The Complete Guide to Video Hooks

By Bell Chen, founder. Updated May 18, 2026.

On March 21, 2024, Alex Hormozi posted a seven-line tweet that read, verbatim, "I lost $10k on my way to my first $100k. I lost $100k on my way to my first $1M. I lost $1M on my way to my first $10M. I lost $10M on my way to my first $100M. The bigger your goals, the more expensive the lessons. It's not a loss, it's the price of tuition." The first eleven words were the hook. The numbers escalated by a factor of ten each line, the structure repeated, and the payoff (the "tuition" reframe) landed only after the reader had already committed to the pattern. That tweet was not an opener attached to a video. It was the entire mechanism of a hook compressed into prose: a specific claim, a number, a structure the reader can feel before they finish reading, and a reason to keep going.

A video hook is the same machine in the first one to three seconds of a short. It is a promise the viewer can recognize before they decide whether to scroll. The hook is not "an attention-grabber." It is a specific image, claim, or sound that names the viewer's reason to stay. The strongest short-form hooks in 2024 to 2026 share four properties: a named noun (not "this" or "something"), a number or visible artifact (a price, a count, a face), a structure the audience can predict will pay off, and zero throat-clearing. The hook does the work the rest of the video pays off.

This page is for the creator who has shot a dozen Reels or TikToks, watched them die at 400 views, and suspects the first three seconds are the problem. The frameworks below come from named operators with verifiable artifacts. None of them are template-fill placeholders.

The three-second metric is real but it is not the only retention signal. TikTok's own creator documentation, summarized in Buffer's 2026 algorithm guide and corroborated by OpusClip's analysis of the platform's ranking signals, confirms that a viewer scrolling past a clip in under three seconds is the strongest negative signal the recommendation engine receives. Total watch time, completion rate, and rewatch ratio are the inputs the system uses to decide whether to widen distribution after that initial test, but the first three seconds are the gating event. If the hook does not hold, none of the downstream signals get a chance to fire.

Instagram's head Adam Mosseri described the equivalent ranking input for Reels in his mid-2024 statement on sends per reach, which was cross-confirmed by SMK's coverage of the same announcement. Mosseri said, "When you're creating content, think about creating something that people would want to send to a friend. Don't force it, but sends are one of the biggest signals we use in ranking and can help your reach over time." The reach mechanism on Reels keys off whether a viewer wants to share the post. The hook on Reels has to do two jobs at once: hold the viewer through the first three seconds and make the post send-worthy by the end. A hook that buys attention but produces nothing to share is half a hook.

Jenny Hoyos, who has shipped multiple YouTube Shorts past 100 million views per video and was profiled at length in Marketing Examined's short-form playbook, articulates the sound-off test. Hoyos said the hook "needs to be so good that you can be watching the video on mute and still know what it's about." The mute test cuts most "curiosity-gap" openers in half. If the video relies entirely on a voiceover saying "the one thing no one tells you," and the visual frame is the creator's face talking, the muted version is illegible. The first-frame visual has to do hook work on its own.

In my own audits of two hundred and change short-form clips across the last six months, the openers that hold a 30-something product manager (the most common buyer in the SaaS pages I look at) put a noun on screen in the first second. A price, a tool, a real person doing a real thing, a visible cost. The openers that die at second two are the ones that ask the viewer to picture a generic concept ("a wall," "a problem," "everything you know").

There are four reliable cognitive mechanisms hooks exploit. Treat them as named patterns with evidence rather than as a personality test.

Curiosity with a real noun, not a vague gap

The "curiosity gap" listicle taught creators to open with incomplete information. The problem is that the gap only works when the viewer can picture the missing piece. Hoyos's mute test names the constraint directly. "The one thing no one tells you" with no visual noun fails the test, because the viewer cannot picture the thing. A hook like "I lost $10k on my way to my first $100k" passes, because the dollar number is the visible noun and the viewer can picture the loss. Hormozi's tweet is a five-stage curiosity escalation where each line names a specific number the reader can hold.

Pattern interruption that fits the platform

The classic "open with silence" pattern interruption was a 2019 to 2022 trick on a feed that defaulted to music. On a 2026 feed where roughly half of viewers watch with sound and many platforms autoplay the previous video's audio for half a second, silence is itself a tell, and the viewers most likely to scroll past silence are exactly the buyers a B2B account is hunting for. The 2024 to 2026 version of pattern interruption is more often visual: an unexpected object on screen (Ramp's transparent glass box in Flatiron Plaza with Brian Baumgartner sitting inside), an unexpected scale (MrBeast's "$1 vs $100,000,000 Car!", which became one of the most-watched MrBeast uploads of its window), or an unexpected register (a corporate Notion account opening on a cluttered employee screen instead of a polished UI walkthrough).

Spiky point of view, not safe summary

Wes Kao's Rad Letters interview on developing a spiky point of view names the move every strong cofounder-led hook makes. Kao said a spiky POV is "a perspective others can disagree with" and "an assertion you're willing to advocate for with conviction." The hook that opens with "Cold email is dead" fails the test because no one in the audience disagrees anymore. The hook that opens with "Cold email is not dead, the 'personalization is everything' crowd just refuses to send volume" passes the test because it picks a fight someone in the audience holds the opposite view on. Spiky POV is the structural reason cofounder-led hooks on TikTok and YouTube outperform anonymous talking-head content. The face says "I will own this take." The take has to be a take.

Five-second moment of change

Shaan Puri, whose podcast My First Million is downloaded by tens of millions of listeners annually, packaged his story-structure principle into a YouTube Short titled "The 5-Second Moment of Change". David Perell quoted Puri's articulation of the principle verbatim in a March 2024 X post: "A story is a five second moment of change. A story is not a sequence of events; it's about [continued]." The Puri move is to identify the five seconds in your idea where something turns (a price collapses, a discovery lands, a character flips) and to build the hook around that turn. The mistake is opening with the sequence (the setup before the turn) instead of with the turn itself.

The pattern below maps real hooks shipped by real accounts to the cognitive mechanism each one exploits. Each row is a verifiable artifact, not a template.

The escalation series

Hormozi's $10k, $100k, $1M, $10M, $100M escalation tweet is the cleanest published version. The structural move is to repeat a sentence shape with one variable escalating across the lines. The viewer is rewarded for predicting the next line and then surprised by the payoff that reframes the entire sequence ("It's not a loss, it's the price of tuition"). The format ports cleanly to short-form video as a series of cuts with on-screen numbers. The mistake is escalating without a payoff that recontextualizes the series. An escalation that ends with a generic "and that's why you should never give up" wastes the structure.

The visible-cost reveal

Ramp's "Brian's Office" stunt remains the cleanest 2025 example. Per Rachel Karten's Link in Bio breakdown, the campaign generated roughly 112 million cross-platform views and the Andy Buckley cameo cut at @ramp.com/video/7561836281752194334 carried 181.9K likes and 600 comments. Kendall Hope Tucker, Ramp's Head of Creative Experimentation, told Marketing Brew the strategic logic: "Accountants have been using the same software for 30 years, they're not looking for alternatives. So we're like, 'How do we make that pain feel visceral?'" The hook was the image, not the claim. Brian Baumgartner in a glass box surrounded by paper receipts is the cost of legacy expense management made visible. The viewer does the math.

The mute-test workflow timelapse

Notion's TikTok account (@notionhq, 142K followers) runs versions of this. The clip at @notionhq/video/7203529954016087342 carries 59.2K likes and 673 comments; the opening frame is a messy employee screen with chaotic meeting notes, and the cut to the cleaned-up Notion AI version arrives by second three. Lexie Barnhorn, Notion's then-Head of Influencer Marketing, told CreatorIQ's Earned podcast Ep. 98 the discipline behind the format: "When you have a set-in-stone strategy, you're setting yourself up to fail, because [TikTok] changes every single day." The mute-test passes because the first frame is a recognizable problem (messy notes) and the cut is a recognizable solution (clean Notion page). No voiceover is required.

The cofounder spiky-POV rant

Cluely's office series is the cleanest sustained example. Per VideoToolkit's breakdown of Cluely's 1 billion views in 3 months, Roy Lee's strategy reduces to a single line he is on record saying: "If half the audience doesn't hate it, it's not viral enough." The Cluely Ep. 1 office-series clip at @cluely/video/7547507694014369031 carries 380.2K likes and 1,592 comments. The hook in each clip is Roy Lee saying something the average B2B audience instinctively flinches at, then earning the next twenty seconds by being right enough to make the flinch land. The format requires a real take. A founder who hedges in private cannot run this hook, because the audience reads the hedge through the camera.

The escalation-of-scale visual

MrBeast's "$1 vs $100,000,000 Car!" (X post and YouTube upload, @MrBeast on X status 1747044525116108854) is the visual cousin of the Hormozi escalation tweet. The hook is the title plus the first-frame wide shot of the $1 car. The audience commits to watching the entire range before the upload has done any explaining. The structural lesson is that a scaled comparison ($1 to $100M, 1 employee to 100, 1 customer to 1 million) gives the viewer the entire arc of the video in the first three seconds, and the rest of the video is just the payoff. The mistake is announcing the range and then not paying off both ends.

The five-second moment of change cold open

Puri's YouTube Short on the five-second moment of change is the meta-example. The hook is Puri himself naming the principle in the first sentence, which is a hook about hooks. The transferable move is to find the five-second turn in your idea (the moment the customer realizes the price collapses, the moment the founder admits the mistake, the moment the data flips) and open on the turn rather than on the setup. The setup can come back in seconds five through twenty. The viewer needs the turn first.

Sound design is half the hook, but most of the audio advice in circulation is generic. Three rules that have held up across the audits I have run:

The first sound the viewer hears decides whether the muted version of your video and the audio version of your video are doing the same work. If the audio is "subscribe for more," the audio version is doing nothing the visual was not already doing, and the platform's autoplay carry-over from the previous video has already eaten the sound budget. Use the first audio second to plant the spiky POV, the price, or the name of the person speaking. Then the visual can do the cost-of-status-quo work in parallel.

Trending sounds are a distribution accelerant, not a hook. On TikTok specifically, a trending sound buys a distribution test from the algorithm, per the standard platform documentation summarized in Buffer's 2026 guide, but the hook still has to do its own work in the visual frame. The accounts that lean on trending sounds without a strong visual hook ship clips that test well in the first 200 views and die at the second distribution wave.

Silence as a pattern interrupt has decayed since 2022. Half a beat of silence inside a strong audio track still works as a punctuation tool, but a full opening second of silence reads in 2026 as either a recording mistake or a desperate hook attempt, and the average viewer interprets it as a reason to scroll. The exception is when the silence is paired with an extreme visual (Ramp's glass-box opening frame is a silent second of an absurd image, and the silence works because the visual is doing all the work).

Two patterns have aged into anti-patterns over the last 24 months.

The abstract POV opener ("POV: when your startup hits a wall") works in niches where the audience can picture the setting in under a second. Gym, kitchen, dating, parenting all qualify. Most SaaS and B2B niches do not, because the viewer cannot picture "a wall" any more concretely than "a problem." In a SaaS audit I ran on a Reels account whose owner had shipped twenty POV openers in a row, the average three-second retention was below 28%, and the three clips that broke that pattern (a named price, a named customer, a named tool) all cleared 50%.

The "AI startup" intro that worked in 2023 has commoditized. By mid-2026 it is the most crowded opener on every platform. When a category name stops doing hook work, the brand has to name a specific take or a specific cost. "We are an AI startup" loses the viewer. "We let GPT-5 write our onboarding emails for a month and 31% of the replies were from prospects asking why our copy got worse" holds them.

The published advice on hook testing usually reduces to "A/B test your openers." That is correct but lossy. The version that works in practice has four steps.

Shoot three openers for the same script before you shoot the rest of the video. The cost of three openers is fifteen extra minutes on the shoot. The cost of finding out the original opener was the weak link is a week. Cut a finished version with each opener, post them at different times across different days, and look at the three-second hold rate.

Look at retention curves, not just total views. Platforms publish retention graphs for a reason. If the drop happens in the first three seconds, the hook is the problem. If the drop happens at second eight, the bridge from hook to payoff is the problem. If the drop happens at second twenty-five, the payoff is the problem. Each of these is a different fix and confusing them wastes the next ten shoots.

Audit the openers in your own competitive set. Pick one account adjacent to yours on the platform, watch their last ten posts back-to-back, and write down the hook for each. The pattern you spot inside your own niche will outperform any generic listicle, including this one. Hoyos, on the Marketing Examined playbook, makes the same point about her own process: she will throw away an entire idea if the hook is not strong enough to hold the mute test.

Treat the hook as one editorial decision, not as a separate craft. The hook fails when it is bolted onto a script that does not match its energy. A hook that promises a number has to pay off the number. A hook that promises a spicy take has to deliver the take. The hook is the contract with the viewer, and breaking the contract costs more than a weak hook would have.

TikTok rewards a visible noun in the first frame and a sound that does not contradict the visual. The clips that perform on the platform in 2026 are the ones that pass the mute test and the autoplay-carry-over test simultaneously.

YouTube Shorts has a slightly higher bar for the bridge between hook and payoff. The Shorts audience, per Hoyos's track record and vidIQ's analysis of her 10M-views-per-Short process, responds to hooks that promise a specific number or a specific result and then deliver it cleanly inside 30 to 60 seconds. The mistake is shooting a 60-second Short with a 12-second-Short hook.

Instagram Reels keys off sends-per-reach as a primary distribution signal, per Mosseri's verbatim 2024 statement above. The hook has to do two jobs: hold the viewer past three seconds, and produce a post the viewer wants to share by the end. A hook that buys attention and produces nothing sendable underperforms a slightly weaker hook that produces a quotable line. The sendable-line constraint is why Reels rewards quotable takes more than TikTok does.

Before you shoot, write the first three seconds in three different ways. Run the mute test on each (would the visual alone tell the viewer what the video is about). Pick the version with the most specific noun on screen. Shoot all three. Post the best after looking at the openers on the three accounts in your niche you most respect. Watch the three-second retention curve. If the drop is at second three, the hook is wrong. If the drop is at second eight, the bridge is wrong. Iterate on the layer that broke. (If you want a second pass at the hook itself, the brand-profile analysis Superdirector runs surfaces the hook-pattern density across a competitive set; one option among several.)

What counts as a "hook" in 2026?

The first one to three seconds of a short-form clip, framed as a promise the viewer can recognize before deciding whether to scroll. Hormozi's March 21, 2024 escalation tweet is the prose equivalent: a specific noun, a number, a structure the reader can feel before they finish reading, and a reason to stay.

Does the three-second rule still apply?

Yes for the gating decision, no for the full retention story. Per Buffer's 2026 TikTok algorithm guide, a viewer scrolling past in under three seconds is the strongest negative ranking signal. Past that mark, total watch time, completion, and rewatch carry the rest of the distribution decision. The hook decides whether the rest of the signals get a chance to fire.

Should I be on camera as the founder for the hook?

Depends on the pattern. The cofounder-led spiky-POV rant (the Cluely shape) requires it. The workflow timelapse (the Notion shape) does not. The wasted-spend reveal (the Ramp shape) explicitly avoids it. The mistake is defaulting to founder face for every hook regardless of which pattern the script needs.

What is the single biggest mistake in hook writing?

Opening with a noun the viewer cannot picture. "POV: when your startup hits a wall" fails because no viewer has a default mental image for "a wall." Hoyos's sound-off mute test catches this: if the muted version of the hook is illegible, the hook is doing nothing the rest of the video could not do.

How many hooks should I test per video?

Three is the floor. Three openers cost an extra fifteen minutes on the shoot and they expose the weak version of the script before it costs a week of distribution. The accounts I have audited that lift retention reliably are the ones that shoot the openers in batches and pick the survivor.

Should I use a tool to grade my hooks?

You can, but the failure mode is grading the hook in isolation when the actual problem is the bridge or the payoff. Tools that grade hooks score the opener; they do not tell you whether the next 40 seconds deliver on it. The retention curve does.