What we actually know about the TikTok algorithm in 2026 (and what's still folklore)

A 19-second video at 92% completion will outperform a 47-second video at 38% completion almost every time, even if the longer one has more likes. That's not a hot take. That's the bar the FYP is grading you on, and most "TikTok hacks" content on TikTok is a distraction from it.

Let's separate what TikTok has actually published from what creators repeat to each other in Discord servers.

What's documented

TikTok's Creator Academy and Transparency Center have been unusually candid about the ranking inputs. The hierarchy hasn't changed much since the 2020 leak that The New York Times wrote up: user interactions (watch time, completion, rewatches, shares, comments, saves) outrank video information (sound, captions, hashtags), which outrank device/account settings (language, location).

What has changed is the threshold. Sources tracking the FYP closely report that the completion rate needed to graduate from the initial test batch (~200–500 users) to broader distribution has crept up from roughly 50% in 2024 to roughly 70% in 2026. Shares and saves now carry visibly more weight than likes. Comment length matters more than comment count.

Rewatches are the under-discussed signal. A 12-second clip looped 1.4 times averages a "watch percentage" north of 100, and the algorithm reads that as a strong endorsement. This is why short, dense, loop-friendly cuts keep beating slower long-form on the FYP — even after TikTok started pushing creators toward 60+ second videos for monetization eligibility.

What's still folklore

Emoji shadowbans. The "don't use a 🍑 in your caption" advice has been around since 2021. There is zero evidence in any TikTok policy doc, transparency report, or moderator interview that specific emojis trigger reach suppression. What does happen: content the model thinks might violate guidelines gets reach-limited, and TikTok doesn't call this a shadowban — they call it "reduced eligibility for the For You feed." The mechanism is real. The emoji-as-trigger story is not.

Posting at 7:43 PM Tuesday. The most-shared screenshots of "best posting times" are screenshots of someone else's audience. TikTok's distribution model is velocity-based: the test batch lands within minutes of publication, and the first 30–90 minutes of engagement decide whether the video gets a second push. If your audience is mostly online at 11 PM, posting at "global peak" 9 AM ET hurts you. The "golden hour" thing is folklore wearing a confidence interval.

Length manipulation. "Make every video 27 seconds because the algorithm likes it." No. The algorithm likes high completion. 27 seconds is a heuristic for "long enough to feel substantial, short enough that most viewers finish." If your edit holds at 53 seconds, post 53.

The follower count penalty. New creators love this one because it's flattering: "TikTok suppresses small accounts." It's the inverse — TikTok is one of the few platforms that under-weights follower count relative to per-video performance. This is why a 200-follower account can hit 4M views and a 2M-follower account can post a flop.

What actually changed in 2025–2026

Three things worth knowing:

Creator Rewards Program update (December 2025). Starting December 4, TikTok began counting search views (with at least 30 seconds of watch time) as qualified views in the monetization pool, alongside dashboard and rewards-structure updates. Search views being monetized is the bigger story. It signals that TikTok is taking its search-engine role seriously as a revenue line, not just a Gen Z behavior pattern.

Community Guidelines rewrite (September 13, 2025). TikTok removed the word "trustworthy" from how it describes its content-moderation goals and tightened rules around AI-generated endorsements and creator responsibility for synthetic content. If you're using AI tools to generate footage of real people, the bar moved.

Follower-first testing. TikTok has been testing serving videos to existing followers before strangers. This is a quiet reversal of the platform's foundational interest-graph promise. If it ships broadly, the practical implication is that audience-building stops being optional for sustained reach — which would make TikTok feel a lot more like Instagram.

The contrarian opinion

Most "TikTok strategy" content overweights the algorithm and underweights the hook. The algorithm is a filter, not a generator. It can decide whether 200 people or 2 million people see your first three seconds, but it cannot make the first three seconds good. I'd rather post a video with a great hook at 6 AM than a mediocre video at "peak time" with the right hashtags and a 27-second runtime.

The folklore industry exists because hook-writing is hard and adjusting your posting schedule is easy. Easy advice scales. That's why it spreads. It is not the same as advice that works.

Practical implications

• Edit for completion, not duration. Cut anything that doesn't move the viewer toward the next beat. If completion drops at the 8-second mark, that's where the dead weight is.
• Watch shares and saves. Likes are vanity. Saves mean someone wanted to find this again. Shares mean someone burned social capital recommending it. The algorithm reads both as much stronger signals than a like.
• Stop optimizing for emoji-superstition. Optimize for whether the first frame stops a thumb.
• If you're chasing Creator Rewards, lean into search-discoverable topics. Search views now count.

One takeaway

Open your last 10 videos in TikTok Analytics and sort by average watch time as a percentage of video length, not by views. The video at the top of that list is the one the algorithm liked. Make more of those.