How the TikTok algorithm works in 2026 (signals ranked)
The TikTok algorithm ranks every video for every viewer by predicting what that person will do: watch it to the end, like it, share it, comment. TikTok has said so since June 2020, and adds that follower count is not a direct factor. In the viral videos I analyzed, shares carry the most weight: the most-shared quarter has a view floor 2.7 times higher than the least-shared quarter.
- TikTok groups its signals into three families: your interactions, video information (caption, sound, hashtags), and device and account settings. Settings weigh the least.
- Videos shared more than once every 43 views top 736,000 views 3 times out of 4. Under 5.7 shares per 1,000 views, 1 in 4 stays below 276,000.
- Comments point the other way: videos with more than 2 comments per 1,000 views get 438,000 views on average, those under 0.25 get 2.7 million.
- Engagement rate doesn't separate a good video from a mega-viral one: 5.7% under 500K views, 6.0% above 10 million.
What TikTok officially says about its algorithm
TikTok published its own explainer on June 18, 2020, and it still holds. The For You feed ranks videos using three families of signals. First, your interactions: videos you like or share, accounts you follow, comments you post, videos you create. Then video information: captions, sounds, hashtags. Last, device and account settings: language, country, device type.
TikTok also gives two weighting rules. A strong signal counts more than a weak one, and its example is finishing a longer video from beginning to end. Device settings get less weight because you don't pick them as a preference. The line that matters most for creators comes later: neither follower count nor a history of high-performing videos is a "direct factor."
In December 2022, TikTok added a "Why this video" button to the share panel. It lists the same reasons plus two more: your searches, and what's popular or recent in your region.
The signals ranked, source by source
No source gives the exact weight of each signal, and they don't even list the same ones. In December 2021, Ben Smith of The New York Times described an internal document called "TikTok Algo 101," which TikTok confirmed was real. It shows a simplified formula: predicted like, comment and play, plus estimated playtime, each multiplied by a value. Shares aren't in it, even though TikTok's 2020 explainer names them right after likes.
| Signal | TikTok (June 2020) | Internal doc (2021) | What my data shows |
|---|---|---|---|
| Watch time, finishing the video | Named as a strong signal | Yes (estimated playtime) | Hidden on other people's videos |
| Shares | Yes | Missing | The clearest link to views |
| Likes | Yes | Yes | Part of a flat engagement rate |
| Comments | Yes | Yes | Reverse link: more per view, less reach |
| Saves | Not named | Missing | Hidden, so excluded |
| Caption, sound, hashtags | Yes | No detail | Not measured |
| Language, country, device | Yes, lower weight | No detail | Not measured |
| Follower count | Not a direct factor | Missing | Not measured here |
Sources: TikTok Newsroom, How TikTok recommends videos #ForYou (June 18, 2020); Ben Smith, The New York Times, syndicated by Boston.com (December 6, 2021); viral videos analyzed one by one by Prooven (2023 to July 2026).
My ranking, heaviest first: completed views, shares, likes, comments, video information, settings. The top spot comes straight from TikTok. Putting shares above likes is my call, based on the data below.
Shares raise your view floor
A heavily shared video is less likely to stall. I split nearly 500 viral videos into four equal groups by shares per 1,000 views. In the least-shared group, 1 video in 4 stays below 276,000 views. In the most-shared group, 3 in 4 clear 736,000.
It isn't a straight line. The two middle groups are about even, and the ceiling barely moves: in every group, the top quarter of videos passes 3.5 million views. Shares mostly protect you from a flop. They don't buy you a jackpot.
The useful bar sits at 23.4 shares per 1,000 views, or one share every 43 views. Above it, the typical viral video gets 1.6 million views. Under 5.7 shares, it gets 1.1 million.
My reading, which TikTok neither confirms nor denies: a share brings in a viewer who skipped the For You feed entirely. A friend picked the video for them, so they have more reason to finish it.
Good video vs. mega-viral: what actually changes
Three things change sharply between a video under 500K views and one above 10 million: length, format, and comment rate. Shares go up too, though less than most people assume. I compared nearly 140 viral videos under 500K views with 46 videos above 10 million.
| Benchmark (typical video in each group) | Under 500K views | Over 10M views |
|---|---|---|
| Shares per 1,000 views | 10.3 | 14.8 |
| Shares, most-shared quarter (from) | 15.4 | 31.5 |
| Comments per 1,000 views | 1.42 | 0.28 |
| Engagement rate | 5.7% | 6.0% |
| Length | 50 s | 26 s |
| Format you can copy as is | 40% | 67% |
| Shared "because it's funny" | 10% | 35% |
Units: shares and comments per 1,000 views; length in seconds; shares of videos in each group, in %. Engagement rate: visible likes, comments, shares and saves divided by views. Source: viral videos analyzed one by one by Prooven (2023 to July 2026, mostly Instagram Reels).
Between 1 and 5 million views, the typical video gets 13.8 shares per 1,000 views. Above 5 million, 13.5. So the share rate can't tell a 2-million-view video from a 20-million one. Length can: half the videos above 10 million views run 26 seconds or less. A short video is easier to finish, and finishing is the first signal TikTok names.
The reasons to share that travel furthest
Videos people send "because it'll help you" or "because it's funny" get the most shares per view. Debate videos get the fewest shares and the most comments.
The clearest case in my data comes from taryn arnold scott, a US personal growth creator. Her July 20, 2025 Reel runs 27 seconds and pulled 23.9 million views at 18.7% engagement. Of all the videos above 10 million views I analyzed, it has the highest share rate. Her opening line asks for the share without sounding like a call to action:
"If someone pops in your head when I say the thing that I'm about to say, I think that that's a sign that you should send this to them."
taryn arnold scott, Reel from July 20, 2025
The viewer has a specific person in mind before the video ends. Truman Kain does the same thing with fear instead of warmth: a 58-second security alert with a fix, posted December 20, 2025, at 13.2 million views. Debates work the opposite way, with 8.3 shares per 1,000 views but 0.84 comments, the highest comment rate of any type. People argue with a debate. They rarely forward it.
Myth: comments push your views up
My data shows the reverse: the more comments a video gets per view, the less far it goes. Videos with more than 2 comments per 1,000 views average 438,000 views. Those under 0.25 average 2.7 million.
It comes down to who comments. Your followers know you and reply. A stranger on the For You feed watches, maybe likes, and scrolls on. Once a video leaves your follower base, its comment rate shrinks. I use this ratio in the shadowban check to spot a video that stayed with your followers.
"Comment a word to get my guide" doesn't fix it. In my data, Antoine Martel's March 19, 2026 Reel asks viewers to comment "blueprint." It got 7,532 views. Comments are in the 2021 formula, but they can't stand in for a video people watch to the end.
Myth: followers and engagement rate decide your reach
TikTok says plainly that follower count is not a direct factor. An account with 800 followers enters the same race as one with 2 million. Viral videos from tiny accounts, and what they share, are covered in why your TikTok has 0 views.
Engagement rate doesn't decide either. Under 500K views, the typical viral video sits at 5.7%. Above 10 million, 6.0%. That gap is far too small to explain 20 times more views. The number still helps you compare your own videos, and the engagement rate calculator does it in seconds. What matters is the mix behind it: shares count for more than comments.
Saves, the signal nobody can see
Nobody outside TikTok can measure how much saves weigh, me included. Instagram hides save counts on other people's videos, and I have them for a single video in my detailed set. I left them out of every calculation instead of counting them as zero. Sites claiming "a save is worth 5 likes" cite no source.
What I can tell you is why people keep a video. Under 500K views, 60% of viral videos get saved as a "useful resource." Above 10 million, that drops to 48%, while checklists and step-by-step guides rise from 29% to 37%. A list is something you come back to.
What changed in 2025 and 2026
The core signals haven't moved since 2020, but viewers got more controls. In June 2025, TikTok launched Manage Topics, which lets anyone turn more than 10 topics up or down in their feed, from travel to sports. Smart keyword filters, up to 200 words, came out the same day. Since March 2023, viewers can also refresh their For You feed as if they'd just signed up.
In the US, the big change is ownership. The TikTok USDS Joint Venture was announced on January 23, 2026. Its stated job includes retraining, testing and updating the recommendation algorithm on US user data, inside Oracle's US cloud. It gave no date. So far, nothing official says the signals themselves have changed, so the rules above still apply.
Europe went another way. Since August 2023, the Digital Services Act has required TikTok to offer EU users a non-personalized For You feed, which shows what's popular in their region and language.
My take
Stop writing for comments and start writing for shares. In my data, shares are the only visible signal that lifts the view floor, while comments climb mostly when a video stays with your followers. Before you post, name the person your viewer will send it to. If you can't, rewrite the hook. Then cut: half the videos above 10 million views run 26 seconds or less. The limit: I can't see watch time or saves on other creators' videos, and my data is mostly Instagram Reels. I can't tell you how much completion outweighs shares on TikTok specifically. For the formats that put these rules into practice, read the guide on how to go viral on TikTok.
Frequently asked questions
How long does TikTok test a new video?
TikTok publishes no test window, and the sites quoting "1 hour" or "48 hours" cite no source. To see what view count is normal for your account size, check how many views counts as viral.
Do hashtags still matter?
Yes, as one input among others. TikTok files hashtags under video information, alongside caption and sound, and doesn't publish their weight. A hashtag helps TikTok understand what your video is about. It won't push it on its own.
Can you reset the algorithm on your account?
You can refresh your viewer feed in settings, a feature live since March 2023. There's no equivalent for your creator account: TikTok judges each video on its own signals.
- TikTok Newsroom, How TikTok recommends videos #ForYou (June 18, 2020)
- TikTok Newsroom, Learn why a video is recommended For You (December 20, 2022)
- Ben Smith, The New York Times, How TikTok reads your mind, via Boston.com (December 6, 2021)
- TikTok Newsroom, Announcement from the new TikTok USDS Joint Venture LLC (January 23, 2026)
- TikTok Newsroom, New features to help you control your scroll (June 3, 2025)
- TikTok Newsroom, Introducing a way to refresh your For You feed (March 16, 2023)
How this article was made. The figures come from nearly 500 viral videos analyzed one by one by Prooven (posted from 2023 to July 2026, mostly Instagram Reels), with likes, comments and shares recorded after posting; saves, which are hidden, were excluded and never counted as zero. Platform rules come from TikTok's official newsroom and the press coverage cited, read in October 2026. Drafted with AI assistance, figures checked and article reviewed by Antonin Girardeau.
Find the formats that work for you
Prooven analyses your account, gives you a score out of 100 and shows you one viral format a day, adapted to your niche. Coming soon to iPhone.