around a testable idea: audience, opening, promise, payoff, and the metric you intend to improve. Keep business outcomes separate from reach. If signups or purchases were not measured, say so. Start a content research project in AdAnt.