The Thing Most People Get Wrong About TikTok's Algorithm
Most people assume TikTok shows you videos from accounts you follow, topped up with some popular content when things go quiet. That's Instagram's model, roughly. TikTok works almost completely backwards from that.
Follower count is largely irrelevant to reach on TikTok. A video from an account with 200 followers can outperform one from an account with 2 million if the engagement signals are right. And a brand-new account posting its first video can hit a million views in 24 hours if the content lands. I've watched this happen to accounts I track. It still feels strange every time.
The reason is that TikTok's system doesn't distribute content to your followers first. It tests content on a small group of users who don't follow you, evaluates their response, and decides based on that data whether to push it wider. Your follower list is almost an afterthought.
What TikTok Actually Measures to Decide What You See
The algorithm cares about a handful of signals more than anything else. Watch time sits at the top. Not just how long someone watched, but what percentage of the video they completed. A 90% completion rate on a 60-second video is a stronger signal than 50% completion on a 10-second video. TikTok's internal research reportedly found that completion rate is the single best predictor of whether a video gets promoted.
Replays are weighted heavily. A video that people watch twice is a very strong signal. When you replay something without consciously deciding to - just instinctively hitting replay because the three seconds of content was that good - the algorithm interprets that as high value. A lot of TikTok's most viral content is engineered exactly for this: short enough that replay happens accidentally.
Shares matter more than likes. Sending a video to someone else is a much stronger intention signal than double-tapping. Saves are significant too, because saving implies the user valued the content enough to return to it. Comments move the needle - not just because volume signals engagement, but because the algorithm apparently weighs videos that generate emotional enough reactions to make someone type.
What it doesn't weight much: follower count, account age, previous video performance (each video is evaluated fresh), and the number of hashtags you use. That last one surprises creators who spend ten minutes loading captions with tags. TikTok's own guidance has walked back how important hashtags are multiple times. They help categorisation. They don't amplify reach the way many tutorials imply.
How TikTok Learns Your Taste So Fast
Open a brand new TikTok account and scroll for forty minutes. By the end of that session, the content will already feel noticeably more personalised than when you started. That's not an accident.
TikTok doesn't wait for you to like or follow anything. It infers preference from behaviour. If you watch three videos about home renovation and stop at each one, it quietly notes that. If you scroll past financial advice videos in under two seconds consistently, it stops serving you those. Every pause, every replay, every exit point in a video tells the model something about what you respond to.
The device-level signals also matter more than most platforms admit. TikTok knows your device type, your location, the time of day you're active, and what content categories were popular on your device in the past. If you've ever switched phones and noticed your TikTok For You Page felt reset and generic for a day or two before getting personalised again - that's exactly what happened. The account retained your explicit history, but the device-level learning had to rebuild.
The Testing Pool System - How Videos Actually Get Distributed
This is the part that makes TikTok structurally different from every other platform.
When you post a video, TikTok shows it to an initial pool of a few hundred to a few thousand users. These aren't your followers - they're a sample drawn based on the predicted audience for your content type, time zone, and other factors. TikTok then measures the engagement rate within that pool, usually over the first few hours.
If the engagement rate passes a threshold, the video gets promoted to a larger pool - maybe ten times the size of the first. If it passes again, larger still. If at any stage the engagement rate drops below what's expected for that audience size, the promotion stops. The video doesn't disappear, but it stops spreading.
This is why TikTok videos can go viral days or weeks after posting. Sometimes a video stagnates after the second pool, then gets picked up again when TikTok re-evaluates it as part of a trending topic or when something similar goes viral and the algorithm looks for related content to promote alongside it. I've seen videos spike at the 10-day mark with no changes made to them. The system never fully abandons content - it just waits for the right moment.
What Actively Hurts Your Reach
Negative signals exist and they're real, even though TikTok rarely acknowledges them explicitly. A high "not interested" tap rate on a video - where people actively tell TikTok they don't want to see it - is apparently treated as a negative quality signal that can affect broader distribution. Not just for that specific viewer, but for the video's overall health score.
Videos that get reported at higher-than-average rates slow down in distribution, sometimes almost immediately. This doesn't mean they get removed, but the testing pool stops expanding. TikTok's trust-and-safety systems communicate with the recommendation system, so a video that's generating moderation reports while also generating views is getting mixed signals that typically result in stalled growth.
Re-uploads hurt. Posting the same video twice, or posting content that TikTok's system detects as a low-quality copy of content that already exists on the platform, gets suppressed. The content fingerprinting is reasonably sophisticated. Mirroring a video horizontally to fool the detection doesn't work as reliably as it used to.
The Consequence You Actually Feel: Content Disappears Fast
Because TikTok's algorithm serves content in time-sensitive testing windows, something you see on your For You Page today genuinely may never appear again. The algorithm moved on. The video is still there, but it's not being served to new audiences anymore. If you wanted to save it, that window has probably already closed.
This is different from YouTube or Instagram, where you can reasonably expect to find something again via search or a creator's profile. TikTok's search is improving but it's still not great for finding specific older content. And if a creator's account gets deleted or suspended, that content is gone entirely - there's no public archive.
Saving content when you see it is just the practical response to this system. MyVideoCity's TikTok downloader is built for exactly this: paste the link before the video stops circulating. What the algorithm shows you is finite. What you save is yours permanently.
For more on the AI systems shaping what you see online, see our piece on how AI is changing video. And if you're trying to save content from other platforms, guides for Instagram, X (Twitter), and Facebook are all here.