We Analysed 2,584 TikTok Videos: The Most Common Hook Isn't the Best One

What we measured

Disclosure. FYPNow sells the analytics product that produced this data. We have an

obvious interest in you finding it credible, which is why the method, the sample bias and

the places the numbers disagree with our own marketing are all stated below rather than

buried. The figures are published under CC BY 4.0 — check them.

2,584 TikTok videos, each run through the same automated analysis: a video model classifies the hook from the opening seconds, and engagement, save and share rates come from the metrics attached to each post.

Everything below is a median, not an average, because view counts on TikTok are wildly skewed and one 9-million-view outlier will drag a mean anywhere you like.

Two caveats worth reading before the numbers, because they bound what you can conclude:

The sample is not TikTok. These are videos FYPNow users chose to track — competitors, creators in their niche, their own posts. That skews toward content someone already thought worth watching. Treat these as benchmarks for videos that found an audience, not for TikTok as a whole. Nothing here is causal. Videos with a particular hook type differ in a hundred other ways too. "Shock hooks have lower save rates" is a description of what we observed, not a promise that changing your hook changes your save rate.

The thing we got wrong first

Worth telling, because it changed a number in this post.

The first time we ranked the catalogue by outperformance, the top of every niche was the same

account: a crypto creator with 2 followers and 813,990 views on one video, scoring a

406,995× multiplier. Six of the top nine results in Crypto were that one account.

It was not a breakout. Outperformance is views divided by follower count, and nothing bounded

the denominator below 1 — so a failed follower lookup produced an astronomically large,

completely meaningless number, and because it was the largest number it sorted first.

The fix was a floor: no multiplier is computed below 1,000 followers. That cleared 684 of

14,599 ranked rows and dropped the maximum multiplier from 406,995 to 10,023. The top of

Crypto became respawnedreality — 2.0M views on 2,194 followers, a real 917× — which is what

the metric was supposed to surface all along.

We mention it for two reasons. It is why the "no hook" category in the tables below is not

polluted by junk rows. And if you are computing views-per-follower yourself anywhere, apply

the same floor, or your top results will be data errors wearing a crown.

What people actually use

The bold claim is more than twice as common as anything else. Open with an assertion — a number, a result, a contradiction — and you are doing what a third of the catalogue does.

What actually performs

Same videos, sorted by median save rate. Only hook types with at least 50 videos.

Three things fall out of that table, and two of them are uncomfortable.

Finding 1: Shock gets the most views and the least of everything else

Shock hooks had the highest median view count in the entire sample — 48,060, more than double the bold claim's 22,211 — and simultaneously the lowest save rate and the lowest engagement rate.

That is a coherent story rather than a contradiction. A shock opening is extremely good at stopping a scroll, which is what generates a view. It is bad at making anyone want to keep the video, send it to a friend, or follow the person who made it.

So the honest reading: if your only metric is reach, shock works and the data says so. If you are trying to build an audience, it is the worst-performing hook we measured — worse than having no hook at all.

This is also a decent warning about optimising on views. A hook can win the metric you are watching while losing the thing you actually wanted.

Finding 2: The second-most-popular hook is one of the weakest

The curiosity gap — withholding the answer to make someone stay — is 17.3% of the sample, the second most common thing creators do.

It also sits near the bottom: 0.38% median save rate against the bold claim's 0.82%, and 6.86% engagement against 8.39%. On saves it is barely ahead of videos with no hook at all (0.33%).

One plausible explanation, which the data cannot confirm: a curiosity gap is a promise, and a promise you underdeliver on costs you the save. A bold claim states the payoff immediately — the viewer already got the value in the first three seconds, and saving it is about keeping the value rather than waiting for it.

Finding 3: Hook length barely varies. Video length varies enormously.

This was the most surprising result. Median hook resolution by type:

Every single type lands between 3.0 and 3.2 seconds. The range across 1,245 videos with timing data is 0.2 seconds.

So "how long should my hook be" has a boring answer — about three seconds, regardless of what kind of hook you are writing. The interesting variable is what the hook is buying time for: a relatable hook buys 16 more seconds, a tutorial promise buys 68.

That reframes the decision. You are not choosing a hook style, you are choosing a video length, and the hook has to be the kind that justifies it. A tutorial promise earns a minute. A relatable opener does not.

What we would actually do with this

If you are starting out: write bold claims. It is the most-used hook, it has the best engagement and share rate in the sample, and it is the easiest to write well — state the result up front. If you make educational content: tutorial promises had the best save rate of anything measured (0.88%) and support the longest videos. Saves are the metric that compounds, because a saved video gets rewatched. If your views are fine and your growth is not: check whether you are leaning on shock or curiosity-gap hooks. Both stop scrolls. Neither, in this sample, makes anyone stay. Do not extend your hook to five seconds because a guide told you to. 1,245 videos say three, across every category.

Method

Every video was analysed by the same pipeline: a video model watches the opening seconds and classifies the hook into one of ten categories, while engagement, save and share rates come from the post's own metrics. Hook resolution timing is detected from the video itself rather than estimated from the transcript.

Medians throughout. Hook types with fewer than 50 videos are excluded from the performance table (transformation, challenge and POV each had fewer than three). Timing figures use the 1,245 videos carrying scene-level analysis.

The sample is drawn from creators FYPNow users track, which is a real limitation: it is videos somebody deliberately watched, not a random draw from the platform.

Related

• The anatomy of a TikTok video — length, pacing and audio across the same catalogue
• 89% of TikTok videos have no call to action — the production-choice half of this study
• Looking at the camera underperforms: presenter data from 2,499 videos — eye contact, expression and delivery
• The best TikTok hooks that stop the scroll — the patterns, with examples
• Find viral TikTok videos in 4 steps — how to run this analysis on your own niche