Creative review · Oct 21 – Nov 3, 2026 · all offers, RedTrack

VID863 "Camera Talk" is the best fresh creative of the last two weeks. It also got rejected.

One TPS talking-head video, launched Oct 22, spent more than any other ad in the account in this window and did it under target CPA. On Oct 28 Meta rejected the scaled ad for Health & Wellness. This page is the whole story: why we call it a winner, what the rejection data actually points at, and the three variants built to find out.

$425K

Spend, 13 days

#1 ad in the account

5,931

Purchases

ROAS 1.89

$71.6

Purchase CPA

target $90

Oct 28

Rejected by Meta

Health & Wellness

Sources: RedTrack (campaigns 10.22 Camera Talk NN Test + US Weekly Wins), ClickUp tasks 17th71z6w47, 17th71z809d, 17th71z7tj6, 17th71z8mte, policy calibration doc (Nov 3). Numbers are purchase-level unless marked.

02 · Why it is the winner

It clears every KPI we judge creatives by, and it does it at scale, not in a test.

Our bar for a TPS creative: purchase CPA at or under $90, IC CPA around $30, IC CR at 3%+, purchase CR at 1%+, CPC under $1. A creative only counts once it runs in a Scale / Wins campaign, not just in its test.

$71.6

Purchase CPA

target ≤ $90

1.47%

Purchase CR

target ≥ 1%

$28.4

IC CPA*

target ~$30

$1.05

CPC

target < $1.00

Against the established TPS winners, same 13 days

Creative	Origin test	Spend	Purchases	CPA	ROAS	CPC
VID863 Camera Talk	10.22.26 (MT)	$424.7K	5,931	$71.6	1.89	$1.05
VID742 (Menopause 4x4 line)	06.19.26	$232.8K	2,910	$80.0	1.74	$1.16
VID601 This Is My Mom	07.18.26	$135.1K	1,601	$84.4	1.72	$1.16
VID636	04.02.26 Wins	$65.2K	834	$78.2	1.89	$1.22
VID467	04.02.26 Wins	$53.9K	631	$85.3	1.70	$0.90

* IC = RedTrack conversions minus purchases (initiate checkout events). VID863 IC CR is 3.7%; Meta-side numbers in the HG brief (Oct 30) were IC CPA $34.8, IC CR 2.7%, so treat IC as directional. Spend and purchases are summed across all HL/AT combinations of the video in both campaigns. VID742 and VID601 totals also include their Wins and RC campaign rows.

What makes it a winner, not an anomalySpend is spread over 10 days of real scale ($25K–$94K/day), the CPA held between $62 and $80 on every day with meaningful spend, and the Wins copy of the ad ($29K) runs at $59 CPA. The only KPI it misses is CPC, by five cents.

03 · Where it came from

A net-new test of three videos. One hook structure won, and it won by a lot.

Test campaign "10.22.26 - TPS - TPS14 - MT - Top 5 - Camera Talk - VID - NN Test - ABO", copy by MT, build by Matt. Three videos (VID861, VID862, VID863), one ad set each, 2 HL × 2 AT per ad set, $180/day per ad set.

Video	Hook	Spend	Purchases	CPA	Read
VID863	H3 "If your jowls look like this… I'm 67, and I've never had a needle in my face"	$395.2K	5,433	$72.7	winner, moved to Wins
VID861	H1 "If your eyes look like this, your jowls like this…"	$36.6K	522	$70.0	strong, lower volume
VID862	H2 "If your foundation sinks into lines like this by lunch…"	$0.2K	2	—	never got delivery

The mechanism

UGC selfie talking head in a car, a 65–70 avatar with visible jowls and neck lines. Hook opens on a problem close-up ("look like this"), then "look at these pictures of me" cuts to her own "before" selfie, then the body is her, now, in the same car. Blind: the product is never shown, the lander is "Top 5 skincare brands for women over 60, skip to #1". Winning ad is HL2 ("Target's #1 Wrinkle Eraser…") × AT2 ("the death of Botox").

VID863 first frame: talking head in the car — 0:00 · talking head, "IF YOUR JOWLS"

VID863 hook contact sheet, first 12 seconds — 0:00–0:12 · hook: problem close-up → her "before" selfie → her now

VID863 contact sheet every 2.5 seconds — full video, one frame every 2.5 s

The brief and shot list: Google Doc "10.18.26 - TPS VID – Nothing Was Working Until I Learned This", ClickUp 17th71z6w47. Reference ad on Atria m880104018170915.

04 · Two weeks, day by day

Scaled from day three. The dip on Oct 29 is the rejection, not the creative.

Daily spend and purchases for VID863 across the test campaign and US Weekly Wins (both HL/AT combos with spend). Oct 28: Matt files the rejection task. Nov 1: a low-delivery day account-wide, not specific to this ad.

$94.2K

Best day (Oct 31)

1,287 purchases · $73 CPA

$62–80

CPA range on scale days

never above target

10 of 13

days above $10K spend

from Oct 25

05 · Iterations on it

Four iteration tracks opened in one week. Only one has data, and none of them is the fix for the ban.

10.26 · MT · cs team

Avatar iteration

Four new avatars on the same script (Caucasian, Black, Asian, Pixar-style 3D). Not launched yet.

10.30 · HG · launched

Visual hook iteration

VID888–892: same script and voice, new first 2–3 s. ~$1.5K spent, 16 purchases, ~$92 CPA. VID889 and VID891 are the only ones with signal.

11.01 · DM · launched

Hook iteration (ban re-upload)

VID910–911 with softened copy ("so many women are skipping their next Botox appointment"). No measurable spend yet.

11.03 · EK · in progress

Ban-fix iteration

Three variants that each change one thing. Built from the policy check calibrated on our own rejected vs approved ads. Section 08.

The HG visual-hook brief already flagged VID863's two soft spots: IC CR 2.7% vs 3% and IC CPA $34.8 vs $30 at the time. Those are also what a cleaner opener should move.

06 · The rejection

Oct 28: the scaled ad (HL2_AT2_VID863_H3_B1) is rejected for Health & Wellness. Small tweaks did not clear it.

Meta's reason

Health and Wellness

The best guess at the time (Matt, task 17th71z809d)

The before-and-after setup: a tight close-up of deeply wrinkled skin with "LOOK LIKE THIS", then the same woman with smooth skin, read as an implied result.
Personal-attribute and result claims in the copy: "women over 60", "the woman in this video is 67", "the death of Botox", "collagen output by up to 258%".
Suggested fix: cut the wrinkled opener, remove the age callouts, keep only the smooth-skin talking head.

What is in the first 12 seconds

Time	On screen
0:00	Talking head in the car, "IF YOUR JOWLS"
0:01	Close-up of another woman's lower face, jowls and neck lines, "LOOK LIKE THIS"
0:05	Close-up of eye wrinkles, "I'M 67 AND I'VE"
0:09	Her "before" selfie video in the car, lines and jowls
0:14	Talking head (the after)

Later in the video: two more skin close-ups with arrows (1:00–1:04) and a smoother-skin "after" close-up (1:15).

Why the guess is not enough on its ownEvery item in that list is also in ads we run approved every day. To know which of them to cut, we compared our own rejected and approved scripts.

07 · What the rejection data says

The wording rules fire the same in rejected and approved ads. The visual rules do not.

Nov 3: the rules from the BANNING ADS doc turned into a script check and run blind on every task with a readable brief. 34 rejected scripts (of 77 rejected tasks) vs 190 approved (launched / scale / fail with no rejection recorded).

Rule	Rejected (34)	Approved (190)	Ratio	Read
#14 body close-up in the shot list	24% (8)	7% (13)	3.4×	separates
#15 before/after visual	29% (10)	13% (24)	2.3×	separates
#3 "Target" in the headline	29% (10)	17% (32)	1.8×	18 of 30 approved TPS carry it
#6 outcome promise	47% (16)	36% (68)	1.3×	weak
#5 "you" + body part	56% (19)	49% (94)	1.1×	no
#5 age / appearance address ("women over 60")	53% (18)	49% (94)	1.1×	no
#7 numbers and timeframes	76% (26)	78% (148)	1.0×	no
#3 NASA / toxins	6% (2)	15% (28)	0.4×	not a trigger

Inside TPS the picture is the sameBody close-ups are in 7 of 17 rejected TPS scripts vs 4 of 30 approved. It matches the "already tested" list: the fix that worked before was replacing close-ups with photography and wider shots.

Limits34 scripts is a small sample. Meta's own reason is known for 8 of 77 rejected tasks. 43 rejected tasks have no script (image tasks, iterations, account-level rejections). "Approved" means no rejection recorded in ClickUp.

The check now runs on every brief before anything is generated: HIGH = a body close-up or a before/after visual, MEDIUM = outcome promise, "you" + body, Target; LOW = numbers and guessed trigger words. VID863's original brief scores HIGH on both visual rules. The age lines and "death of Botox" that the first fix went after are MEDIUM at most.

Full table and variants: ai-video-kit / Skills/Video/reviewer/references/policy-calibration.md · summary doc shared with Diana Nov 3.

08 · Ban-fix test: three variants

Same ad, three rebuilds, one variable each. Meta's review is the result.

All three keep hook H3, avatar, voice (her clone from the final), and the same safe headline + ad text. Built Nov 3–4 from the original VID863 start frame, ~$18 total; all three go to test together. The question each one answers:

v1 · wording

Third person, same pictures

Script rewritten so she talks about "her" and "women like me" instead of "you" and "your jowls"; age callouts out. Visuals untouched, including the close-ups.

Tests: is it the words?

v2 · life stage

Same words, no close-ups

Hook and body word for word, but every skin close-up replaced with lifestyle B-roll of the same woman; no "look like this" shots.

Tests: is it the body close-ups (#14)?

v3 · then / now

Same words, no before/after

Her head and voice from VID863 untouched; the "before" selfie and arrows replaced with "then" stills where she keeps her real skin and only the moment changes, plus 12 s of VO. Re-edited in Resolve.

Tests: is it the before/after staging (#15)?

Why one hook and one copyIf a variant changes two things at once we learn nothing from the approval. More hooks only after one variant gets through. Launch all three on the same account (TPS14) so the review outcome is comparable; that is the ask for media buying.

Rules that came out of building these, now in the kit: a talking head keeps one start frame per character and set-up (never regenerated); in a then/now split the problem stays in the frame at medium shot, never retouched; no lander or article on a phone screen.

09 · Next

What happens with the answer

All three approved: the trigger was wording after all, or the review is noisier than the data suggests. Scale v1 (closest to the original) and keep the close-up rule as MEDIUM.
v2 and v3 approved, v1 rejected: visuals confirmed. Close-ups and before/after go HIGH in the check for every TPS brief; v2/v3 become the scale copies of VID863.
Only one approved: that rule is the trigger; the other is noise. Update the calibration and rebuild the avatar-iteration briefs (10.26) before they launch.
All rejected: the trigger is outside the script (lander, page, account history). Escalate with the three builds as evidence.

Every rejection from here adds a data point to the same table. The skill that built these three variants (ban-fix) is the deliverable of Sprint 21 task "Avoid ad ban".

Evgenii · Nov 4, 2026