How the TikTok Algorithm Works: A Founder Explanation, From the Source

The TikTok algorithm ranks every video against each individual viewer using signals TikTok itself has published: how that viewer interacts with content (watch time, likes, shares, comments, follows), what the video contains (captions, sounds, hashtags), and low-weight device settings like language and country. Watch time dominates. Follower count, by TikTok's own statement, has no direct effect. Each new video gets shown to a test audience first, and its performance with that group decides whether it spreads.

That is the answer in four sentences. The rest of this article is what those four sentences mean when you are a founder trying to turn views into pipeline, because the algorithm most founders complain about is better documented than almost any other distribution channel they use. TikTok explained its recommendation system publicly in a 2020 newsroom post, and journalists have stress-tested it since. What follows draws on those sources plus what we see running founder accounts daily at JOLT.

What Signals Does the TikTok Algorithm Use?

TikTok's own explanation groups the inputs into three buckets.

User interactions. The videos a viewer likes, shares, and comments on, the accounts they follow, the content they create, and how long they watch. This is the heavy bucket. TikTok gives the explicit example that finishing a longer video from beginning to end is a strong indicator of interest, weighted well above weaker signals.

Video information. Captions, sounds, and hashtags. This is how the system figures out what your video is about and which viewers to test it on. Your caption and on-screen text are targeting inputs, so writing them in the language your customer uses matters more than stuffing them with tags.

Device and account settings. Language preference, country, device type. TikTok says these receive lower weight because the viewer never chose them as preferences.

The system scores a video's likelihood of interesting one specific viewer, then builds that viewer's feed from the ranking. There is no single For You page. There are as many feeds as there are users, which is why "the algorithm hates me" is rarely the right diagnosis. The system holds no opinion about you, only data about how strangers responded to your first three seconds.

The Wall Street Journal ran an experiment with over a hundred automated accounts to watch this personalization happen in practice, and it is worth the roughly thirteen minutes of your time to see how fast watch time alone steers a feed:

The WSJ's finding matched TikTok's stated design: the system learned each account's interests primarily from how long it lingered on videos. Nobody liked or commented on anything. Watch time was enough.

Does Follower Count Matter on TikTok?

Here is the sentence from TikTok's newsroom post that should change how every founder thinks about this channel: neither follower count nor whether the account has had previous high-performing videos are direct factors in the recommendation system.

Read that as a founder. On LinkedIn, distribution compounds on the audience you already built, so a new account shouts into a void. On TikTok, your next video gets judged on its own merits against a fresh test audience. An account with 40 followers can outperform an account with 400,000 on any given video, and we watch it happen across client accounts routinely.

This is the mechanism behind the results that sound implausible until you see the design. When JOLT built the Real American Beer campaign with Hulk Hogan, 90 days of daily short-form content across Instagram, Facebook, and X produced 15 million organic impressions and over $1 million in sales with zero paid media, and the launch reel alone pulled 4.8 million views. Different platforms, same structural truth: recommendation systems that rank content per-video reward the quality of each post over the size of the account behind it. TikTok is the platform that says so in writing.

For a founder starting from zero, that is the entire strategic case: you're paying for good videos, not for years of audience-building debt.

How Does TikTok Test a New Video?

TikTok does not publish exact test batch sizes, so treat any specific number you read elsewhere as folklore. What the published design does tell you: new uploads get evaluated, some content is ineligible for recommendation entirely (spam, videos still under review), and distribution follows demonstrated viewer interest rather than a schedule.

In practice, across the accounts we run, the pattern is consistent. A new video reaches a small initial audience. If those viewers watch, rewatch, share, or follow, the system widens the circle and tests again. If they scroll past in the first seconds, distribution stops, which is exactly the wall founders describe when every video stalls around 200 views. The stall is a measurement: the test audience gave its verdict on your hook, and the system believed them.

Two practical consequences follow. First, the first three seconds carry absurd weight, because they decide whether the test audience delivers the watch time that everything else depends on. Second, a video that flops teaches you something specific about its opening, and the correct response is a sharper next video rather than deleting and reposting the same file into the same verdict.

What Should a Founder Do About It?

Everything above compresses into four working rules. This is roughly the operating system JOLT installs for clients, and none of it requires an agency to start.

1. Optimize for completion before anything else. Watch time is the dominant signal, so length discipline beats production polish. A tight 25-second video someone finishes outranks a beautiful 90-second video they abandon. Cut every sentence that does not earn its seconds.

2. Spend your effort on hooks. The test-audience mechanic means the opening seconds determine whether the rest of the video ever gets seen. Founders have a natural advantage here: sharp opinions about your industry, spoken plainly, are hooks by default. Save the company backstory for viewers who already stopped scrolling.

3. Post enough for the system to learn. Each video is one data point in the algorithm's model of who your audience is. Sporadic posting starves it. The accounts we run publish around seven videos a week, and the compounding usually shows up over a quarter rather than a weekend, which is why we map the ramp in a founder's first 90 days on TikTok. If daily is unrealistic solo, batch a week of filming into one session and protect a floor of three or four posts.

4. Ride sounds and formats before they peak. Video information is a targeting input, and a trending sound or format is a lane the system is already distributing heavily. Trendspotting is half of what clients pay JOLT for: catching a format early, then fitting the founder's actual message into it instead of chasing the trend for its own sake. You can do a manual version by spending ten minutes a day on your own For You page noting which formats keep repeating.

Where the Algorithm Explanation Stops Helping

An honest limit: understanding the recommendation system does not make weak content perform. The algorithm is a distribution engine sitting on top of a judgment it outsources to viewers, and viewers are not confused. If your videos are corporate, hedged, or indistinguishable from every other founder reading bullet points at a camera, perfect algorithmic technique will distribute that boredom to a test audience who will decline it, accurately.

The founders who win this channel treat the algorithm as settled infrastructure and pour their effort into the two things it measures: an opening worth stopping for and a video worth finishing. Start there. Pick one sharp opinion you defended in a meeting this week, film it in under 30 seconds with the claim in the first sentence, and post it. The system will tell you the truth about it within a day, which is more honest feedback than most marketing channels give you in a quarter.