Start with 1 task or 10 hours — scale when ready. View pricing →
Women's health supplement · DTC · Meta ads

One story per person. Four hooks each. Zero reshoots.

A supplement brand in a strictly moderated health category needed video ads that show what to scale, not just look good. We turned their full brand analytics into a 12-video test and edited it from footage they already had.

September 2026Test batch, round 1
2 UGC creators4 hooks each, 9:16, 86–96 s
Our scopeResearch, angles, scripts, approval doc, editing, sound, graphics
Not our scopeFilming and media buying: the client shot and runs the ads
real frames from the ads
Creator 2 with hair-loss overlay
Creator 1, head in hands at laptop
Creator 2 with two-options card
Creator 1, tired with phone
Cascade graphic: stress to symptoms
Creator 1 talking to camera
Creator 2 with camera-off card
Creator 1 outdoors
12videos delivered for the test
2creators, one story each, hooks swapped in the edit
+4founder videos rebuilt from existing footage
0reshoots, licensed tracks or stock sounds
Hook = the first seconds that make people stop scrolling.Body = the main part, shared by all four hooks.Angle = the reason to care that a video tests.
Q1 · Why not just make a few good videos?

Because a nice video doesn't tell you what to scale.

The brand sells a daily hormone-support supplement through paid Meta traffic. Its best ads were raw founder and customer stories. Now it needed a test that shows which opening works, and every line had to pass Meta's rules for health ads.

For marketersHook isolation needs identical bodies; 3 angles pre-validated by the client's ad data; no cure or treatment claims anywhere.

What we got
  • A survey of 834 customers and persona research
  • 10 best-performing past ads
  • Long continuous takes, full of pauses
  • Founder voiceover recorded apart from her video
The brief
  • 2 creators × 4 hooks, plus founder videos
  • Hooks built on the 3 angles that already sell
  • Every story follows the beats of past winners
  • "Feeling like yourself again" as the through-line
Q2 · How do you know what to say?

We started with what 834 customers already told the brand.

The client handed over everything: survey results, personas, the founder's story, product facts, brand voice, the creator guide and their winning ads. We read all of it before writing a line.

For marketers10 winners transcribed and broken down by hook type, angle, emotion and where the product enters; our own early ideas checked against the brief, weak ones cut.

1 · We read

Everything the brand knows

Customer survey, personas, founder story, product facts, brand voice and the creator guide.

2 · We decode

Why the winners won

Ten past ads transcribed: which hooks stop people, which emotions carry, when the product appears.

3 · We write

Scripts in customers' words

Every beat and hook points back to something a real customer said.

"I want to feel like myself again."#1 PHRASE IN THE SURVEY · 147 MENTIONS
Creator 2 talking to camera

Every story ends on getting yourself back, not on a symptom score. The endings are small, visible moments: the camera back on, the open collar, feeling like herself by week 8.

Doctors who couldn't figure it out.STRONGEST EMOTION ACROSS THE WINNING ADS
Creator 2 with acne photo overlay

"Who failed her" became a beat of its own. Not a vague "nobody listened", but named specialists and named prescriptions, so viewers recognise their own path.

"Too many pills."#1 REASON BUYERS LEFT OTHER BRANDS · 64 MENTIONS
Creator 1 reading phone

The product never shows up as "one more supplement". It enters as the thing that replaces the stack. 81% of customers said simplicity is what they value most.

Q3 · Why 12 videos?

3 stories × 4 hooks = 12 tests for the price of 3.

Each creator films her story once. Four different openings are cut in during the edit. Inside each group the main part is exactly the same, so the test compares openings, not editing accidents. The founder's videos were rebuilt the same way.

For marketersCreator angle coverage: root cause ×3, undiagnosed ×2, diagnosis and its mechanism ×2, identity ×1. All three angles from the brief covered.

See the 8 creator hooks
VideoAngleHook (shortened)
Creator 1.1Diagnosis"My doctor put me on birth control 'for my PCOS' and told me to give it three months."
Creator 1.2Root cause"Ask your gynecologist the words: insulin resistance. If she just shrugs…"
Creator 1.3Undiagnosed"I loaded six years of my labs into ChatGPT. It answered in 40 seconds with one word."
Creator 1.4Root cause"'Eat less, move more.' Turns out my insulin doesn't work."
Creator 2.1Identity"I didn't know that same day was going to break my body."
Creator 2.2Diagnosis"Have a baby right now, or take the pill until menopause. I'm 29."
Creator 2.3Root cause"'Take a long leave, it's just stress.' Not in month four of the job I chased for eight years."
Creator 2.4Undiagnosed"Four months into my promotion I started turning my camera off on every call."
+ 4 founder videos, rebuilt from existing footage with the same hook-and-body structure
3 angles from the client's ad data 4 hooks per story 1 body per creator, filmed once 4 rounds of hook rewrites 5-point audit before sending
Q4 · What does every story have to hit?

Five beats, taken from the ads that already sold.

The client asked for stories built on the beats of their winning ads. We mapped those beats and wrote every story to the same skeleton, then sharpened the hooks with real, specific details.

Details we added
  • Hair coming out in clumps
  • Camera off on every call
  • Down to three hoodies in the closet
  • Labs read by AI in 40 seconds
  • A choice of two options, neither of them hers
Checked before sending
  • Matches the brief
  • Believable and logical
  • Has potential to win
  • Length close to winners
  • Hook leads into the body
beat 1One pain"Six years. Chin acne that no facewash touched."
beat 2Who didn't help"The dermatologist said acne. The gynecologist said stress."
beat 3The real cause"Stress raises cortisol. Cortisol breaks your insulin sensitivity."
beat 4One instead of many"Her whole list, in one sachet."
beat 5Result and hope"Week 10, first regular cycle in a year. Month 3, camera back on."
Q5 · Who tells which story?

Each creator tells the story her audience lives.

We didn't give both creators the same script with a different face. Each story was written for a real customer group from the research, so viewers recognise themselves in the first seconds.

For marketersCreator 1 maps to the diagnosed persona (51% of buyers, clinical language, "insulin resistance"); Creator 2 to the burnout pathway ("running on empty", corporate roles).

Creator 1 pointing
Diagnosed, still stuck

Creator 1

  • 8 doctors, 8 diagnoses
  • Her labs decoded by AI
  • Off the pill only with her doctor
Creator 2
Career woman, burnt out

Creator 2

  • Promotion, then the camera goes off
  • Two options she refused
  • Stress → hormones, explained
Q6 · Can you even advertise this on Meta?

Yes, as long as no line promises a cure.

Health is one of the most restricted ad categories. Safety was built into the scripts from the first draft, not fixed in moderation.

Rule 1

No cure language

Stories describe how someone feels, never what the product treats.

Rule 2

Said out loud

"It doesn't replace your doctor. It's not a treatment." Spoken in the video, not hidden in the caption.

Rule 3

Doctor in the loop

Any change to medication happens "under her supervision", never on the viewer's own.

approval

One round of comments

The client got a short doc: batch structure, angle matrix, full hooks, bodies as beats and in full. Their edits were minor wording, made in the same doc with comments kept.

Q7 · How do you make four hooks a fair test?

We edit in code, so only the hook changes.

The edit is written frame by frame, not dragged on a timeline. The main part is rendered once, and each variant is the hook joined to it without re-encoding. Four files, one identical body, and a fix in the body is one rebuild instead of four.

For marketers[hook]+[body] stream-copy concat on a forced keyframe; bodies byte-identical across variants; every segment's frame count checked against the plan.

Hook A
Same body86–96 s
Hook B
Same bodybyte for byte
Hook C
Same bodybyte for byte
Hook D
Same bodybyte for byte
lip-sync0.10–0.26 sworst gap between lips and voice where the voice was recorded apart from the video. Simple stretching gave 0.89 s.
sound vs picture≤0.09 smax drift. Building the timeline from 19 audio files as delivered would have pushed the voice 0.70 s ahead.
loudness−14 LUFSthe level of the feed. The raw videos sat at −28, about half as loud as everything around them.
See how the edit was engineered

Pauses cut by speech, not by eye

Gaps over 0.45 s shrink to 0.22 s: still a breath, not a hole. A map of original time to edit time drives every B-roll, caption and graphic.

Lip-sync in three steps

Word-level alignment at 10 ms, matching words in order so short words don't latch onto the wrong spot, then a smooth time-warp with a knot every ~2 s so motion never jitters.

A hidden offset, found and fixed

In every talking-head clip the picture started 0.2–0.8 s after the sound. Measured clip by clip; the missing frames went to the previous B-roll shot.

Two shots no warp could save

Replaced with pop-up cards on a clean background that land exactly on their word, so the meaning reads with sound off.

Mastering without delay

The standard limiter adds 5 ms of lookahead and shifts the track. Peaks are tamed by a gain curve instead, and checked again after AAC encoding.

Picture never re-encoded

Every sound fix is a remux with the video copied. Delivery: 1.4 GB in 74 parts, rebuilt on the client side, 8 of 8 matched by checksum.

Q8 · What about music, captions and graphics?

Nothing borrowed, so the ads can run for months.

The client's music reference came from a free library, and those licences almost always forbid paid ads. So we made everything ourselves.

Music

Written from scratch

Same tempo, key and mood as the reference, no samples inside. It drops under the voice and opens up between lines.

Graphics

In the product's colours

Picked with an eyedropper from the product footage, so overlays match the pack in the frame.

Captions

Built for sound-off

Readable on any background, broken by real text width, a number never split from what it counts.

sound effects synthesised too
Cascade step 1
Cards build word by word
Cascade step 3
…in the product palette
Cascade final
…ending on the spoken word
Hair overlay
Proof shown, not told
Camera off
Native UI, not a banner
Option card
Lands on the spoken word
Q9 · What did the client get?

Twelve finished videos, each one a clean test.

Two creators with four hooks each, plus four founder videos rebuilt from footage the brand already had. Every video checks one angle and one opening, so after the test the brand knows what to scale.

12videos delivered: 8 creator + 4 founder, 1080×1920, 86–96 s.
0reshoots, and no licensed assets in sound or graphics: nothing to clear, credit or renew.
1 rebuildis what a fifth hook or a new market costs. Not a new project.
ad results come after launch
Q10 · What would this look like for you?

You bring the brand data. We turn it into a test.

For DTC brands in women's health (hormones, menopause, fertility, skin, digestion) that sell on Meta and TikTok and keep hitting health-ad limits.

step 1Research

Your survey, reviews, personas and winning ads.

step 2Test plan

3 stories × 4 hooks across your proven angles.

step 3Scripts

Meta-safe, approved in one round of comments.

step 4Edit

From your footage, founder lip-sync, AI voice when needed.

step 5Delivery

12 variants, no licences, rebuild-ready.