How to find viral TikToks before they peak

How to get 400 million views on TikTok by turning virality into a science. Most people scroll TikTok endlessly, hoping to stumble across the right trend or idea. We stopped hoping and built a system.

First, you have to redefine virality. It isn't about raw views. It's about the velocity of views relative to the account's follower count. The slope of the line matters much more than the height. A 500K-view video on an account with 1,000 followers is a strong signal. Finding 20 of those videos across 20 different small accounts? That's an absolute goldmine. There are video formats that have achieved this level of repeat success, but they're incredibly rare. They're also basically impossible to find if you're trying to do it manually.

The current way most creators find inspiration is slow: create a new account, search for niche videos to "warm it up," and then scroll the For You Page waiting for winners. This works, and the advantage is that you only see videos TikTok is actively pushing on the FYP at that moment, which acts as a quality filter. But the disadvantages outweigh it. You only ever see a small slice of your niche, and it would take hours of scrolling to fill a library of viral videos.

The system we built fixes this. It lets us find formats on the cusp of going viral before everyone else does, and more importantly, it identifies formats that consistently work across creators big and small. You can run this exact system by hand using these steps.

Step 1: Define your search corpus

You can't monitor all of TikTok, and you don't need to. What you need is a fixed list of search terms that covers your niche, because a repeatable process needs a defined universe to sample from. We call this the corpus.

Start with 30 to 50 obvious keywords for your niche. Then add the hashtags your audience actually uses, the slang, the product names, the phrases people type when they're looking for this content rather than describing it. The list should feel slightly too big. It'll shrink itself in step 3.

One thing to set up on day one: a stoplist. Tags like #fyp, #viral and #trending appear on everything and tell you nothing. Ban them from your corpus now so they never sneak in later.

Step 2: Set up your search cadence (breadth over depth)

Every term gets searched twice: once filtered to videos posted in the last 24 hours, once to the last week. Sort by most liked. Look at the first two pages and stop.

The instinct is to go deep, to page through hundreds of results so you don't miss anything. That instinct is wrong. A video that's going to matter shows up near the top of a likes-sorted search within a day of being posted. Page 9 of last month's results is where trends go to die. Two shallow passes over 30 terms beats one deep dive on any single term, every time.

We run keywords every 2 hours and hashtags every 4. By hand, once or twice a day is fine. What matters is that the cadence exists at all, because the 24-hour window only works if you actually come back tomorrow.

Step 3: Curate the term list like a portfolio

Your terms are positions, and they should earn their place. A term that keeps surfacing winners gets checked more often. A term that's produced nothing in weeks drops to a weekly glance. We tier ours into hot, core, and tail, and the tiers reshuffle themselves based on yield.

The list also grows itself if you let it. When a high-performing video carries a hashtag you don't track yet, and then another one does, and then a third, that tag has earned a spot in the corpus. Our rule is three appearances on strong videos within a week. You don't have to be that precise, but you do have to have a rule, or your list stays frozen at whatever you guessed on day one.

Step 4: Track on a decay schedule

Not every video you find deserves tracking. The bar for entry: it's doing at least 300 views per hour since posting, or its views are 3x the account's follower count. That second one matters more than people think. A 2,000-follower account doing 500K views is the purest signal on the platform. A 5M-follower account doing 500K views is Tuesday.

For anything that clears the bar, check it again at 6 hours, 24 hours, and 72 hours, then stop. Log views, likes, comments, shares and saves each time. If it gained fewer than 100 views per hour between checks, cut it from the schedule.

The point of all this is that a snapshot lies and a curve doesn't. Total views tell you what already happened. The difference between two checks tells you what's happening right now, and right now is the only thing you can still act on.

Step 5: Score it

Once you have curves, ranking is almost mechanical. Weight what's spreading now over what has the most views. Our order: recent velocity first (views gained per hour between your last two checks), then overall views per hour, then follower lift. If the same video reaches you through more than one route, a keyword search and a hashtag and your own feed, bump it up. Independent sightings mean it's actually moving.

Then the penalties. Under 5 seconds or over 90, halve the score, because you won't be able to do anything with the format. Dance, music, gaming, pets, anything where the visuals are the whole point, halve it. And the celebrity penalty: if the account has over a million followers and the video didn't even do half their follower count in views, halve it again. That's reach, not a format. There's nothing in it for you to take.

Step 6: Filter by what you can reproduce a version of

This is the filter almost everyone skips, and it's why people end up with folders full of "inspiration" they never use. A viral video is only worth something to you if you can produce your own version of what made it work.

Our fail list, in the order things actually show up: multiple people on camera, lip-sync performances, choreography, screen recordings, slideshows, animation, videos where a physical product is the star, and anything with chaotic rapid cuts you'd need an editor to match. The pass list is short: one person talking to camera, a silent person with a text overlay, or a single static shot with text and music.

Two quick tests settle the edge cases. The focal point test: what is your eye actually on for most of the video? If it's not the person, you can't replicate it by being a person. And the first frame test: could you recreate roughly the opening frame? If the answer's no, the rest doesn't matter. Judge what you can see and hear, never the topic. Topics transfer between formats. Formats don't transfer to you unless you can shoot them.

The whole thing in 8 lines

Build a list of 100+ niche search terms and hashtags. Stoplist the junk.
Search each twice, last 24 hours and last week, sorted by likes, two pages max.
Promote terms that produce, demote ones that don't, harvest new hashtags off winners.
Track anything doing 300+ views/hour or 3x its follower count.
Re-check at 6, 24, and 72 hours. Cut anything gaining under 100 views/hour.
Rank by momentum, not totals. Discount celebrities.
Keep only what one person could re-shoot talking to a camera.
Come back tomorrow and do it again.

That last line is the honest one. Every step here is simple, and none of it works if you do it once. We know because we did it by hand until the checking schedule started running our lives, and then we wrote software to run the loop for us. It now searches every 2 hours, re-checks views on the decay schedule around the clock, scores everything hourly, and screens every candidate against the reproducibility rules before a human ever sees it.

But the software isn't the edge. The loop is the edge. We just got tired of being the loop.