The hook rate benchmark you are using is eleven TikTok accounts from 2022

01The benchmark is eleven accounts, one agency, the first half of 2022

The figure is 30.66%, published on 6 July 2022 in a post titled Benchmarking TikTok Video Ad Hook Rates, and the sample sits in the post itself: eleven TikTok accounts that agency managed, measured from 1 January to 30 June 2022. The sentence reads "The average ad hook rate across accounts was 30.66%".

Each hop subtracts. The denominator goes first, then the sample, then the date, and by the fourth hop a figure describing one shop's clients in early 2022 is being read as what TikTok does. The only readable copy of the original now sits in the Internet Archive. That is the whole benchmark. Eleven accounts.

02The trail is walkable, and it is a fifteen-minute job

03A real study's sample, attached to numbers it never measured

This is the most citable finding on this page. Several sites publish hook rate bands of roughly 25% to 35% and credit them to Motion's Creative Benchmarks 2026, quoting the sample in the same breath: 550,000 ads, over 6,000 advertisers, around $1.3 billion in spend. The report is real. The bands are not in it.

Motion's hook rate ranges do exist. They sit on a separate blog page about creative performance metrics, and that page states no sample at all. The ranges have no dataset under them, the dataset has no ranges in it, and downstream the two arrive in one sentence. Motion ran a real study and did nothing wrong here. Other people are borrowing its credibility.

A real dataset's credibility gets transplanted onto numbers it never measured.

The pattern the audit kept running into

04Hold rate is four quantities wearing one name

In one day of fetching we found four different fractions published under the words hold rate, and published benchmarks for it running from 15% to 60%. That spread is not a disagreement about performance. It is four pieces of arithmetic being compared as though they were one.

Write your denominator into the column header and hold it there. Then never compare a figure computed one way against a figure computed another way, however close the two names sound.

A benchmark quoted without its denominator is not a number, it is a mood.

The rule we ended the audit with

05What does Meta actually publish?

A definition, and no benchmark. Meta's own help page defines hook rate as the number of three-second views divided by the video's total impressions, times 100. That is the strongest citation available in this whole area, and it attaches no target, no average and no range to the metric it defines. Meta also labels it "in development". Taylor Sicard's 2026 roundup says the same thing from the other side: "Meta publishes no official benchmark for hook rate, hold rate, or creative hit rate".

Two more things we checked. hook_rate and hold_rate are not fields in the Ads Insights API, so hook rate is a metric you build rather than a column you request. A search of Meta's help center for hold rate returns no article at all, which makes hold rate a practitioner invention with no platform definition behind it. The official counts underneath are three-second video plays, ThruPlays, and the milestones at 25, 50, 75, 95 and 100 percent. Everything above them is arithmetic somebody chose.

06How do you compute your own baseline?

Export ninety days at ad level, pick one denominator and hold it, drop everything under five thousand impressions, take the median. Twenty minutes, and that is the whole job. Your own median beats any published figure, because it is computed on your product and your placements.

1. PICK ONE DENOMINATOR AND WRITE IT DOWN.
   Impressions, not reach. Reach flatters, and it flatters unevenly
   across placements, which is the worst combination.

2. EXPORT 90 DAYS AT AD LEVEL, with:
   impressions | 3-second video plays | ThruPlays | spend | placement

3. COMPUTE, per ad:
   hook rate = 3-second video plays / impressions
   hold rate = ThruPlays / 3-second video plays

4. DROP every ad under ~5,000 impressions.
   Then take the MEDIAN. Not the mean.

5. SPLIT by placement and aspect ratio before you compare anything.
   A feed number and a Reels number are different populations, and
   averaging them produces a benchmark you cannot act on.

Then treat that median as a floor, not a target. The worst self-inflicted defect in our own archive came from optimizing to a check rather than to the thing the check stood in for, and it shipped.

07A hook rate counts stops and prices none of them

An opening that arrests the wrong person scores well and costs you money twice: once for the impression, again for the click that was never going to convert. Four more things the number cannot separate are below.

The hold rate card is why any film over roughly ten seconds has to declare a second hook and name the frame it lands on. In several films we measured, the re-hook carries more production value than the opening does, and no single-number retention metric will show you that. The structural version is in how a fifteen-second ad is actually built, and the repair job is in what to do when your hook rate is below benchmark.

08The metric we built ourselves, which ranked backwards

We are not auditing other people's numbers from a clean position. We built our own hook metric and it was wrong, in the most dangerous way available: confidently, and in the reassuring direction. A hook is change, so measure how much the picture changes in the first two seconds and gate on it. One study on one rejected film supported it. Then we ran it across all five judged films, and the two accepted ones came back as the least changeful of the set.

What separated them was the last row of the same table, and it was not motion. It was information. The accepted films state their one idea before frame 12. The rejected ones said their first word at 9.92 seconds, or never. A metric that agrees with you is the one to check hardest.

In every case the wrong answer was the reassuring one.

Our own audit of four failed detectors and five failed measurers

09What we hold, what we do not, and what we could not verify

We hold measurements of files: first-sound timings across 48 films, cut rhythms, contrast under glyphs, cap heights on delivered frames. We hold a verdict log with kill reasons. What we do not hold is one ad with a reported hook rate, hold rate or return on ad spend attached to it. Not one, in the whole corpus. That makes the doctrine a prior about craft and not evidence about conversion.

Two figures you may have seen are missing on purpose. One is a Reels hold rate average, on a page that refused every request we made. The other is a range for what counts as a good hold rate, which not one page we could fetch actually states. Naming what you could not verify is the part nobody publishes.

Two of the checks above are one idea in different clothes: a number you plan to act on needs its sample decided before you look, not quoted afterwards. The same reasoning sits behind how many creatives you can actually learn from and why twenty creatives usually produce noise.

Run the trail yourself on whatever benchmark is about to go into your deck. Not because vendors are dishonest, but because a number with no origin behaves exactly like a number with one, right up to the moment you spend against it. Eleven accounts, six months, 2022.