What Is a Good Hook Rate on Meta and TikTok Ads?

Short answer: 30% or above on Meta, according to Taylor Sicard’s 2026 Meta creative benchmarks, which is a top-quartile reading rather than a typical one, because the two largest published 2026 panels put the middle of the distribution at 20.9% and 25.44%.

On TikTok there is no comparable number, because TikTok does not measure the same thing. Hook rate is the most-quoted creative metric in paid social and one of the most misread. It is useful, it is genuinely diagnostic, and it tells you considerably less than most accounts act like it does.

What is a good hook rate for Meta ads?

A good hook rate for Meta ads is 30% or above, the working 2026 benchmark according to Taylor Sicard’s Meta creative benchmarks for 2026, with elite creative at 35 to 45%. Below 25% needs attention. Below 20% the creative is effectively invisible, and no amount of quality further into the video will rescue it.

Hook rate	Read
35-45%	Elite. This creative is doing its job.
30-35%	Healthy. The benchmark band.
25-30%	Functional, improvable.
20-25%	Fix zone for an established account.
Below 20%	Effectively invisible.

One qualification belongs with those bands before you use them, and the section below gives it in full: 30% is a good target and not an average, because the measured median on the largest 2026 panels is nearer 21 to 25%, per Biddyco and Billo. An ad at 22% is behind the target and in the middle of the market, which are two different conversations.

Hook rate is 3-second video plays divided by impressions. Thumbstop rate is the identical calculation under a different name, which causes more confusion in reporting than it has any right to. Different tools and agencies picked different labels for the same number, so if two dashboards disagree, check the definition before you check the creative.

Work it through by hand once and the ambiguity goes away.

An ad served 240,000 impressions that recorded 76,800 three-second video plays has a hook rate of 32%, which is 76,800 divided by 240,000. Meta counts a 3-second video play only when the video actually plays for three seconds, and replays inside a single impression are excluded.

Build it in Ads Manager from those two columns rather than trusting a pre-computed figure, because tools differ on whether the denominator is impressions or reach. Reach is always the smaller number, so a reach-based hook rate reads higher on the same asset without anything having improved.

What the metric captures is narrow but real: of the people the platform showed this to, what share stopped long enough to register it. That is a measurement of your first three seconds and nothing else. Not your offer, not your product, not your landing page, not your price.

What is the median hook rate on Meta, and is 30% actually normal?

The median hook rate on Meta is lower than the benchmark almost everyone quotes. Two published 2026 panels put the middle of the distribution between 20.9% and 25.44%, one a median and one a mean, according to Biddyco and Billo, which places the familiar 30% target around the top quartile rather than at the centre of the market.

Biddyco’s 2026 benchmark study measured 479 Meta video ads from 17 US ad accounts carrying $10.1 million of spend between June and August 2026, and found a median hook rate of 20.9%, with the middle half of ads between 14% and 30% and the top 10% at 45% or above, according to its hook rate and hold rate benchmarks for Meta ads.

Billo’s H1 2026 benchmark is far larger and lands in the same region: a cross-industry average of 25.44%, up from 24.42% in the second half of 2025, across 88,329 sales-objective Meta video ads with more than 1,000 impressions each in 14 industries between January and June 2026, per its H1 2026 video ad benchmarks and the September 2026 report of its findings.

The strongest categories were Toys and Games at 28.8% and business-to-business video at 28.7%, per the same Billo benchmark, which is the finding the trade press picked up, because business-to-business creative is supposed to be the boring end of the market.

Published 2026 figure	What it is	Sample behind it
20.9% median, 14% to 30% middle half, 45%+ top decile	a distribution of measured ads	479 video ads, 17 US accounts, $10.1M of spend, June to August 2026, per Biddyco
25.44% average, up from 24.42%	a cross-industry mean	88,329 sales-objective ads, 14 industries, January to June 2026, per Billo
30%+ working benchmark, 35-45% elite	a target for strong creative	no disclosed sample, per Taylor Sicard
18-28% in-feed, 24-36% Reels	placement bands on the three-second view	no disclosed sample, per AdSights

Those four figures reconcile rather than contradict each other once you notice that two of them are distributions and two are targets.

A 30% reading sits at roughly the 75th percentile of the measured spread per Biddyco, and about five points above the mean per Billo, while the 35 to 45% elite band lines up with Biddyco’s top decile at 45% and above.

So keep the bands at the top of this page as a target, and stop reading a 22% as a failure, because per the samples in the table above it is an ordinary ad that happens to be behind a demanding benchmark.

One result in the Biddyco data is worth more than the medians.

It reports that the ads Meta spent the most on did not have higher hook or hold rates than the lower-spend ads in the same accounts, which means the platform’s own budget allocation does not track the metric most teams use to predict winners.

Treat hook rate as a diagnostic for a weak opening, which is what Biddyco concludes, and not as a forecast of which ad will scale.

What is a good hook rate on TikTok?

A good hook rate on TikTok cannot be read off the Meta number, because TikTok does not report a three-second play at all. Its video play reporting counts 2-second video views and 6-second video views, so the TikTok version of the metric is 2-second video views divided by impressions.

The definitions are stricter than most dashboards imply. TikTok’s video play metrics documentation counts a 2-second video view when the video played for at least two seconds within one impression, and replays do not count. If someone swipes away and comes back, that is two impressions and two separate views rather than one continued view.

That missing second changes the number. The same asset clears a two-second bar more often than a three-second one, so a library imported from Meta will look better on TikTok before anyone touches the creative. An account that sets a 30% target from a Meta benchmark and reports 34% on TikTok has measured a definition, not an improvement.

The closer analogue to hold rate is the 6-second video view, with one caveat. TikTok counts one when the video played for at least six seconds, or was watched in full if it is shorter than six seconds, or received an engagement within the first six seconds.

That last clause means it is not purely a watch-time measure, so an ad collecting early likes can post a 6-second rate that overstates how much of it anybody watched.

So compare TikTok to TikTok. Hold a separate median for each platform, each placement and each audience temperature, and treat the two platforms as two measurement systems that happen to share a vocabulary. If you are still deciding where the budget goes in the first place, that is a different question with a different answer: TikTok ads versus Meta ads for a consumer brand.

What is hold rate and why does it matter more than people think?

Hold rate measures whether the people who stopped actually stayed. Using Meta’s ThruPlay definition, which is ThruPlays divided by 3-second plays, 20 to 25% is healthy on cold traffic, according to AdLibrary’s 2026 hold rate benchmarks.

A definitional trap sits inside that. Meta counts a ThruPlay when someone watches at least 15 continuous seconds, or watches the whole video if it runs shorter than 15 seconds.

Hold rate on a 12-second ad is therefore a completion rate, and hold rate on a 45-second ad is the rate of reaching the 15-second mark. Those are different questions, and comparing them across a mixed-length library is a reliable way to conclude that short creative holds better when what actually changed was the definition.

The relationship worth internalising: hook rate gets you into the auction, hold rate largely determines what you pay in it. Platforms reward completion with cheaper distribution, so two ads with identical 32% hook rates and different hold rates will not cost the same. One is teaching the algorithm that its audience was well matched. The other is teaching it the opposite.

Reading the pair together is far more useful than reading either alone:

• High hook, low hold. The opening promised something the video did not deliver, usually a hook disconnected from the content behind it. Common with borrowed formats: the hook was lifted from a video about something else.
• Low hook, high hold. The content is good and nobody is seeing it. Usually the cheapest problem in the account to fix, because the asset already works for everyone who gets there. Rewrite the first three seconds and leave the rest.
• Low hook, low hold. Retire it. There is nothing here to diagnose.
• High hook, high hold. Your scale candidate.

Most accounts track hook rate and ignore hold rate, which means they can identify that an ad failed but not where. The two-metric read tells you which part to rewrite, and that is the difference between iterating and guessing.

Which hold rate are you measuring?

Four different hold rates are in circulation on Meta, they are built from different columns, and a published hold rate benchmark quoted without its denominator cannot be compared against your account at all. This is the reason two people can both be right while disagreeing by twenty points.

What it is called	How it is built	What it answers	Published band
Hook rate, or thumbstop rate	3-second plays divided by impressions	did anybody stop	30%+ on Meta, per Taylor Sicard
Hold rate, 15s over 3s	15-second views divided by 3-second views	did the body keep the people who stopped	about 15 to 25% on cold prospecting, per AdSights
Hold rate, ThruPlay over 3s	ThruPlays divided by 3-second plays	the same question on Meta’s own ThruPlay event	20 to 25% on cold traffic, per AdLibrary
ThruPlay rate	ThruPlays divided by video plays	hook strength and hold collapsed into one number	roughly 18 to 28% on 15 to 30 second prospecting creative in Reels and Stories, per AdSights
Hold rate, ThruPlay over impressions	ThruPlays divided by impressions	what share of everyone shown the ad reached a ThruPlay	median 4.6%, middle half 3 to 7%, per Biddyco

The middle two rows agree with each other on long creative and part company on short. Meta counts a ThruPlay at 15 continuous seconds, or at a complete view if the video runs shorter than that, so on a 30-second ad the ThruPlay version and the 15-second version are asking the same question.

On a 12-second ad the ThruPlay version has quietly become a completion rate, and completion rates on short creative run far higher. A library of 10 to 45 second assets reported on one hold rate column is therefore ranking video length as much as retention.

The ThruPlay rate row is the one that causes arguments, because it is a different shape of metric. AdSights sets ThruPlay rate against video plays rather than 3-second plays, which folds the hook back into the number: a weak opening drags it down even when the body holds everyone who stayed.

It works as a one-line summary of a video and fails as a guide to whether the first three seconds or the middle needs rewriting.

The last row is the widest departure of the four conventions, and it is the one most likely to reach you from a study rather than from an account.

Biddyco measures hold rate against impressions rather than against three-second plays, which gives a median of 4.6% with the middle half between 3 and 7% on its panel of 479 Meta video ads, per its 2026 benchmarks.

That number is not a worse version of the 20 to 25% band per AdLibrary, it is a different fraction of a different denominator, and the two get quoted against each other in the same meeting.

How far apart do the hold rates read on the same ad?

The hold rates read anywhere between 7.5% and 25% on one unchanged asset, per the four conventions in the table above, which is the clearest argument there is for writing the denominator next to the number. Work a single ad through them and the spread is hard to miss.

Take a 30-second asset carrying the four columns AdSights says to add: 200,000 impressions, 180,000 video plays, 60,000 three-second plays and 15,000 ThruPlays. Hook rate is 30%, comfortably at the 2026 Meta benchmark and at roughly the 75th percentile of Biddyco’s measured distribution.

Hold rate on the ThruPlay convention is 25%, at the top of AdLibrary’s healthy band. ThruPlay rate is 8.3%, well under AdSights’ 18 to 28% band.

Hold rate on the impressions denominator is 7.5%, which is 15,000 ThruPlays over 200,000 impressions, above the 4.6% median per Biddyco. One asset, four verdicts, and two of them look like failures while the other two look fine.

So build the columns yourself rather than trusting a pre-computed field. AdSights’ guide to reading video metrics in Ads Manager says to add impressions, video plays, 3-second video views and ThruPlays as columns and compute the three ratios from them, segmented by placement and by creative length.

Ten minutes of column setup removes an entire category of reporting dispute, and it makes the benchmark you quote checkable by the person you are quoting it to.

One habit worth adopting in reporting: write the ratio next to the number, as “hold 24% on ThruPlays over 3-second plays”, which is the convention behind AdLibrary’s band. It looks fussy for about a week and then saves an argument a month.

The same discipline applies one step further down the funnel, where Meta’s CTR (all) and TikTok’s destination-based CTR are also two different metrics wearing one name: what CTR means on TikTok, and which clicks it counts separates the two before giving the benchmarks.

What is the hook rate formula, and how do you add it to Ads Manager as a custom metric?

The hook rate formula is three-second video plays divided by impressions, and Ads Manager will calculate it for you as a saved column if you build it once as a custom metric. Neither hook rate nor hold rate is a Meta metric.

Both are ratios you define, which is why they arrive in meetings with different denominators attached, and why the fix is to define them once in the account rather than in a spreadsheet every Monday.

The builder lives behind the Columns dropdown in Ads Manager, under Customize Columns, where a create custom metric control opens a formula field.

You assemble the formula from Meta’s own metrics, name it, choose a format, and set whether it is visible to you alone or to everyone with access to the business, per Jon Loomer’s walkthrough of the builder, last updated in March 2025. Madgicx’s guide, dated September 2025, places the control in the bottom left of the customisation panel and gives the same sequence.

Ads Reporting carries the same builder under the Metrics tab in a pivot table, per Loomer, which is where to build it if you report out of saved reports rather than the table.

The four ratios worth saving, in the conventions this page has been separating:

Metric	Formula	What it answers
Hook rate	three-second video plays divided by impressions	did anybody stop
Hold rate, ThruPlay over 3s	ThruPlays divided by three-second video plays	did the body keep the people who stopped
Hold rate, 15s over 3s	fifteen-second video views divided by three-second video plays	the same question without ThruPlay’s short-video completion quirk
ThruPlay rate	ThruPlays divided by video plays	hook and hold collapsed into one number

Pick each metric from the dropdown rather than typing the name. A formula assembled from the picker stays bound to the metric itself, while a typed column name is only as durable as the label Meta happens to be using.

What goes wrong when you build hook rate as a custom metric?

One setting produces most of the wrong-looking results people then blame on the creative.

Meta’s percentage format multiplies the result by 100 on its own, per Madgicx’s guide, so a formula written as the raw ratio and set to percentage reads 32%, while the same ratio multiplied by 100 and then set to percentage reads 3,200%. Either convention works as long as you do not apply both.

Madgicx recommends the number format with the multiplication written into the formula, and the choice matters far less than using the same one in every account you report on. It describes two options, Number and Percentage.

Neither guide is Meta’s own documentation and the panel has changed before, so read the dropdown in front of you rather than the screenshot in a guide.

A few limits are worth knowing before you build a library of these:

A formula needs at least two metrics and an operator, and it cannot divide by zero, per Loomer’s walkthrough. An asset with no three-second plays returns nothing rather than a zero, which matters the moment you sort the column to find the worst performers, because the worst ones can be the blanks.

Custom event names containing spaces break formulas, per both guides, which bites on accounts using custom conversions in the denominator. Map the event to a custom conversion with a single-word name, or rename the event.

Creating a custom metric while a column set is open resets your columns to the default selection, per Loomer, so build all four metrics first and arrange the view afterwards rather than losing a layout halfway through.

Save the four alongside impressions, video plays, three-second plays and ThruPlays rather than instead of them.

A hold rate of 26% on 400 three-second plays is not a finding, for the reasons set out in how many impressions a hook rate needs, and leaving the inputs on screen next to the ratios is the only way to notice without opening a second view.

How many impressions either number needs before it means anything is worked out further down this page, under how many impressions a hook rate needs.

Does a high hook rate mean the ad will be profitable?

No, a high hook rate does not mean the ad will be profitable, and that is the trap the metric sets. Hook rate measures attention. It does not measure intent, and the two come apart constantly.

A striking opening can hold a viewer perfectly while giving them no reason at all to want the product. Shock, curiosity and pattern interrupts all produce excellent hook rates, and none of them transfer to purchase intent on their own. Accounts that optimise on hook rate alone reliably end up scaling attention-grabbing creative that converts at nothing, then concluding the platform is broken.

Treat hook rate as a diagnostic for one specific failure mode: nobody is seeing this. It is excellent at identifying that condition and useless for anything else. It should never be the metric a scaling decision is made on, and an ad should not get budget because its hook rate is high. It gets budget because it converts, and hook rate explains part of why.

The failure has a recognisable shape once you have seen it. An account finds a hook format that reliably produces 40% hook rates, well clear of the 35% line AdLibrary’s 2026 bands treat as scalable, standardises on it, and watches CPA get worse while every creative dashboard improves.

The hook is doing its job perfectly and selecting for the wrong people: viewers drawn by curiosity rather than by interest in the product.

The check is straightforward. Compare hook rate against conversion rate across your library rather than looking at either alone. If your highest-hook assets are not among your best converters, the hook is buying attention from an audience that was never going to buy, and standardising on it will make the account worse in a way the creative metrics will never show.

Why do published benchmarks mislead?

Published benchmarks mislead because hook rate moves with variables that have nothing to do with creative quality. The bands above are useful orientation and poor targets.

Placement mix shifts it substantially, and by more than a single round target can absorb. AdSights’ published Meta bands put in-feed prospecting at 18 to 28% on the three-second view and Reels at 24 to 36% for the same metric.

Vendors disagree with each other for the same reason, so a figure lifted from one of them without checking which placement it describes will be wrong for most of your account.

Audience temperature shifts it more. Retargeting audiences hook far better than cold prospecting, for obvious reasons, and comparing a warm-heavy account against a cold-heavy one produces a conclusion about audience composition dressed up as a conclusion about creative.

Creative style matters too. Static and video are not measured the same way, and mixed accounts produce blended numbers that describe nothing in particular.

Platform differs as well. TikTok’s autoplay behaviour is not Meta’s, so the same asset yields different numbers on each, and a shared target across both is a target that is wrong on at least one.

The benchmark that actually predicts your next winner is your own account’s historical median for the same audience, the same category and the same creative style. Industry figures tell you whether you are in the right postcode. Your own median tells you whether this specific ad is better than the last one, which is the only question that changes what you do next.

How should you actually use hook rate?

Use hook rate as a gate rather than a target, and read the three metrics in sequence so each one gates the next.

Hook rate first, as a gate. Anything under 25% on cold traffic, the baseline in AdLibrary’s 2026 hook rate bands, gets a new opening or gets retired. There is no point diagnosing the body of a video nobody watched, and no point debating the offer in an ad that never got seen.

Hold rate second, on whatever cleared the gate. This tells you whether the body of the video earned the attention the hook bought. A strong hook with a weak hold is a rewrite of the middle, not the opening.

Conversion metrics last, and they are the only ones that decide budget. Hook and hold explain why an ad performed as it did. They do not determine whether it deserves more money.

Run them in that order and each number means something specific. Run them together and you end up scaling whatever looked most impressive on the dashboard.

Two practical notes on reading the numbers. Judge an asset on enough impressions to be stable, because hook rate is volatile early and a promising figure at low volume regularly regresses, and the section below puts a number on what “enough” means.

And compare like with like: an asset tested against a warm audience is not comparable to one tested cold, so a library ranked on raw hook rate mostly ranks audience temperature.

The habit worth building is recording your own median for each audience and format as you go. After a couple of months that private benchmark is worth more than any published figure, because it accounts for your category, your placements and your creative style at once.

How many impressions does a hook rate need before it means anything?

A hook rate needs roughly 10,000 impressions on the asset, per the 95% interval on a proportion worked through below, before it can be trusted to place that asset in one of the bands above, and around 1,800 impressions on each side before two openings can be told apart at all.

Both numbers fall out of arithmetic rather than opinion, because hook rate is a proportion, and the uncertainty on a proportion is calculable per the standard error formula rather than a matter of judgement.

Hook rate is three-second plays divided by impressions, as Meta defines the play, so an asset at n impressions is a sample of n chances to stop someone.

The spread on a measured proportion p is the square root of p times one minus p, divided by n, per the standard error formula, and the conventional 95% range is a little under twice that either side. At a reading of 30%, that works out as follows.

Impressions on the asset	A 30% reading really means	Width of the range
1,000	27.2% to 32.8%	5.7 points
2,500	28.2% to 31.8%	3.6 points
5,000	28.7% to 31.3%	2.5 points
10,000	29.1% to 30.9%	1.8 points
25,000	29.4% to 30.6%	1.1 points

Read that against the bands at the top of this page, which are five points wide and come from Taylor Sicard’s 2026 Meta creative benchmarks.

At 1,000 impressions, per the table above, the uncertainty is wider than a whole band, so a 30% reading is genuinely consistent with “functional, improvable” and with “healthy” at the same time, and an account that pauses one asset and scales another on that difference has made a coin toss with a budget attached.

At 10,000, per the same calculation, the range fits comfortably inside one band and the classification holds.

Comparing two openings needs more, because two uncertain numbers are being subtracted from each other. Per the same formula applied to a difference between two proportions, calling a three point gap between two hooks real at 95% takes about 1,800 impressions on each, and calling a one point gap real takes about 16,100 on each.

That second figure is the useful one, per that same calculation, because it prices a common habit. Chasing a hook from 30% to 31% is a test nobody runs long enough to resolve, and the result will be noise that reads like a finding, no matter how carefully the opening was rewritten.

These are floors rather than targets, because the formula assumes impressions are independent and they are not. The platform serves the same people repeatedly and clusters delivery by audience, and both of those widen the true range beyond what the table shows. So treat 10,000, per that calculation, as the point at which a number becomes worth discussing rather than the point at which it becomes exact.

How long should you run a hook test before reading it?

Run a hook test until the impressions arrive rather than until a number of days has passed, and convert that impression target into money before launching so that you know what the test costs.

At a $20 CPM, a figure to take from your own account rather than from a published benchmark, 10,000 impressions is $200 of spend per asset. A five-hook batch is therefore a $1,000 test, and a fifteen-hook batch is not a small decision.

Time is the wrong unit here because it is standing in for budget.

Per that same CPM arithmetic, a $50 a day ad set and a $2,000 a day ad set reach the same 10,000 impressions days apart, so a rule that says “read it after 48 hours” is really a rule about how fast you spend, not about the creative.

Let the impression count trigger the read and the same instruction works at every budget in the account.

One timing caveat sits alongside the arithmetic and pushes in the same direction. Meta’s own documentation describes an ad set as learning limited when it is unlikely to receive around 50 optimisation events in the week following a significant edit, and delivery during that period is deliberately exploratory. An early hook rate is therefore partly a measurement of an audience mix that will not be the same one next week.

How do you fix a hook rate below the benchmark?

Fixing a hook rate below the benchmark starts with three checks that have nothing to do with the creative, and only then with swapping openings in batches against footage that already works.

Delivery first. A three-second play requires three seconds of playing, and autoplay itself carries a condition: Meta’s specifications for video ad playback by placement set a minimum share of the ad’s pixels that must be in view before autoplay triggers at all, and that minimum differs by platform and placement.

Part of every hook rate is therefore a measure of whether the ad got a chance to play, and no rewritten opening line moves that part.

Placement second. AdSights’ published Meta bands put in-feed prospecting at 18 to 28% on the three-second view and Reels at 24 to 36%. An account reading 22% that is feed-heavy is sitting inside its own placement’s band.

An account reading 22% that is Reels-heavy is below it, per those same AdSights bands. The same number supports opposite conclusions, and the placement breakdown that settles which one you are looking at is a single column in Ads Manager.

The crop third. A 4:5 feed master delivered into Reels gets cropped or letterboxed, and the vertical frame reserves its top and bottom for interface: profile row and sponsored label above, captions and the call-to-action button below.

A hook carried by on-screen text positioned for the feed can be sitting underneath that furniture in the placement where it is being judged.

Sound has a version of the same problem, since automatic sound on feed video is a setting the viewer controls, so an opening that exists only in spoken words is not reaching everyone it is being measured against.

Only once those three are ruled out does the creative become the suspect, and then the fix is a swap rather than a reshoot. Keep the body that already holds and cut five to ten new openings against the same footage.

That isolates the variable properly, because the body, the audience and the offer stay constant while the first three seconds are the only thing moving, and it costs almost nothing, being an edit rather than a shoot.

Work the arithmetic once and the size of the job becomes obvious. An asset at 240,000 impressions and 43,200 three-second plays reads 18%, which is inside the invisible band per the bands at the top of this page.

Reaching the 30% benchmark on the same impressions means 72,000 three-second plays, so 28,800 more people have to get past the opening without a dollar of extra spend. That is a lot to ask of one rewritten line, which is why a batch of openings outperforms a single rewrite.

Write the alternates at briefing time rather than after the result arrives. The brief format we use carries a HOOK line with two alternates for exactly this reason: three openings specified before the creator films cost nothing extra and turn one shoot into three tests, where three openings commissioned after an 18% result cost three briefs, three fees and three weeks.

One case where none of this is the answer. If hold rate stays low across every opening you try, the opening was never the problem. The body is not earning the attention the hook buys, and sharpening the first three seconds further only delivers more people to the same disappointment.

How do you improve your hook rate, and which openings actually work?

You improve a hook rate by testing openings in batches across different structures rather than rewording one, and the public evidence on which structures to use has a catch worth understanding before you use it: it measures which ads survive, not which ads stop the scroll.

The dataset is Brandsearch’s 2026 analysis of winning Meta ad patterns, built from 523 ads across fitness, beauty, home goods, supplements and fashion, every one with at least 30 days of confirmed run time and real daily spend behind it, with the hook type, format, call to action, video length and copy length logged for each. Three opening structures accounted for 71% of those long-running ads.

Opening structure	What it is	Where it concentrated in the 523-ad sample
Pain point first	Opens on the problem the viewer has today, before the product exists in the frame	Fitness, 44% of hooks; beauty, 38%
Authority claim	Opens on a credibility signal: a number, a credential, a study	Supplements, 52%; beauty, 18%
Visual pattern interrupt	No words at all, a jarring visual: a texture shot, a before and after flash cut, a product doing something unexpected	Home goods, 61%; fashion, 34%
Question	Opens by asking the viewer something	8% of ads that lasted 30 days or more

Why does the opening with the best hook rate often fail?

The opening with the best hook rate often fails because hook rate measures who stopped, not who bought, and those are not always the same people. The 8% in the table above, from Brandsearch’s study of 523 Meta creatives, is where that shows up in the data.

That number is the one people misread, because of how the study was built. A 30-day survivor was selected by profit, not by hook rate. An ad runs for a month because somebody kept funding it, which means it converted, and the opening only had to get it into the auction for that to happen.

So a structure can be rare among survivors either because it fails to stop the scroll or because it stops the wrong people.

Question openings look like the second failure rather than the first. “Struggling to sleep?” is a question almost anyone will answer yes to, which is exactly what makes it good at collecting three-second plays and bad at collecting customers.

It widens the top of the funnel without qualifying anybody, so the hook rate goes up and everything downstream of it goes down.

That is the same argument this page has been making since the profitability section, now visible in somebody else’s data: the openings that maximise hook rate and the openings that survive are not the same set.

Which gives you the rule for the batch. When you cut five to ten new openings against footage that already holds, spend them across structures rather than across rewordings of one.

Three pain-point variants are one test, not three, because they share whatever is wrong with the approach. One of each structure above, plus two variants of whichever one your account has historically favoured, is a batch that can actually tell you something.

Then read every opening against both numbers: the hook rate says whether it stopped anybody, and the downstream cost per acquisition says whether it stopped the right ones.

An opening that wins on the first and loses on the second is the most expensive result in creative testing, because it looks like progress in the creative report and shows up as a problem two weeks later in the account.

One structural note on the sample. Three of the five categories in the study, supplements, beauty and fitness, are heavily regulated categories where claims are constrained, which pushes creative towards testimonial and authority framing for reasons that have nothing to do with attention. If you sell in an unregulated category, weight the home goods and fashion rows, where the visual pattern interrupt dominated, more heavily than the overall 71%.

The short version

30%+ hook rate on Meta per Taylor Sicard’s 2026 benchmarks, 20 to 25% hold rate per AdLibrary, both read against your own account’s history rather than a table from a blog. On TikTok, build the number from 2-second video views and compare it only to other TikTok numbers. Use all of them to diagnose creative and never to decide spend.

Hold the benchmark and the average apart in your head, because they are not the same claim.

The measured middle of the Meta distribution is 20.9% per Biddyco’s 2026 study of 479 top-spending video ads and 25.44% per Billo’s panel of 88,329, so a 30% target is a demanding one and an ad at 22% is ordinary rather than broken.

Quote every hold rate with its denominator, since per the four conventions in circulation the same asset reads anywhere from 7.5% to 25%.

What none of these numbers do is tell you which ad will win. Biddyco found that the ads Meta chose to spend the most on did not carry higher hook or hold rates than the rest, so the metric diagnoses a weak opening and nothing further.

Decide budget on conversion metrics, rewrite openings on hook rate, rewrite middles on hold rate, and keep your own median per audience and placement as the only benchmark that accounts for your category.

Having enough creative in the system for any of this to be worth diagnosing is a separate question, and it has an arithmetic answer: how many ad creatives to test per month.