Writing hooks that stop the scroll

Short-form feeds are zero-sum. Every second of watch-time your video earns is a second stolen from the next creator in the stack. The algorithm knows this — it measures hook-rate (viewers who watch past the 3-second mark) as a direct proxy for whether your opening line delivered on its implicit promise. A hook-rate above 40% is strong across TikTok, Reels, and Shorts. The average paid ad lands around 25–30%. The gap between those two numbers is almost entirely a writing problem.

The contract is simple: in the first line, tell the viewer one specific, believable thing that will happen to them if they keep watching. Not two things. Not a brand introduction. Not a disclaimer. One thing. The more specific that thing is, the more the viewer believes you can actually deliver it, and the more likely they are to stay past the 3-second mark.

The benchmark to write against is not your last ad. It is the five videos that will appear directly after yours in a cold-start feed. A viewer who's never seen your brand, scrolling at full speed, deciding in under a second whether to pause. Write for that person, not for the media buyer who already knows the product.

These six structures account for the majority of high-scoring hooks in the Ad Bench database. They work because each one creates a specific psychological reason to keep watching — curiosity, fear of loss, disagreement, social proof, urgency, or a concrete value promise.

• —Pattern interrupt. “You're wasting money on this every day.” Opens with an accusation or unexpected frame that forces the viewer to pause and self-assess. The key is specificity: “wasting money” lands harder than “making a mistake.”
• —Curiosity gap. “The reason your ads keep getting rejected.” Names a problem the viewer probably has, then withholds the answer. The gap between the named problem and the withheld fix is what keeps them watching. Works best when the problem is embarrassing or costly enough to feel urgent.
• —Controversy / hot take. “Hot take: UGC is dead.” A strong, arguable position that makes the viewer want to agree, disagree, or at least hear the argument. Do not soften it. “Hot take: UGC might be losing effectiveness” is not a hook; it is a hedge.
• —Social proof. “100K brands already do this.” Numbers create authority fast. The specific number matters more than the claim — “100K” beats “thousands of brands” because it reads as a real measurement rather than marketing language.
• —Problem-agitate-solution. “Your hook rate is 8%. Here's what 40% looks like.” Diagnoses the gap between where the viewer is and where they could be, then promises to close it. The benchmark numbers are load- bearing — they turn a vague pain into a quantified one.
• —Value promise. “In 30 seconds: the only 3 words that matter on a hook.” Bundles a time commitment (“30 seconds”), a scope limiter (“only 3”), and a specific deliverable. Works best for educational content where the viewer needs permission to invest their time.

The Ad Bench auto-flags four hook patterns as high-risk. Each one fails for a specific algorithmic reason, not just a creative one.

• —“Hey guys, today I want to tell you about…” This opener uses 7 words to say nothing. It names the format instead of delivering value. The Ad Bench scores it below 30 in the hook category because it sets a passive, low-urgency frame before the viewer has any reason to care. The 3-second window is already half gone.
• —“I can't believe I'm saying this but…” Overused to the point of pattern-fatigue. Feed-trained viewers have seen this prefix on thousands of videos — their brain classifies it as “creator-content filler” and scrolls past before the actual hook lands. What was a curiosity-gap mechanism two years ago is now a skip trigger.
• —Brand name + static logo on frame 1 with no movement. Signals “ad” before a single word is spoken. The algorithm's ad-detection signal fires; the viewer's skip reflex fires shortly after. Brand recall does not compensate for a lost 3-second window. Brand logos belong at second 5 or later, after the hook has paid off.
• —Slow zoom on a product with no text or spoken hook. A cinematic technique that works in a 30-second broadcast slot and fails in a 0.8-second scroll window. Slow visual movement reads as “nothing is happening yet” in a feed context. Without a written or spoken hook in the first 2 seconds, the viewer has no reason to wait for the zoom to resolve.

There are three hook types: spoken (the creator says the hook out loud), written (text-on-screen in the first frame), and pure visual (a pattern interrupt — an unusual action, object, or setting that stops the scroll without words). Each has a different reliability profile across platforms.

Sound-off viewing rates are roughly 85% on Facebook Feed, 60–65% on Instagram Feed and Reels, 40% on TikTok, and 30–35% on Shorts. A spoken hook with no on-screen text is invisible to more than half your Reels audience and nearly all of your Facebook audience. Written hooks do not require sound. That makes them the more reliable choice for cross-platform distribution — the same hook lands whether or not the viewer has audio on.

The highest-performing hooks usually combine all three layers: a strong spoken line, the same line (or a compressed version) as on-screen text, and a visual frame that reinforces the claim rather than contradicting it. The text reinforces the audio for sound-on viewers and carries the whole message for sound-off viewers. The visual prevents the “nothing is happening” problem that text-only hooks can produce on a static background.

If you can only do one, choose written. It reaches 100% of viewers regardless of context. If you have the budget for all three, layer them — the combined hook-rate lift over spoken-only is consistently measurable in the Ad Bench rubric's hook-rate scoring category.

The fastest way to understand hook principles is to see them applied to real weak openers. Three common patterns and how to fix them:

• —Weak: “Check out our new skincare line.” Rewrite: “My skin looked 10 years older until I changed one thing.” The original names the brand's intent (promote the line); the rewrite names the viewer's pain (aging skin) and promises a specific fix (one change). The viewer who has that pain will stay. The viewer who doesn't is not the target anyway.
• —Weak: “We're offering 20% off this weekend.” Rewrite: “Most people overpay for this by $200/year. Here's the fix.” The discount opener requires the viewer to already know and want the product. The rewrite works on cold audiences by quantifying the pain ($200/year) before revealing the solution. The sale becomes the fix, not the feature.
• —Weak: Product demo starting on frame 1, no spoken or written hook. Rewrite: Open with the problem statement as text-on-screen — “Every blender I've tried leaks. Until this one.” — then cut to the demo. The curiosity gap (“what blender?”) earns the 3 seconds needed for the demo to start paying off. Without it, the demo is just footage with no context for why the viewer should care.

In each case, the rewrite moves the opening line from brand-centric to viewer-centric. The brand's interest (promote, discount, demo) becomes the viewer's problem (aging, overpaying, leaking). That shift is the core of what The Ad Bench's hook score measures: does the first line serve the viewer or the brand? The algorithm does not care about the brand. The viewer decides in 0.8 seconds whether they do.