Since late May, I have had a script save the Meta Ad Library pages for DraftKings, FanDuel, BetMGM, Hard Rock Bet, Fanatics and William Hill every night, which sounds obsessive until you see what falls out of it.
The library was built so regulators could track who pays for political ads; but it captures everything, including the accounts of the six companies that spend more on creative testing than the rest of the industry combined.
Watching those accounts night after night is the closest thing the business has to a free research lab. Over 12 weeks, the six brands ran at least 890 different creatives past me, roughly 10 new ones a night, with about 100 live at any moment.
Almost all of them died within days, and that churn is the whole point. An ad still running after several months has survived the most expensive A/B test in gambling, decided in real spend rather than a focus group. I started writing the survivors down, and they had more in common than I expected.
What the survivors look like
The oldest on my list is a BetMGM video that has run for more than four months: 15 seconds of slot reels turning, "Your $25 Bonus Awaits!" on screen from the opening frame, no dialogue to speak of. In that time, hundreds of slicker ads appeared beside it and were gone inside a fortnight.
Once you notice the shape, you see it everywhere. Every acquisition video that lasts runs between 15 and 24 seconds, most of them exactly 15, and every one opens on the offer rather than easing in with a scene.
They are built to be watched with the sound off, because that is how anyone actually watches a feed, so the on-screen text carries the message while the product does the performing: reels spinning, a game screen, a bet slip filling itself in.
Nobody is acting out a big win, and that absence is the real tell. Nothing with a storyline, a famous face or an obvious production budget lasted the quarter.
The Wizard of Oz effect
There is one more habit the casino survivors share, which is that they borrow their fame instead of building it. BetMGM's most durable video spots are all licensed slots, The Price Is Right, Wizard of Oz and Survivor, each with its own 15-second cut, several of them running for months.
The logic is just arithmetic. You cannot teach someone a new game in 15 seconds, but you do not need to teach them a game they already know, so the licence fee ends up buying the storytelling the ad has no time to do for itself.
A testing programme shaped like this now sits within reach of a mid-sized operator or a serious affiliate, which quietly changes who gets to find their own winners and who is left guessing at what works
The trade has argued for years about whether branded slots are worth what studios charge for them. I will leave the question of floor performance to someone else; but on paid social it looks settled, because the branded titles are the ones left standing.
The banner doesn't die
Video gets all the attention and yet the numbers keep pointing at the humble banner. The three oldest ads in the whole dataset are plain BetMGM images, a bonus figure and a logo and a compliance line, and when I stopped counting they had been live for 158, 157 and 132 days.
With static ads having an average age of 45 days against just 23 for video. The banner lasts because it is shelf stock, quietly restating the welcome offer to whoever wanders past, and an offer does not go stale the way a piece of creative does.
Video burns faster precisely because it is where the experimenting happens, and most of it exists to lose that experiment and be replaced. The banners never entered the contest; they were still there, saying the same thing, when the quarter ended.
Reading someone else's homework
The obvious objection is that all of this is American, built on American products under American rules, and that an ad lasting a long time tells you how long someone kept paying for it, not what it earned. Both points are fair.
Neither travels badly, though, because the things actually being tested are not American at all. Sound off viewing, a three-second attention span and an offer that has to land first behave the same in Rotterdam as they do in a living room in New Jersey, and only the compliance layer really changes when you cross a border.
That makes the whole pattern a free head start for anyone working a regulated market. The recipe is hardly a secret: 15 seconds, the offer in the first frame, legible with the sound off, the product doing the acting, and a familiar piece of IP behind it wherever you can license one.
The winners were dug out of a heap of dead creatives, and finding a handful that lasted took hundreds that did not. That burn rate used to mean an in-house team and an eight-figure budget; and it does not any more.
A testing programme shaped like this now sits within reach of a mid-sized operator or a serious affiliate, which quietly changes who gets to find their own winners and who is left guessing at what works.
The six biggest spenders in the industry have published a full quarter of their creative R&D in a public library, timestamped and free to read for anyone who bothers to look. It would be rude not to.