The examples
20 creatives from 117 advertisers, as the public libraries returned them. Click any to open it here.
Benchmarks
| Company | Channels | Ads observed | Meta | Longest run | ||
|---|---|---|---|---|---|---|
 Lovable |    | 1,251 | 1,251 | 748 | 330 | 153d |
 Function Health | | 1,236 | 1,236 | 324 | 27 | 108d |
 Superpower | | 999 | 999 | 285 | 33 | 204d |
What the data shows
Video is the default, and the exceptions are deliberate
67 of the 117 advertisers with a readable Meta library run video in at least 60% of their creatives, and Lovable sits at 66%. That is not a stylistic preference so much as a production decision: a video pipeline that can cut one concept into feed, story and reel formats produces more testable variants per unit of creative work than static design does. The advertisers that break the pattern break it hard rather than partially. Gamma, Attio, Substack, Whop, Harvey, OpenEvidence, Gumloop, Polymarket, Oushk Pharmacy, Mosh, Moshy, Moshy NZ, Cloudways, BigCommerce run carousel-heavy libraries, which almost always signals an offer that needs several panels to explain, such as a template gallery or a multi-feature product tour. Between those two poles there is very little middle ground, which suggests format choice is being made once at the account level rather than tested creative by creative.
Launch cadence separates the testers from the maintainers
The sharpest division in this set is not size but freshness. 7 advertisers replaced at least half their visible library in the seven days before the scan, while 13 launched nothing at all. A high-churn library is a volume-testing posture: creative is produced faster than any human can judge it individually, and the account sorts winners in-flight on delivery signals. A static library of comparable size is the opposite. The testing already happened, the winners were chosen, and what remains live is a settled set being run for reach. Neither is better in the abstract, but they imply completely different creative budgets, and comparing an advertiser's raw ad count without checking its launch rate reliably misleads.
What the copy actually does
The hooks in this set cluster into a small number of shapes. Demonstrations open on the product doing something specific and measurable. Confessional first-person openings, the ones that read as a person rather than a brand, are almost exclusively used by consumer-adjacent tools, and almost always in vertical video. Claim-led openers put a number in the first six words. What is nearly absent across the whole set is the feature list: with a few carousel exceptions, no advertiser here opens by enumerating what the product has. The transferable part of a competitor's library is this structure, not the wording, and it survives even when the product does not resemble yours.
Read the destination, not just the creative
Landing-page paths are the most underrated field in the Meta Ad Library, because they reveal segmentation the creative hides. An advertiser sending every ad to one path is testing message against a single audience. One running four or more distinct paths is segmenting before the click, usually by persona or by market, and has built separate landing experiences to match. That is a materially more expensive operation and it shows up nowhere else in public data. UTM parameters survive in the library too, which means campaign naming conventions, and therefore account structure, are frequently readable straight off the ad.
Do this yourself
Sprites reads any advertiser’s Meta library the same way this page was built, then builds and runs your own campaigns against what it finds.
analyze and optimize Meta ads with AI →Every advertiser featured here
Run ads like these, without the team
Sprites reads competitor libraries the same way this page was built, then builds and runs your campaigns against what it finds.