Creative Intelligence
Creative Intelligence Library
186 beauty & skincare ads broken into their 7-beat structure, hook type, psychology, and creative format. Updated as new winners land.
Motion's 2026 Creative Benchmark Report put a number on something every performance marketer already felt: only 4 to 6 percent of ads ever shipped to Meta or TikTok earn enough sustained spend to scale. The rest never clear the algorithm's quality bar long enough to find their audience. That 4 to 6 percent is the live training set for what actually works in 2026. The other 94 percent is noise.
What separates the winners is not a single trick. It is structure. The ads that scale share a small number of recurring architectures, executed with category-specific hooks and a known pacing rhythm. They open the same way, escalate the same way, validate the same way, and close the same way. The aesthetics are different. The architecture repeats. Once you can see the architecture, the same brand can ship a hundred winners instead of waiting for accidents.
This library is the curated set of those winners across Meta and TikTok in 2026. Every ad in it was either actively scaling at the time of decoding, or shipped on a structural pattern that continues to scale across the category. Each one is broken into the parts you can rebuild against your own product, your own audience, and your own offer.
Every ad in this archive runs through the same pipeline: a structural pass that segments the ad into a 7-beat architecture, a classification pass that labels each beat by subtype, and a context pass that pulls out the psychological mechanism, the marketing angle, the algorithm intent, and the creative format. The output is a machine-readable formula for that ad. Not a recap. Not a summary. A formula.
The 7 beats are Opening, Context, Tension, Delivery, Validation, Shift, and Close. An Opening can be a direct question, a pattern interrupt, a stat shock, a founder-to-camera confession, a B-roll cold cut, or one of roughly forty other classifiable subtypes. Each subtype carries different scaling math. A direct-question hook on a supplement brand has different three-second view-rate behavior than a stat-shock hook on a SaaS brand. Knowing which subtype you are reaching for is more useful than knowing the ad "had a good hook."
The Tension beat is where most scaling happens or dies. Tension is the moment the viewer has a question the ad has not yet answered. Strong ads escalate the tension before they pay it off. Weak ads pay off too fast and lose Meta's three-second view rate signal, which Andromeda treats as the strongest single creative quality input. Every decoded ad in this library carries its tension structure so you can copy the shape without copying the words.
Also in the shop
Browse decoded ads by category
You don't need another platform.
You need every ad to start from a formula that already works.
Pick your Heists. Build the stack that gets you there.
Build Your Stack7-day free trial.
Filter by category to find ads in your vertical. Filter again by product type to narrow into the exact sub-niche you are building for. Click any card to see the full structural breakdown, beat by beat, with the hook subtype, the psychological mission, and the creative format laid out as a recipe you can rebuild against your own brand.
The 15 product categories cover the bulk of D2C and B2B advertising on Meta and TikTok in 2026: Health and Supplements, Beauty and Skincare, Fashion and Apparel, Food and Beverage, Fitness, Home and Living, Tech and Gadgets, B2B and Software, Education and Courses, Kids and Parenting, Pet, Sports and Outdoors, Travel and Hospitality, Automotive, and Real Estate. Health and Supplements is the largest cluster (578 decoded ads) because it is also the most competitive paid acquisition category on Meta, and the volume there gives the structural patterns more statistical weight.
The library is not a swipe file. A swipe file gives you something to copy. A decoded library gives you something to rebuild. The difference matters because Meta's Entity ID system now flags ads that share visual or copy DNA with prior creative as the same entity, which means a literal copy of a winning ad fails to escape creative fatigue. A structural recreation, executed with your brand's own assets and your own copy against the same beat architecture, reads to Meta as a brand-new piece of creative and earns fresh exploration spend.
Each ad is selected because it cleared a measurable scaling bar at the time of decoding. The bar is a combination of Meta Ad Library active-time signals, public engagement floors, and confirmed campaign duration. We do not decode ads that disappeared within the first 72 hours, because those ads never gave Andromeda enough data to call them winners. We do not decode ads that ran only in test markets.
The decoding pipeline itself is multi-agent. The transcript and visual track are first segmented into the 7 structural categories. A second agent runs a subtype classification per beat against a 350-dimension taxonomy. A third agent extracts the linguistic cadence pattern: sentence length, rhetorical move sequence, syntactic stress patterns. A fourth agent pulls the psychological mission. The outputs assemble into a single structural formula that the Decoder Heist, Ad Script Heist, and Hooks Heist all read from when generating brand-fitted recreations.
The framework you see on each decoded ad page is the same framework an internal agent would see if it were generating a new ad based on this winner. That equivalence is the point. The library is not a gallery for browsing. It is the training surface for your own creative output.
Start by picking an ad that is winning in your category or in an adjacent one. Read its 7-beat structure. Notice the hook subtype, the tension escalation, the validation move. Then rebuild that structure against your own product. The Ad Script Heist will write a new script using the same architecture and your brand voice. The Hooks Heist will give you twenty variations of the opening beat. The Visual Studio Heist will produce the visual assets to match the format. The result is your own ad, on a structure that already cleared the 4 to 6 percent bar in your category.
The same library powers the weekly Trend Reports, the Category Deep Dives, and the Brand Decodes elsewhere on this site. Those surfaces aggregate the library into editorial reads on what is scaling. This page is the raw archive itself: every ad, every formula, browsable.
