What Does TikTok's Algorithm Actually Measure?
The algorithm operates on two levels simultaneously: it builds a profile for every user based on their behavior, and a profile for every video based on the signals it generates. Its only job is to match these two profiles — finding the right viewer for each video as fast as possible. Every signal you send either widens the distribution circle or narrows it. This article explains how each signal works mechanically, with real numbers. For the comparative weights of each signal, see TikTok algorithm signal priority rankings.
How the Algorithm Reads Your Signals: The Test Mechanism
When you publish a video, it isn't shown to everyone immediately. The algorithm starts with a small test sample — typically 200 to 500 viewers — and measures their behavior during the first 30 to 60 minutes. Based on the signals collected from this sample, it decides:
- Expand distribution to the next wave (5,000 to 10,000 viewers)
- Or stop support and keep the video within its limited initial audience
This process repeats across waves — each wave that succeeds opens a larger one. Read how TikTok distributes videos across waves to understand this mechanism in detail.
1. Completion Rate — The Core Gateway Signal
What the algorithm measures: The percentage of viewers who watch the video all the way through — the strongest signal the system reads, because it reflects genuine interest and cannot be faked.
How it works in practice: The algorithm treats the first 3 seconds as the "entry gate." A viewer who stays past the third second has a high probability of completing the video. Full completion is the green light to advance to the next distribution wave.
Real example with numbers:
- 22-second video — 42% completion rate → jumped to FYP and reached 1.8M views
- 45-second video — 4.5% completion rate due to a long slow intro → stopped at 2,300 views
How to earn this signal: A strong hook in the first 3 seconds is not optional — it's the condition for staying in the distribution cycle. The minimum worth targeting is 30% completion rate. Below that, the algorithm stops support early. See our TikTok hook guide for the formulas that work.
2. Replays — The Exceptional Value Indicator
What the algorithm measures: How many times the same viewer replays the video in a single session. When retention rate exceeds 100%, it means each viewer watched the video more than once.
How it works in practice: The algorithm classifies content with a high replay rate as "content too valuable to watch just once" — and pushes it with unusual force because it keeps users inside the app longer.
Real example with numbers:
A 7-second "visual illusion" video — 145% retention rate (viewers replayed it twice to understand the trick) → 3.2 million views in 24 hours alone.
How to earn this signal: Add a subtle detail that's hard to catch on first watch, information that appears and disappears quickly, or an unexpected ending that makes viewers go back to verify. Very short videos (under 10 seconds) naturally earn replays because the loop restarts automatically.
3. Shares — The Distribution Multiplier
What the algorithm measures: How many times the video is sent to someone else — whether inside TikTok or externally via WhatsApp, Instagram, or Telegram.
How it works in practice: External shares carry higher weight than internal ones for a specific reason: when someone sends your video outside the app, they're bringing a new user to TikTok for free. That's exactly what the platform wants — and what it rewards most generously.
Real example with numbers:
A "secret AI tools" video — 500,000 views with a 5% share rate. Once share rate crossed 3% in the first hour, the algorithm multiplied distribution speed by 4x.
How to earn this signal: Design content that makes the viewer think "someone I know needs to see this" — useful information for people they know, a situation that describes their friends' lives, or an opinion they want to prove to someone specific. Effective CTA: "Send this to a friend who keeps making this same mistake."
4. Saves — The Reference Value Signal
What the algorithm measures: How many times a video is added to the viewer's favorites list. A save tells the algorithm something a like never can: the viewer wants to return to this video later — proof of lasting, not just momentary, value.
How it works in practice: The algorithm classifies content with a high save rate as "high-value reference content" — a classification it rewards with wider distribution because TikTok wants to be a useful resource, not just entertainment.
How to earn this signal: Present information that can't be absorbed in a single viewing — lists, step-by-step guides, tools. Effective CTA: "Save this video — you'll need it tonight" instead of "give it a like."
5. Comments — The Video Lifespan Extender
What the algorithm measures: Comment count and depth, but more importantly their indirect effect on watch time — a viewer reading comments while the video plays in the background raises the "dwell time" signal on that video.
How it works in practice: A new comment revives the video in the algorithm's eyes. While the typical lifespan of an average video is under 48 hours, a video with active comments can sustain distribution for two weeks.
Real example with numbers:
A career advice video ending with "Do you support quitting without a backup plan?" — 1,400 comments vs. an average of 50 for similar videos. The algorithm kept pushing it for 14 consecutive days.
How to earn this signal: Ask a question with two opposing answers, or leave a deliberately controversial point that begs a response. Read do comments help on TikTok for the full techniques.
6. Negative Signals — The Instant Brakes
Negative signals don't just slow distribution — they stop it almost instantly, and sometimes penalize the next video from the same account.
| Negative Signal | How It Works | When It Becomes Destructive |
|---|---|---|
| Immediate skip (first 2 seconds) | Tells the algorithm the hook failed | When it happens to 75%+ of the test sample |
| "Not interested" tap | Heaviest negative weight — resets targeting negatively | 5% of the sample is enough to isolate the video immediately |
| Report | Triggers immediate manual review, may affect entire account | Any volume triggers the review process |
This explains why posting content outside your niche punishes you even with your existing audience — followers who aren't interested in the new topic hit "Not interested," and that signal collapses the video's score instantly. See our Shadow Ban guide to understand the effects of accumulated negative signals.
7. User Profile vs. Content Profile: How Matching Happens
The algorithm builds a detailed profile for every user based on: the topics they watch, how long they spend on each content type, the sounds that repeat in their viewing history, and the accounts they engage with. This profile determines who will see your video in the initial test phase.
At the same time, the algorithm reads every video through: caption text and spoken words (TikTok auto-transcribes audio), hashtags, the sound used, and computer vision analysis. Together these signals form the video's "identity card" that determines who it gets shown to.
The Practical Formula
Based on these mechanisms, the algorithm's decision can be summarized in one equation:
- Design every video so that completion is the viewer's natural path — don't leave weak spots in the middle
- Create shareable content: useful information, relatable situations, or opinions people want to prove to their network
- End with a controversial question to spark comments and extend the video's lifespan
- Add a subtle detail to encourage replays
- Stay in your niche — leaving it triggers negative signals even from your current followers
For the precise numerical weights of each signal and how they compare, see TikTok algorithm signal priority: the real weights. And to measure how your videos are actually performing against these signals, see how to read TikTok analytics correctly.
Frequently Asked Questions
How long does TikTok take to evaluate a new video?
The first decision is made within 30 to 60 minutes based on the initial test sample's performance. But a video doesn't necessarily "die" after that — unlike Instagram, TikTok can bring an older video into a new distribution wave if it receives unexpected external engagement. Videos have gone viral days after posting.
Do the words spoken in a video affect its distribution?
Yes — TikTok automatically transcribes spoken words and uses them to classify the video and direct it to the right audience. This means using the right keywords in your actual speech inside the video helps the algorithm match it to people searching for related content — not just the hashtags and caption.
Does an account's average completion rate affect new videos?
Yes — the algorithm builds "authority" for each account based on the performance history of its videos. An account that maintains an average completion rate between 15% and 22% gets wider initial distribution for new videos, because the algorithm treats it as a reliable source of quality content.
Does the sound used in a video affect distribution?
Yes — a trending sound connects your video to an existing discovery wave and increases the chances of it appearing to users already engaging with that sound. Original audio carries a different advantage: if your video goes viral and the sound becomes trending, every video that later uses it automatically references your account as the source.
Does posting a new video reactivate old videos on the account?
Indirectly — yes. When an account posts a new video that performs well, more visitors land on the profile, exposing them to older videos. This is why an old video might suddenly pick up views after a new video goes viral — but it's an indirect effect from increased profile traffic, not a direct algorithmic reactivation.