Viral reels are not luck. They are a hook family.
How to make viral reels: hook families with measured lift from 4.5 billion TikTok posts, the share signal that predicts a hit, and the new-account warm-up.
Most people describe a viral reel the way they describe a lottery ticket: it happened to someone else, nobody knows why, and the next one is as unpredictable as the last. The data from 4.5 billion TikTok posts disagrees in a specific way. Viral reels cluster around a small set of opening structures, and those structures have measurable, slowly changing lift.
So which hook families actually carry lift, which one is still growing, what tells you on day one whether a reel is in contention, and why do new accounts need a warm-up before any of it applies?
First, the definition we measure against
A "hit" here is a post in the top 1% by views within its own country and week. Everything below is based on our analysis of 4.5 billion TikTok posts (2014 to 2026) with that definition, so a hit in a small market in a quiet week counts the same as a hit in a big one. "Lift" means how many times more often a feature appears among hits than among average posts.
The hook families, ranked
We tagged opening hooks by type and measured lift globally:
| Hook family | Lift among top 1% posts | Trend |
|---|---|---|
| GRWM, vlog, ASMR openings | 3.32 | Losing 10 to 15% a year |
| "Part N" series | 2.90 | Growing: 2.81 in 2024, 3.01 in 2026 |
| How-to | 2.65 | Losing 10 to 15% a year |
| Tips | 2.49 | Losing 10 to 15% a year |
| Question | 2.08 | Losing 10 to 15% a year |
| Storytime | 1.81 | Losing 10 to 15% a year |
| POV | 1.39 | Losing 10 to 15% a year |
Two things to read off that table. First, every family above 2.0 roughly doubles or triples the odds compared with an opening that belongs to none of them. Second, most families lose 10 to 15% of their lift every year as audiences learn the pattern. The one exception is the series: "part 3 of the kitchen renovation" creates a reason to come back, and the platforms reward return behaviour. It is the only hook whose lift went up between 2024 and 2026.
For a business account this is an instruction, not a trend note: do not post twelve unrelated reels. Post three series of four parts.
The formula inside the family
A hook family is the shape of the opening. The sentence still has to work. The rule we apply to every script comes from Derek Muller's experiment, described in the video "Veritasium: why he still gets views": students scored 6 out of 26 on a physics test before watching; after a clear, well-explained video they scored 6.3, because they heard what they already believed and stopped listening; after a video that opened by stating the misconception, they scored 11. Break the belief first, then ask the question, then explain.
Our best-performing reel so far did exactly that. One reel made by the first version of the Ultim agent on our 3D printing channel, posted 25 September 2026, reached 181K views on Instagram and 79.6K on TikTok, with 4,211 shares between them. The reel was about a record printed on a home 3D printer. The hook broke a belief, the question came before the answer, and the shares followed. One reel is one reel, but it is the pattern the data predicts.
The signal that tells you on day one
Comments feel like traction. The data says they are not. A post with many comments is not meaningfully more likely to be a hit than one with few. Shares and saves are different:
- Shares per view between 0.5% and 1%: a 2.5% chance of reaching the top 1%.
- Shares per view of 2% or more: 5.6%.
- Zero shares: 0.01%.
- Saves per view of 2% or more: 4.0%.
Views arrive fast, too. In our own 100-reel experiment on a fresh TikTok account, almost every reel got its views in the first 12 to 24 hours and then went flat; the 100 reels collected 12 shares in total, which told us on day one that none was in contention. The useful habit is to check the share rate at 24 hours and decide what to make next from that, not from how the reel felt to post.
The levers that are free
Hook family is the biggest creative lever. The rest are metadata, worth about 3x combined at best, and all of them are free:
- Length: 61 to 120 seconds, lift 3.08; 8 to 15 seconds, 0.63.
- Caption: 21 to 100 characters with no hashtags, 1.7 to 1.8; empty, 0.27.
- Sound: own voice or own recording, 1.09; someone else's music, 0.79.
- Posting hour: 10:00 to 18:00 local, 1.04 to 1.12, peak at 15:00 to 16:00; 21:00 to 02:00, 0.78 to 0.84.
The post on what an AI reels generator can and cannot change goes through each of these with the hashtag exceptions by country.
The warm-up: why the formula fails on reel one
Here is the part most "viral formula" posts leave out. Platforms show new or dormant accounts to very few people. The first three to eight reels on a fresh account usually earn little, whatever the hook. That is the account warming up, not the content failing.
Ultim plans this phase on purpose: safe formats first, then the experiments. A reel that tests a bold hook in week one, on an account the platform is still sizing up, tells you nothing about the hook. The same test in week three, with recent posts on the account, is a real test. In our 100-reel experiment, Instagram with zero followers returned about 800 views across 20 reels while the same reels on TikTok got a median of 303 each; the platform decides how big the first pool is, and a new account's pool is small.
Volume is the formula's other half
Among 2.87 million creators with at least 3 posts, the median number of posts before a first top 1% hit is 5, the third quartile is 16 and the 90th percentile is 38. The chance an account has at least one hit is 1.6% at 3 to 4 posts, 10.9% at 20 to 49 and 39% at 100 or more. One percent of creators make 77% of all hits, and in a dataset where volume is the strongest measurable predictor, that one percent is mostly the people who did not stop.
Frequency has a ceiling: 1 to 3 and 3 to 7 posts a week give the same share of accounts with a hit, and above 7 a week the chance per post drops. The formula is therefore not "one perfect reel". It is a hook family, a share check at 24 hours, and 3 to 7 posts a week until you pass post fifty. The full numbers are in how many posts before a video goes viral.
What this means for your reels
Pick two or three hook families and build series inside them, because the series hook is the only one gaining lift. Open every reel by breaking a belief and asking the question before you answer it. Keep the caption to one plain sentence, use your own voice track, post in the afternoon. Treat the first eight reels on a new account as warm-up and judge nothing by them. Then watch shares per view at 24 hours and let that choose the next reel.
That loop is what Ultim runs for a business: it writes hooks in these families from your website, tests one variable per reel, reads the results at 24 and 48 hours, and benches what fails. The free first step is the hook itself. Paste your site into /tools/reel-hooks and see which of the five openings you would stop for.
FAQ
Which hook type goes viral most often?
By lift among top 1% posts: GRWM, vlog and ASMR openings at 3.32, the part N series at 2.90, how-to at 2.65, tips at 2.49, question at 2.08, storytime at 1.81 and POV at 1.39. Only the series hook is gaining lift year over year.
How do I know within a day if a reel has a chance?
Watch shares per view, not comments. At 0.5 to 1% shares per view a post has a 2.5% chance of reaching the top 1%; at 2% or more, 5.6%; at zero shares, 0.01%.
Why do my first reels on a new account get almost no views?
Platforms show new accounts to very few people. Expect the first three to eight reels to earn little. That is the account warming up, not the content failing, so test safe formats first and run experiments once the account is warm.