TikTok Hooks Ranked by Data: We Measured 2,421 Real Openings

Nearly every guide to TikTok hooks opens the same way: ask a question. It is the first category in almost every list, on the theory that a question makes the viewer supply an answer and therefore commit.

We had the data to check, so we checked. Across 2,421 real TikTok posts where we could read the on-screen hook and the play count, question hooks were the worst performing family we measured at every account size above 1,000 followers. At accounts over 100,000 followers they did 40% worse than the alternative.

This is not a claim that questions never work. It is a claim that the single most repeated piece of hook advice does not survive contact with a corpus, and that something else does.

What we measured, and what we threw away

We hold 4,404 real TikTok posts collected across nine verticals: SaaS, beauty, dental, fitness, food, e-commerce, home services, real estate, and a general pool. Each record carries the play count, like and comment and share counts, duration, follower count at collection time, and — the part that matters here — the on-screen text and the speech transcript.

On-screen text is the visual hook: the words burned into the first frames. The transcript is the spoken opening. Those two fields are what a hook actually is, and they are why this analysis is possible at all.

From 4,404 we dropped 132 posts flagged as paid ads, because their reach is bought rather than earned and they would flatter whatever hook they happened to use. That leaves 4,265 scored posts, of which 2,421 carry usable on-screen text. Play counts in that set run from 984 to 42.2 million, with a median of 34,800.

The first result we got was wrong

Our first pass compared posts that had on-screen text against posts that did not. Posts with a written hook showed a median of 34,800 plays against 24,700 — a tidy 1.41x, exactly the kind of number that becomes a marketing statistic.

Then we checked whether it was really about the text. Accounts using on-screen text had a median of 11,800 followers; accounts not using it had 8,955. Bigger accounts get more views regardless of what they write. So we split the corpus into follower bands and re-ran it inside each one.

The 1.41x is almost entirely an artefact of the smallest accounts. Above 1,000 followers, adding on-screen text is worth nothing measurable — slightly negative in two bands, flat in the third.

The honest reading of the under-1,000 band is not that text is magic for beginners. It is that among very small accounts, the ones bothering to burn a hook into the frame are the ones taking the whole thing seriously, and the other habits travel with it. Selection, not causation.

We are leading with the result we got wrong because it is the one most likely to be quoted back at us, and because a 1.41x that dissolves under a single control is what most "we analysed N videos" content is made of.

Question hooks lose, and lose harder as you grow

Same method, applied to the hook families themselves. A question hook is one whose first line ends in a question mark or opens with what, why, how, when, who, which, is, are, do, does, did, can, should, would, will, or if. There are 459 of them.

The pattern is the mirror image of the on-screen-text result, and it carries the same warning: the only band where questions win is the band where everything wins. Above 1,000 followers the effect is consistently negative, and the penalty deepens as the account grows.

A plausible mechanism: a question defers the payoff. It asks the viewer to hold a thought for a beat before anything is delivered. On a feed where the decision is made in about two seconds, deferral is a cost. Small accounts are compared against an audience that mostly is not watching anyway; large accounts are competing against their own back catalogue, and there the cost shows.

What actually wins: the scenario hook

The family that beat the corpus was the scenario opener — a hook starting with POV, "when", "that moment", or "me when". It puts the viewer inside a situation before it asks them for anything.

Head to head against question hooks, restricted to posts that have on-screen text, scenario openers win in all four follower bands:

Four bands, four wins, ratios between 3.0x and 5.9x with no band anywhere near parity. That is about as stable as an effect gets in a corpus this size.

The caveat that kills the clean version of this

Scenario hooks are not universally good, and we would be doing exactly what we criticised if we stopped at the last table. Split the same effect by vertical:

In SaaS the effect is enormous. In beauty it is mild. In the general pool and in dental it reverses — scenario hooks do roughly half as well as everything else.

SaaS is 2,193 of our 4,404 posts, so it carries a large share of the pooled result. The correct conclusion is narrower than the headline: scenario hooks are a strong default for business, software and professional-services content, and an open question everywhere else. If you sell to consumers in a visual category, test it rather than adopting it.

Say something. Silence costs you at every size.

The most consistent effect in the whole corpus is not a hook family at all. It is whether anyone is speaking.

Four bands out of four, same direction, no reversal. The advantage shrinks as accounts grow — from 3.66x to 1.07x — which is what a real effect usually looks like once other advantages accumulate. Text-on-screen with a silent track is the format this data likes least.

The 31-to-60-second dead zone

Duration produced the other four-out-of-four result, and it is not the monotonic "shorter is better" rule that gets repeated. Posts between 31 and 60 seconds underperform everything else at every account size:

Across the unsplit corpus, posts under 15 seconds median 45,200 plays and posts over 60 seconds recover to around 32,000 — but the 31-to-60 band sits at 15,200. It is a trough, not a slope. Our read is that 31 to 60 seconds is long enough to require a real structure and short enough that most creators do not give it one: a clip that would have been tight at 20 seconds, padded.

If you are cutting a long video down, the actionable version is to land under 30 seconds or commit past 60, and treat the middle as the length you have to justify. That is a rule we apply in our own AI clip generator when it picks clip boundaries.

How to use this without overfitting

Six findings, ranked by how much we would bet on them:

1. Put speech in the video. Four bands, four wins, no reversal, and the only finding with no vertical that contradicts it.
2. Avoid the 31-to-60 second band unless the structure earns it. Four bands, four wins.
3. Prefer a scenario opener to a question — but if you are not in SaaS or professional services, treat it as a test rather than a rule.
4. Do not reach for a question hook by default just because every list starts there. Above 1,000 followers it cost every band we measured.
5. Do not expect on-screen text alone to do anything above 1,000 followers. Write it because it helps sound-off viewers, not because it buys reach.
6. Hook length is close to noise. Seven-to-ten words did best at 1.38x baseline and four-to-six words worst at 0.75x, but the pattern is not monotonic and we would not act on it.

If you want the raw templates rather than the measurements, our viral hook formulas post has the copy-paste versions, and the free hook generator will draft openers against your topic. For how the same corpus behaves on reach rather than hooks, see what actually goes viral on TikTok.

What this data cannot tell you

It is TikTok only. We have no Instagram Reels corpus, so nothing here should be read as a Reels finding — the platforms differ enough on watch-time weighting that transferring the numbers would be a guess dressed as evidence.

It is a snapshot, not a time series. We cannot tell you whether question hooks used to work and stopped, only that they are behind right now.

Play count is not the only goal. A hook that gets fewer plays from a better-qualified audience can be the correct business decision, and no view-count analysis will ever show that.

And hook family is one variable among many. We controlled for follower count and excluded paid ads; we did not control for posting time, sound choice, thumbnail, or the quality of what follows the first three seconds — which is, in the end, what decides whether anyone stays.

Frequently asked questions

What is the best hook for TikTok?

On our corpus, a scenario opener — POV, "when", "me when" — beat question hooks in all four follower bands by 3.0x to 5.9x. But the effect is strongest in SaaS and professional-services content and reverses in dental and our general pool, so it is a strong default rather than a universal rule.

Do question hooks work on TikTok?

Less well than their reputation suggests. Among the 459 question hooks we measured, they underperformed every other hook family at 1,000-10,000 followers (0.82x), 10,000-100,000 (0.83x) and 100,000+ (0.60x). They only won among accounts under 1,000 followers, where nearly every hook family wins for reasons unrelated to the hook.

Should I add text on screen to my TikToks?

For accessibility and sound-off viewing, yes. For reach, our data says it does nothing measurable above 1,000 followers — the apparent 1.41x advantage disappears once you control for account size.

How long should a TikTok be?

Under 30 seconds or over 60. Posts between 31 and 60 seconds underperformed at every account size we tested, from 0.34x among the smallest accounts to 0.77x among the largest.

How many posts is this based on?

4,404 collected posts, 4,265 after removing paid ads, and 2,421 with usable on-screen hook text. Every group we report has at least 40 posts in it; smaller groups were suppressed rather than published with a caveat.