How to Write a 3-Second Hook: the 3 Layers and the Thresholds

A 3-second hook is written in three simultaneous layers: the first frame (what you see at 0.0 s), the first spoken sentence (6 to 9 words at most) and the on-screen text (5 words at most, readable without sound). You write for seconds 0 to 2 and measure at 3: hook rate ≥ 25% on Meta, 2-second view rate ≥ 27% on TikTok. Below that, you rewrite the frame, not the sentence. What follows is the full protocol: ten hooks written in twenty minutes on a single video body, the thresholds by placement, and the matching quote line. According to Médiamétrie’s Année Internet 2025, relayed by CB News in February 2026, mobile accounts for 80% of French people’s browsing time, up from 66% in 2020, and 92% among 15-24 year olds.

Why is “3 seconds” a Meta metric and not a writing brief?

The 3 seconds are not a writing length: they are the moment Meta decides to count a view. At Meta, a 3-second view means 3 continuous seconds of playback, or 97% of the video if it is shorter than 3 seconds (AdSights, Thumbstop Rate glossary, updated August 2026). TikTok measures at a different point: a view is counted as soon as playback starts, and the separate “2-second views” metric is conventionally used as the basis for hook rate (Reloop, Hook Rate: Formula, Benchmarks, 2026). Same creative, same edit, two different counters.

The consequence is plainly mechanical. If you write for “the first 3 seconds”, you write a sentence that ends at 3.4 s. The viewer, meanwhile, has already made up their mind while your subject was saying the third word. So you write for the 0–2 s window, you measure at 3, and you judge the asset’s survival at 15 s (ThruPlay). Three clocks, three uses. If the definition of hook rate and how it is calculated are not clear to your whole team, start again from the definition, measurement and cost of a video ad hook; we do not redefine them here.

Second misunderstanding, and a costlier one. The figure you read in a blog post is not necessarily yours. A paid hook rate (3-second views ÷ impressions) and an organic 3-second retention (viewers still present ÷ viewers who entered) do not measure the same population. The first includes every impression served, including those nobody watched. The second starts from people who have already entered the video. Comparing the two is like comparing a click rate with a conversion rate. You still see it at serious advertisers who set a 50% paid hook rate target on Meta because they read an organic TikTok retention benchmark.

Last point, and an uncomfortable one: Meta publishes no official threshold for a “good” hook rate. The platform documents the metric, not the bar (Sepia-lab, Hook Rate Benchmarks 2026). Every figure in circulation comes from vendors, agencies or account aggregators. That does not make them useless; it means you must cite them with their source and date, and build your own internal reference after three months of delivery.

Which 3 layers should you write separately for a hook?

A hook is made of three layers, frame 0, voice and on-screen text, and each one must be written separately, then stand on its own. This is the rule that changes results the most in our writing workshops. As long as you write “the hook” as one block, you produce a sentence and then illustrate it. When you write three distinct lines, you often realise the sentence is good and the image is dead.

Layer 1, frame 0

This is the still image shown at 0.0 s, before any movement. It is described in one written sentence, in the brief, just like a piece of copy: “close-up of a hand unscrewing a bottle, bathroom light, white tiled background”. Not “product hook shot”. If someone who hasn’t seen the video can’t describe your frame 0 in one sentence, it isn’t written, it is hoped for. On our e-commerce campaigns, when a hook underperforms, the frame is what we rewrite first in nine cases out of ten; the sentence comes after. A TikTok and CreatorIQ report, relayed by Marketing Brew in December 2023, indicates that showing the creator, or anyone else, in the very first seconds increases hooking power by 50% and ad recognition by 32%.

Layer 2, the first spoken sentence

6 to 9 words. Beyond that, it runs past 2 seconds at a natural pace. The constraint seems brutal; it is arithmetic. A creator speaks 3 to 4 words per second in French when comfortable, fewer when reading. Nine words is already 2.4 s. The sentence must also carry the brand’s useful sound: according to the System1 × TikTok study relayed in August 2025, audio assets present in the first two seconds are the branding element with the strongest impact on brand awareness uplift, at +191%. Two seconds of silence on a product shot is two seconds wasted twice over.

Layer 3, the on-screen text

5 words maximum, placed outside the interface zones, readable at arm’s length on a 6-inch screen with the sound off. It is not the subtitle of the sentence. If it repeats the voice word for word, you have lost a layer: you have two instead of three. The on-screen text must supply the information the voice does not yet say, or frame who is speaking. TikTok Business documented in 2024-2025 that brand recognition in the first two seconds generates +57% “happiness” and −19% attention loss: on-screen text is often the cheapest vehicle for that recognition. The TikTok Business blog on creative effectiveness (May 2025) adds that content that is well branded from the start of the ad gets a 25% rise in brand choice.

Validation test for the three layers: turn the sound off, the hook must remain understandable. Hide the image, the sentence must remain intriguing. Remove the text, the shot must remain legible. If a single layer does not survive on its own, the hook is not written, it is illustrated.

A caveat, because a lot of fantasy is sold on this subject: the hook opens, it does not sell. An internal TikTok analysis relayed by Marketing Brew in December 2023 placed most of an ad’s memory impact in the first six seconds. The first two buy the right to exist until the sixth. A perfect hook on top of a limp video body produces a cliff-edge curve and a burnt budget.

How do you write 10 hooks in 20 minutes on the same video body?

The protocol comes down to one equation: 1 promise × 5 attack entries × 2 wordings = 10 hooks written, on an unchanged video body. Twenty minutes on the clock, provided you never go back while writing. You write the ten lines, then you judge. The classic mistake is to polish hook no. 1 for fifteen minutes and rush the other nine.

Step 1, the promise. One sentence, the one the video body already delivers. Not the one you would like to deliver. Working example: “this serum fades post-sun marks in eight weeks, €39”. If you cannot write it, the problem is not the hook, it is the angle, and we deal with that elsewhere.

Step 2, the five attack entries. These are not hook types to copy; they are five doors through which to enter your own promise: the visible symptom, the quantified cost, the head-on objection, the precise moment, the material proof. If you want to enrich this bank, the 8 hook archetypes that stop the scroll serve as a reservoir of entries; here, we turn them into sentences written about your offer, not templates to trace.

Step 3, two wordings per entry, in the table below. One row = one complete hook, all three layers filled in, plus a verdict column “works without sound?”.

Of these ten lines, you will keep four to six. The others fall at the test in the next section. How many hooks per angle before changing angle? The logic of variation is the same as for creative variants, detailed in the 3 then 5 to 8 rule: you start tight, and only widen if the angle shows a sign of life.

Eight psychological levers to vary the two wordings

The five attack entries show where to enter the promise. The levers show which reflex you trigger in the viewer. Crossing the two avoids a frequent trap: ten variants that all play on curiosity and look alike in delivery. For each entry, make the second wording rest on a different lever from the first:

• Curiosity: you open a question without closing it. “What the label doesn’t tell you about this serum.”
• Narrative tension: the viewer recognises an uncomfortable situation they live through. “You’re probably making this mistake every evening.”
• Visual shock: a frame 0 out of context, which forces you to watch one second longer before you understand.
• Counter-intuitive number: a figure that contradicts a widespread belief, with the context arriving right after.
• Direct promise: a concrete, immediate benefit, stated in a few words.
• Emotional identification: “That moment when...” followed by a precise, slightly awkward situation the target has already lived.
• Challenger authority: “Stop believing that...” applied to a received idea in your sector.
• Sound break: an unexpected or very familiar sound before the image, followed by a sharp cut. This lever fails the muted test by construction: reserve it for placements where sound is likely.

Two safeguards. Visual shock and pure curiosity are the levers most exposed to the hook disconnected from the offer: on an acquisition campaign, check that the product category is still named before second 2. And the choice of lever depends less on the sector than on your target’s state of mind at the moment they scroll. On our B2B or institutional accounts, narrative tension and direct promise more often come out on top; in mainstream fashion and beauty, visual shock and emotional identification take over. It is a starting point to test, not a rule. In the System1 × TikTok study relayed by Chief Marketer in August 2025, logos shown in context, for example a person drinking from a can during the ad, come just behind audio assets with a 182% rise in awareness uplift.

How do you know whether a hook is good before shooting it?

Four written tests eliminate half the hooks before a single euro of shooting is committed. None of them predicts performance; they detect the flaws that guarantee failure. That is already a lot.

The muted test

Read only the “frame 0” column and the “on-screen text” column. If you can’t tell what the video is about, the hook depends on sound. Part of the feed is consumed without sound, and on Meta in-feed placements the proportion remains high enough that you shouldn’t bet against it. Binary verdict: the hook passes or it is rewritten on the image side.

The single-line test

The hook must fit on one table row. If you need two sentences to describe what happens between 0 and 2 seconds, you have written a sequence, not a hook. Opening sequences of three shots in two seconds produce jittery edits and mediocre hook rates: the brain had no time to fix on anything.

The legibility test

Five words maximum, outside the interface zone (the bottom of the TikTok screen, the side banner on Reels), with a font size large enough to be read at a glance. On-screen text cut off by the account name or by the “Buy” button is on-screen text that does not exist. You check that in the ads manager preview, not on the editor’s timeline.

The invisible third party test

Have someone who knows neither the brand, nor the product, nor the brief read the ten lines. Ask them to tick the three they would watch. Not “the three best”, “the three you would watch”. The nuance matters: a marketing team’s aesthetic judgement correlates poorly with scroll behaviour. In our workshops, the client’s favourite hook rarely ends up in the trio picked by the third party.

What we don’t know, and must own up to: none of these tests says which hook will win. The final ranking is decided in delivery, at volume. The pre-shoot tests exist only so that you don’t pay to produce hooks you already know will fail.

The 3T test

Reread each line with three questions. The Trigger: what reflex does it set off, curiosity, a feeling of being concerned, a wish to know what comes next? The Tension: what is left unfinished or uncomfortable at the end of the second second? The Tease: does the line start the answer without giving it away? The most frequent gap is tension: the hook explains instead of questioning. “This serum fades marks” informs and closes the subject; “My marks came back after every summer” leaves a question open. A line without tension gets rewritten before moving on to the invisible third party test.

The kept promise test

Read the hook, then the full script, putting yourself in the place of a viewer who stayed to the end. Do they feel rewarded for waiting, or cheated? In the second case, you have written clickbait. A curiosity hook whose answer never arrives, or arrives in a disappointing form, creates a frustration that falls back on the brand, in the comments as in recall. This test is done on paper, before the shoot, with the whole script in front of you: it costs five minutes, against an entire campaign when you discover it in the results.

Which numerical thresholds validate a hook after delivery?

The bar depends on the placement, not the platform, and a hook should not be judged before roughly 30,000 impressions. On our accounts, Meta Reels generally shows a higher hook rate than in-feed: the gap varies and must be recalculated on your own data. The same asset, served to the same audience, shows two different hook rates because the viewing environment differs. Comparing a Reels creative and an in-feed creative on a single table produces wrong decisions.

Two benchmarks to keep in mind when these figures seem low to you. Across all formats, Prooflytics (June 2026) puts the average thumb-stop between 25 and 35%, which amounts to saying that 65 to 75% of paid views stop before the third second. If the median 2-second view rate hovers around 27%, that mechanically implies that about seven impressions out of ten scroll past without stopping. A 30% hook rate is not half a failure; it is a good asset.

Hook rate alone validates nothing. It is read together with hold rate: Adlibrary (May 2026) sets a floor at 25%, a “good” at 30% and a scalability threshold at 35%+ on Meta, provided hold rate is 40 to 50%. Some players put forward hold floors around 60%; without a verifiable source, build your own internal reference. For control ratios beyond the hook, CTR, CPC, outbound click rate, go through the 3 ratios to measure on Meta Ads video.

That leaves the question of volume. A common order of magnitude is several tens of thousands of impressions per variant; the real volume depends on the difference you are trying to detect. A hook judged on 2,000 impressions is judged by chance. That is why we test 4 hooks and not 12 when the media budget is tight: better four reliable verdicts than twelve hunches.

For organic, read saves and shares too

On an organic account, 3-second retention remains the first indicator of the hook, but it does not tell you what the viewer understood of the promise. Three signals complete the reading, in TikTok Analytics as in Instagram Insights:

• Completion rate: an indirect indicator, it shows whether the tension opened by the hook pulled the viewer through to the end.
• Save rate: from the opening, the viewer perceived a useful promise they want to find again later. It is the typical signal of direct promise hooks.
• Share rate: the hook activated an identification lever strong enough for the viewer to think “that’s my friend”. It is the typical signal of emotional identification hooks.

To compare two openings in organic, keep everything else identical: same body, same length, same call to action. If one variant gains in retention but loses in saves, it caught the eye without getting the promise across.

Why can a high hook rate wreck your CPA?

A hook that stops everyone also attracts those who will never buy, and you pay for them. We observed, at a DTC client, a “mystery box” hook with strong retention and high CTR but no conversion at all (internal observation, unpublished). The creative passed every top-of-funnel check. It sold nothing.

The mechanism is well known and rarely named. A hook disconnected from the offer inflates hook rate, then CTR, because it creates a curiosity that the product page does not satisfy. Stackmatix (November 2026) describes the same chain: high hook rate + low hold rate = flattering CTR and rising CPA. You are buying curiosity clicks at the price of intent clicks.

Hence the rule we have applied with our e-commerce clients since 2025: the hook must name the product category or the problem solved before second 2. Not the brand, not the price, the category. “Serum”, “mattress”, “my gym membership”. It costs one or two points of hook rate and it brings CPA down, because those who stay know what is being talked about. We have stopped defending pure “pattern interrupt” openings (the slam, the scream, the falling object) on lower-funnel acquisition campaigns: they perform on the first ratio and nowhere else.

To isolate the weak link cleanly, hook, hold, click, conversion, we use the 4-stage creative measurement protocol. The order matters: you never touch the hook until the stage below has been checked. Rewriting a hook to fix a product page problem is repairing the front door of a house that leaks through the roof.

How much does a hook cost, and on which quote line?

An additional hook is an editing unit, not a shooting unit, and that is the whole point of the model. On a video body that is already edited and already validated, producing four extra openings means replacing the first two to three seconds, resyncing the on-screen text and re-exporting. On our quotes, this line falls in a range of €25 to €60 excl. VAT per hook when the rushes already exist, against several hundred euros as soon as a creator has to be called back to shoot a new shot. The gap explains why we refuse quotes where a “creative variant” and a “hook variant” are billed at the same price.

What that implies upstream. The shoot must plan for the hooks: five openings filmed back to back, same set, same light, same outfit. Ten extra minutes on set, not half a day. If your provider delivers a video with a single opening, you have bought one creative where you could have bought five entries into the same creative.

On the UGC creator side, three requirements to put in black and white in the brief: the number of hooks shot (3 minimum, 5 for a decent budget), the maximum length of each opening (2 seconds, timed, not “short”), and the matching constraint, same framing and same sentence starter, so that the edit can swap them without a visible jump cut. How to lock all this down is detailed in the 4 locks of a brief that works for a UGC creator.

A fair clarification: multiple hooks are not always profitable. On a campaign with less than €1,500 of monthly media, you will never get the tens of thousands of impressions per variant needed for a verdict. In that case, two hooks are enough, and the money saved goes into media. Multiplying openings on too small a budget produces pretty reporting lines and no decision.

What should you do when all your hooks pass and nothing sells?

If your hook rates exceed the thresholds and CPA does not budge, the problem is no longer in the first three seconds. And yet that is where 80% of teams keep looking, because it is the easiest part to change.

The diagnosis is made link by link, from the bottom up. Hold rate below 40% while hook rate is good: the hook’s promise is not kept between second 3 and second 10, you have a matching problem, not an opening problem. Decent hold rate and CTR under 1%: your video body informs but gives no reason to act now. CTR above 1% and CVR under 1.5%: the problem is on the page, the price or the promise; the video did its job. Improvado (2026) also offers a simple triage hierarchy: 6-second view rate ≥ 10%, CTR ≥ 1.0%, CVR ≥ 1.5%. Three thresholds, three verdicts, in that order.

That leaves the most uncomfortable case: everything passes, and the account doesn’t take off. Here it is no longer creative, it is unit economics. An average basket of €29 with a CPM of €14 and a CVR of 1.8% will never fund cold acquisition, whatever the quality of the hook. No opening with a 40% thumb-stop repairs an insufficient gross margin. We would rather say so before selling twenty variations.

One last thing, and it is the one we repeat most in steering meetings: knowing how to write a 3-second hook is useless if you only write one a month. The skill is not in the sentence, it is in the cadence. Ten hooks written on Monday, four shot on Wednesday, delivered on Friday, judged the following Friday on sufficient volume. The brands that win in 2026 are not those that write the best hooks, they are those that kill them the fastest.

Frequently asked questions

What is a good hook rate in 2026?

On Meta, a 30% hook rate is good and 35%+ allows you to scale, provided hold rate is 40 to 50% (Adlibrary, May 2026). On TikTok, the median 2-second view rate hovers around 27%, within a range of 20 to 35% (AdSights, July 2026). Meta publishes no official threshold: all these figures come from vendors.

How long should a hook last?

Two seconds of useful content, measured at three. The first spoken sentence is 6 to 9 words, about 2 seconds at a natural pace in French. TikTok counts its view at 2 seconds, Meta at 3 continuous seconds (or 97% of the video if it is shorter than 3 s). Writing longer amounts to writing after the viewer’s decision.

Do you need a different hook for each platform?

Rarely by platform, often by placement. Meta Reels beats Meta in-feed by about 7 points on the same 3 s threshold (TTS Vibes, September 2026): the same hook doesn’t read the same depending on the environment. The real adaptation concerns the on-screen text, which must avoid the interface zones, different across TikTok, Reels and Shorts.

How do you test several hooks without reshooting the video?

By filming 3 to 5 openings back to back on shoot day, same set and same framing, then swapping them in the edit on a single video body. The cost becomes a post-production line, around €25 to €60 excl. VAT per hook on our quotes when the rushes exist, against several hundred euros for a new shoot.

My hook rate is good but I’m not selling, why?

Look at hold rate first. A high hook rate combined with a low hold rate inflates CTR and pushes CPA up (Stackmatix, November 2026). A DTC case documented by Novapixel in October 2026 showed 70% retention past 3 s, 1.8% CTR and zero conversions: the hook promised something other than the product.

How many impressions before judging a hook?

Around 33,000 impressions per variant. With a 1.5% CTR and a 2% CVR, you need roughly 500 clicks to reach 95% significance (Improvado, December 2026). Below that, the ranking of your hooks is statistical noise. With a tight media budget, test 4 openings rather than 12.

Which type of hook should a B2B brand choose?

Start with narrative tension and direct promise: your target scrolls with a problem in mind. Naming a common mistake in the sector or a concrete result catches a decision-maker faster than a visual shock, which needs a context they don’t have time to interpret. Challenger authority, which attacks a received idea in the trade, also deserves a place in your first test.

Can you reuse a TikTok hook as is on Instagram Reels?

The mechanic transposes, the execution rarely does. TikTok tolerates a rawer, more direct opening; Reels, especially on an established brand account, expects a more polished first shot and a slightly calmer tone. Recycling without reworking frame 0 or the hook text is the mistake we see most. Keep the sentence, rework the image, check the interface zones.