Sell hook volume, not polished single videos
The AI-native offer is 30 hook variants against one validated body, sold as a testing package. Deliverable structure, naming convention, and how to price it.
A brand asks for a quote on "a UGC ad for our new product." The default response is one number for one finished 30-second video, and you're now competing against every other supplier on the same axis: who makes the nicest single asset for the least money. That's a race with a floor, and the floor keeps dropping — published rates for AI video freelancers span $150–$1,200 per deliverable or $50–$150 per finished minute, and buyers know the range.
The better answer is a different product entirely. Not one polished video, but 30 hook variants running against one body that already works, delivered as a testing package. It's a stronger offer for the buyer, a much harder one to price-shop, and it's the shape AI production is actually good at.
Why the hero video is the wrong unit
Media buyers stopped picking a winner upfront years ago. The workflow is: put many openings in front of the algorithm, let it find the one that holds attention, scale that one. Which means the thing a buyer needs is not the best single asset — it's enough variance to find out what works.
The economics used to prevent this. When each creator video was reported at $150–$600, testing 30 openings was a five-figure exercise and nobody did it. AI production drops the marginal render to a small fraction of that, and once the marginal unit is cheap, the optimal number of variants goes from one to many. That's the actual shift — not that AI makes videos faster, but that it makes the right number of attempts affordable for the first time.
Two more reasons the hero video underperforms as a product. It has no diagnostic value: if it fails, nobody learns anything, because there was no controlled comparison. And it decays — creative fatigue sets in on any single asset as frequency builds, so a one-video deliverable has a shelf life in weeks while a variant library refills itself.
What's actually in a hook-volume package
The structure that makes this sellable rather than just "more files":
One validated body. This is the load-bearing part. The body is the middle and end of the ad — product reveal, demonstration, objection handling, CTA — and it stays fixed across every variant. Validated means it's either already performing in the client's account or it's the output of a prior round. Selling 30 hooks against an unvalidated body is selling 30 rolls of the same dice.
Thirty hooks in four or five families. Not 30 rewordings. Families are structurally different openings: a question, a flat contrarian claim, a negation ("stop doing X"), an immediate visual demo, a price or number callout, a social-proof line. Six to eight variants inside each family, so the test can tell you which kind of opening works before it tells you which specific line does. This is the difference between a batch and an experiment — batch generation makes the point that varying two things at once yields files rather than conclusions.
One controlled cut point. Every hook ends and hands off to the body at the same timestamp, so the only difference between two variants is the first three seconds. If hook lengths drift, you've changed pacing as well as copy and the result is uninterpretable.
A ratio set. Whatever the buyer's placements need, applied uniformly across the variants.
A read-out. A short document naming what to measure, which is not CTR. Hook rate — the share of impressions surviving the first two or three seconds — is the metric the test is built to move, and stating it up front makes the package a service rather than a file drop.
On the production side, the mechanic that makes 30 variants practical is generating them as a planned set rather than one at a time. Versely's agent will create a branded hook pack: describe the brand and the offer and it plans N distinct hooks so they don't repeat each other, renders each as a 1080p portrait clip with the line burned on screen, and charges once for the whole pack rather than per hook. Once the winning openings are known, adding a hook to your video is the step that puts a chosen line onto the fixed body. The AI ad generator covers the same set-based build when the body is being produced alongside the hooks.
A naming convention that survives a real test
This sounds like housekeeping. It isn't. A 30-variant test dies at the analysis step, not the production step, because two weeks later the ads manager shows a winner called final_v3_REALFINAL.mp4 and nobody can say which hook family it belonged to. Every file name has to be self-describing.
Use a fixed field order, underscores between fields, hyphens inside them:
{client}_{offer}_{body}_{hookfamily}-{hookid}_{ratio}_{date}
A real one:
acme_sleepgummy_B02_question-H14_9x16_260820.mp4
Reading it back: Acme, the sleep gummy offer, body version 2, hook family "question", hook 14, 9:16, 20 August 2026. The two fields that matter most are body and hookfamily. body is what stops you comparing hooks across different bodies and drawing a false conclusion. hookfamily is what lets you group 30 results into five meaningful rows instead of staring at 30 unordered lines.
Two rules that keep it usable. Hook IDs are permanent and never reused — H14 means the same line in round four that it meant in round one, so a hook library accumulates instead of resetting. And the name goes in the ad-platform ad name too, not just the filename, because the platform's report is where the analysis actually happens.
Pricing it against a hero quote
The framing that closes this: the buyer is comparing your package against one hero video, so price it so the comparison is favourable and honest.
| Hero video | Hook-volume package | |
|---|---|---|
| What's delivered | 1 finished asset | 1 body + 30 hook variants, ratio set included |
| What the buyer learns | Nothing, unless it wins | Which hook family works, ranked |
| Shelf life | Until fatigue, then re-buy | Winners scale; library refills the next round |
| Comparable spend | Per-deliverable rate | Priced as one project, not 31 files |
Three pricing principles that hold up.
Price the package, never the file. Thirty variants priced per file produces an absurd number and invites the client to cut the count, which destroys the experiment. One project fee, with the variant count stated as a spec line rather than a multiplier.
Anchor above a single hero quote, not against 30 of them. Something in the range of one and a half to two and a half times what you'd charge for a single premium deliverable is defensible: the client is buying more useful output than a hero video and materially less than 30 hero videos, and the value is in the learning rather than the file count. Package discounts of 15–20% at five or more deliverables are the published norm, and a variant package sits well past that threshold, so build the discount into the flat number instead of showing your working.
Sell the second round in the first contract. The test produces winners; winners fatigue; the next round tests new families against the surviving body. A two-round or quarterly structure converts a project into a retainer without a separate sales conversation. It's also the honest description of what the client needs: nobody buys creative testing once. The trigger for round two is a measurement — the leader's hook rate sliding — not a calendar date, which is why the contract should name the metric rather than a refresh interval.
For grounding on what a finished ad involves end to end, what a finished 30-second ad costs breaks the components down in credits.
Running the test so it produces an answer
The package is only worth what the analysis is worth. Four rules, all of them boring and all of them load-bearing.
- One variable. Hook copy and hook visual change; body, voice, captions, music, length and ratio do not. If the client wants to test a second body, that's a second test.
- Equal exposure. Variants need comparable impressions before any of them is judged. Killing a variant at 200 impressions is reading noise.
- Judge families before lines. Rank the five families first. That's the finding that transfers to the next product and the next client; the specific winning line usually doesn't.
- Retire the winner deliberately. When hook rate on the leader starts sliding, that's the fatigue signal and the trigger for round two, not a reason to change the body.
A framework for ad hook testing goes deeper on the measurement side, including how long to run before calling a result.
FAQ
Isn't 30 variants overkill for a small brand?
The count should track ad spend, not brand size. A brand spending a few hundred a month can't buy enough impressions to resolve 30 variants, and 8–12 across three families is the honest version of this package at that level. The structure is the same; only the width changes.
What if the client has no validated body yet?
Then round one is a body test, not a hook test: three or four structurally different bodies with one shared hook. Hook-volume testing assumes the back half already converts. Running it before that is measuring openings into an ad that was never going to close.
How do I stop 30 hooks being 30 rewordings of the same line?
By planning the set together rather than generating them one at a time, and by requiring family coverage in the brief. If every variant is a question with different nouns, the test can only tell you which nouns performed — which is the least transferable thing it could have told you.
Does the client get to keep all 30 files?
That's a licensing question, not a production one, and belongs in the contract like any other usage term. A common structure delivers all variants but names surfaces and window in the licence, with winners eligible for extension at a stated renewal rate.