How to Write Scroll-Stopping Hooks That Keep Viewers Past 3 Seconds

On January 8, 2025, Adam Mosseri posted a Reel on @mosseri (instagram.com) naming the three signals that decide which Reels get unconnected reach. "Watch time, likes per reach, and sends per reach," per Mosseri, with sends carrying the most weight for unconnected reach. The order matters more than most operators read it. If the first three seconds collapse, watch time never accrues, the likes-per-reach denominator never fills, and the sends signal never has a chance to fire. The hook is not a creative flourish; it is the prerequisite that determines whether the rest of the post is allowed to compete.

Jenny Hoyos, who has crossed 6 million subscribers on YouTube primarily through Shorts and is the most-studied short-form retention operator working today, told Marketing Examined (marketingexamined.com) that she rewrites her opening line until the but and the then arrive inside the first sentence. "I went to Chick-fil-A but they were closed, then a man walked up to me," per Hoyos, is the verbatim example she gave. That is what a working hook looks like. This page is the procedure for writing them on purpose, the test rig for separating which channel is failing, and the diagnostic for the harder problem, which is realising the hook is not the bottleneck and the script is.

What a hook actually has to do in 2026

A hook in 2026 does three jobs in roughly 1.5 seconds. It has to interrupt the scroll, register a specific promise, and earn the next 1.5 seconds. The interrupt is visual. The promise is verbal or textual. The earn is structural; it depends on whether the viewer believes the payoff is specific enough to be worth the wait. The order is fixed.

The platform-side reason it is fixed is the one Mosseri named on January 8, 2025. The verbatim line from his Reel (instagram.com), as transcribed in Buffer's March 24, 2026 algorithm update (buffer.com) by Shivani Shah, is the three-signal frame: watch time, likes per reach, and sends per reach. The Socialinsider 2024 benchmark report (socialinsider.io), built on more than 35 million Instagram posts across 447,613 active accounts from January through December 2024, recorded a 47.46% average Reels completion rate for accounts under 5,000 followers and 39.74% for accounts above 200,000. The median Reel does not finish. The opening seconds are where the failure happens.

The category-wide reason is competition density. Buffer's 2026 State of Social Media Engagement report (buffer.com), which analysed 52 million posts across ten platforms, found that Reels still distribute 2.25 times wider than single-image posts on the same account. That is the upside the format pays out when the hook works. The downside, in the same Buffer dataset, is that median engagement rates dropped 24% year over year. Wider distribution on a working post, lower floor on a failing one.

Hooks do not just decide first-three-second retention. They decide what the algorithm thinks the post is about. Paddy Galloway, who consults for MrBeast, Logan Paul, and Mike Tyson and is among the most-cited short-form strategists, told the Colin and Samir interview in 2024 (youtube.com) that "the opening seconds are how the algorithm decides what your video is about," per Galloway, and how it assigns the post to a topic cluster for distribution. A hook that misnames the topic gets shown to the wrong viewers, who do not finish, which kills retention, which kills distribution. The hook is not just retention. It is targeting.

Step by step

1. 01

   Step 1. Pull the 3-second retention curve on your last 20 posts (15 minutes)

   Open Instagram's Professional Dashboard or TikTok's Creator Center analytics tab. For each of the last 20 posts, write down the 3-second retention number. Group them into above-60% and below-60%. The threshold is not a published Instagram number; it is the working benchmark senior operators converge on, and it tracks the 47.46% / 39.74% completion-rate split Socialinsider 2024 (socialinsider.io) recorded across the same window. If more than half your posts are below 50%, the hook is the bottleneck and the rest of the procedure applies cleanly.

   Deliverable

   A 20-row tracking sheet with 3-second retention bucketed above and below 60 percent.

2. 02

   Step 2. Identify the two posts in your top-five and bottom-five that share a topic (15 minutes)

   If one of your top-five posts is a recipe Reel and one of your bottom-five is also a recipe Reel, the topic is controlled and the variable is the opening. Compare them side by side. Watch the first 3 seconds of each on mute (the Hoyos test). What did the top one do in frame 1 that the bottom one did not? The answer is usually one of three things: a face on screen versus a B-roll opener, a specific noun on screen versus a generic establishing shot, or a movement-anchored first frame versus a static one.

   Deliverable

   A two-row controlled comparison naming the single first-frame difference between the matched top and bottom posts.

3. 03

   Step 3. Rewrite the failing opener using the three-channel rule (20 to 30 minutes)

   A hook runs on three channels in sync. VISUAL is what the viewer sees in frame 1. AUDIO is the first words or sound. TEXT is the overlay that anchors the promise. The visual channel goes first: open with a specific noun on screen, ideally a face, a product, a number, or a recognisable workspace. The audio channel goes second; the first word should be mid-sentence or specifically unexpected, never "Hey guys". The text channel goes third; the overlay should be 4 to 8 words and visible inside the first frame, not fading in over the first second. The Hoyos but/then structure is the cleanest verbal pattern. The Hormozi catalyst-sentence pattern is the cleanest declarative pattern. Pick one. Write 8 to 12 versions. Keep the one that passes the mute test.

   Deliverable

   Eight to twelve hook variants in writing, one selected via the mute test.

4. 04

   Step 4. Build the testing rig (10 minutes the first time, 0 minutes after)

   Set up a tracking sheet with one row per post and columns for hook style, first-frame description, opening line verbatim, 3-second retention, completion rate, and send count. The sheet is the swipe file Hormozi and Hoyos both describe. After 30 to 40 posts, the patterns in the sheet are personal. They are not generic. "Catalyst sentences land 8% better than but/then on my product audience" is the kind of finding the sheet produces, and that finding is the entire point of running the rig.

   Deliverable

   A six-column tracking sheet template ready to log every shipped post.

5. 05

   Step 5. Cut the first 1.5 seconds of three under-delivered posts and re-upload (15 minutes)

   This is the most under-used diagnostic in short-form audit work. If reach lifts on the re-uploads, the opener was the bottleneck and the rest of the post was fine. If reach stays flat, the opener was not the problem and the script or topic is, which means the audit you need next is on the catalyst content of the post, not the hook. I have watched this single test resolve more "I do not know what is wrong with my reach" conversations than any other diagnostic. It is the cleanest counterfactual short-form has.

   Deliverable

   Three re-uploaded posts with the original 1.5 seconds trimmed, plus a 7-day reach comparison row.

6. 06

   Step 6. Roll the winning patterns into a templated library (ongoing)

   After 4 to 6 weeks of running the rig, you will have 5 to 10 hook patterns that consistently land for your specific audience. Document them as fill-in-the-blank templates. "[Surprising number] [things] that [authority figure] [never tells you]" is the canonical example. "I tried [thing] for [time period] and the third one [surprising outcome]" is another. Templates are starting points, not finishers. Refresh the library monthly; retire the patterns that landed three to five times in a row, on the working theory that the algorithm pattern-matches and discounts repeated openings on the same account. (A hook analyser that surfaces the patterns winning on competitor accounts in your category gives you a starting set for the library rather than discovering them from your own data first; one option among several.)

   Deliverable

   A 5-to-10 entry template library with each pattern dated and reach-tagged.

The named-operator playbook

Jenny Hoyos. Hoyos's Marketing Examined playbook (marketingexamined.com) names the rule that decides whether her hook ships. She calls it the mute test. If the first frame, played on mute, does not communicate the topic and the stakes, the hook is broken before any audio fires. She rewrites the opening line until the but and the then arrive inside the first sentence. The verbatim Chick-fil-A line is the canonical example: "I went to Chick-fil-A but they were closed, then a man walked up to me," per Hoyos. The but introduces a problem in 0.8 seconds; the then opens a curiosity loop that the viewer cannot resolve without watching. Her vidIQ creator profile (vidiq.com) confirms the method is the load-bearing reason her channel grew past 6 million subscribers primarily through Shorts retention.

Alex Hormozi. Hormozi, whose Acquisition.com clips routinely cross 10 million views on short-form, runs what Opus Clip's breakdown (opus.pro) calls the catalyst-sentence pattern. The opening line is a single declarative claim that contains its own stakes. "Most founders die on this one mistake" is the structural shape Hormozi uses repeatedly across his clip library. He told Lex Fridman in their 2024 conversation (youtube.com) that he rewrites his opening sentence 8 to 15 times before recording, sometimes more. The volume is the work.

Roy Lee. Lee, who runs the Cluely office series that has become the most-studied short-form brand experiment of 2026, opens Episode 1 of the series (instagram.com) with a cold cut on a face talking mid-sentence in a recognisable office setting. The post hit 380.2K likes and 1,592 comments, per the visible counts on the live post. Lee told VideoToolkit's breakdown of the series (videotoolkit.io) that the first-frame rule for the series is no logos, no intro card, no scene-setting.

MrBeast. The leaked How to Succeed in MrBeast Production handbook (simonwillison.net), a 36-page internal document that surfaced in September 2024, names the rule for the production team: "The first minute is the most important minute of each video," per the leaked document. Every opening is rewritten until the hook is in the first 10 seconds, the stakes are clear, and the visual on screen matches the verbal claim.

Where this typically breaks

The script does not have a payoff specific enough for any opener to sell. This is the most common failure and the hardest to admit. If the post is "here are 5 social media tips," no hook will save it, because the payoff is generic by construction. Hoyos's mute test catches this version of the failure inside the first frame. The fix is not on the opener; it is on the topic.

The visual hook is fighting the verbal hook. A common pattern in scripted Reels is a face talking to camera in a static frame while the voiceover is delivering a curiosity gap. The visual is doing nothing; the audio is doing all the work. The mute test fails immediately. The fix is to put the catalyst on screen visually as well as verbally. If the line is "most founders die on this one mistake," the visual should be the mistake (a redacted email, a crossed-out diagram, a face mid-grimace) not a neutral B-roll of an office.

The opening is a logo, an intro card, or a "Hey guys." Lee's no-intro rule on the Cluely series is the cleanest expression of the fix. The viewer enters the scene already in progress. Hoyos's vidIQ profile (vidiq.com) documents her version of the same rule. The hook starts at the moment of action, never at the moment before action.

The send beat is missing, so the hook works but distribution still drops off. Watch time clears the floor, completion rate clears the floor, and the post still under-distributes. The diagnostic is sends per reach. Kendall Hope Tucker, who ran Ramp's Brian's Office campaign and was profiled in Marketing Brew's October 22, 2025 piece (marketingbrew.com) by Phoebe Bain, told the publication that the campaign crossed 112 million cross-platform views in part because the scripts ended with a specific named referent who would forward it. The Andy Buckley cameo cut hit 181.9K likes. The hook earned the open; the closing send-beat earned the distribution. If your hook is strong and your reach is still flat, the failure is at the close, not the open.

Metrics to track

3-second retention. Target: 60% or above on unconnected reach. Below 50%, the hook is the bottleneck and Steps 2 to 5 of the procedure apply.

Completion rate. Target: 30 to 45% on a working hook for short clips, calibrated to the Socialinsider 2024 47.46% / 39.74% bands. Falls steeply between second 3 and second 8 when the script does not pay off the hook's promise.

Sends per reach. Target: 0.5% on Instagram for clips with a designed send beat. Per Mosseri's January 8, 2025 signal hierarchy, sends-per-reach is the load-bearing signal for unconnected reach. Below 0.2%, the close is not earning the share.

Hook rewrite volume per piece. Target: 8 to 15 variants per recording, per the Hormozi published practice. Below 8 means the opening has not been stress-tested.

Re-upload lift on trimmed-opener test. Binary: did reach lift on the re-upload? Yes confirms the opener was the bottleneck. No confirms the failure is downstream of the hook.

Where a planning-first tool fits

The procedure above runs in a sheet, the platform-native analytics, and 90 minutes of focused rewriting per piece. The point where most operators ask for tool help is in surfacing which hook patterns are winning on adjacent competitor accounts, because the manual version (pull 20 to 30 competitor posts, label by hook pattern, identify the repeating archetypes) costs roughly six hours per category audit. Superdirector's hook analyser collapses that step to about one hour by indexing public competitor posts and surfacing format archetypes by niche. One option among several alongside Foreplay's ad library, Crayon, or a hand-built Notion archive of competitor posts. The Hoyos mute test, the Hormozi 8-to-15-variant rewrite, and the re-upload counterfactual are the load-bearing diagnostics regardless of toolchain.

Disclosure by Bell Chen, founder of Superdirector: the brand-profile, hook-analysis, and script-generation features mentioned in this piece are part of the product I build. Methodology and benchmarks here are sourced from the linked platform documentation, industry reports, and named-creator interviews; treat the tooling note as one input among several.

Frequently asked questions