How manyadsshould an app run?

A founder opens the Meta Ad Library, types in the name of the biggest app in their category, and counts. ChatGPT: 580. TikTok: 531. Canva: 482. Netflix: 463. The conclusion writes itself: the winners run hundreds, so run hundreds.

The chart is what that conclusion looks like when you check it against 1,505 apps making $100k or more a month whose advertiser page we could match. Yes, the stairs go up. The median app with no ads takes 110k installs a month; with 10 to 24 creatives, 160k; with 400 or more, 600k. But look at the shape. Almost all of the climb happens in the first two steps. From 25 creatives to 400, the median app gets three times the installs for sixteen times the ads. Fit a line through all 1,492 apps and doubling the creative count goes with 11% more installs, and it explains 3% of the variance. Meta's own product page caps a campaign at fifty creatives and claims a 7% improvement from its automation. Nobody publishes the curve for what happens above fifty. This is that curve.

Why does ChatGPT need 580? Because the Ad Library counts variations, not ideas. An independent tracker that reads the same library breaks OpenAI's pile down: more than half video, a third carousel, eight languages, three product lines (the app, ChatGPT for Work, Codex), and 46 new creatives in a week. Hundreds live is languages times formats times products times weekly churn. Canva's pile is the same size and the opposite strategy: ten new creatives in a week, one ad that has run for 212 days. A long-lived, localised catalogue next to a fast test bench, and the library gives them the same number.

Then there is Esti. A photo editor with 470 live creatives and 200k installs a month. Aspect, an AI social app: 458 creatives, 90k installs. Fable, a book club app: 451, 120k. These are not failures; each makes $100k or more a month. But they are running ChatGPT's creative count at one four-hundredth of ChatGPT's scale, and that is the clearest way to see what the count measures. It measures how many ads you can afford to make. It does not measure what they buy.

An app with 1 to 9 creatives takes about 50k installs per creative. An app with 400 or more takes about 1k. This is arithmetic, not attribution: installs come from everywhere and we divide all of them by the ad count. But the industry's own numbers have the same shape. AppsFlyer read 1.1 million creatives from 1,300 apps and found the top 2% of creatives took 43% of non-gaming spend; the other 98% were tests. Singular's 2026 benchmark has the median advertiser launching 13 new creatives a week, the top quarter 53, and says in so many words that more creatives does not mean better click, conversion or return. Motion, on half a million Meta ads, found the hit rate is 5 to 8% at every spend tier, so the big accounts get more winners only because they buy more tickets.

The reason the pile has to keep growing is that each ad is a wasting asset. A benchmark of 368 Meta ads this spring put the median creative's life at 18 days; a quarter were dead in six. Meta's own guidance calls it fatigue when an ad's cost per result doubles. At that lifespan a 580-creative pool needs about thirty launches a day to stand still, which is roughly what OpenAI does. RevenueCat wrote up a founder testing 500 new creatives a day and called it the creative volume trap; Eric Seufert's version is that going from twenty variants of one idea to two hundred buys vanishingly little. The apps at 400 are not getting forty times the installs of the apps at 10. They are getting about 3.8 times, and paying for the rest in localisation, testing and the cost of being big.

Every band starts the year at the January peak and slides. The apps with no ads ended May at 0.84 of their January installs, a 16% fall; the apps with 1 to 9 did about the same. From ten creatives up, every band held between 0.92 and 1.00. The market did not fall in those months: Sensor Tower has non-game downloads up 4.9% on the year in the first quarter and flat in the second. What fell was the New Year's audience of apps with no paid floor under it.

There is now an experiment that says the same thing. Three researchers at MIT studied a large game developer that switched its ads off worldwide and found organic installs fell 20 to 30 percent, because paid installs hold store rankings and rankings bring the organic ones. Every hundred dollars bought about 32 installs and 2 more for free. Turning ads off does not remove the paid installs; it removes the floor. Whether the ads held the installs or the healthier apps kept advertising, our data cannot say. Both readings end in the same advice.

Cal AI is what the step looks like in a P&L. Its creator programme plateaued at about $2 million a month. The founders added paid on top, on Meta, TikTok and Instagram, and by January were spending over a million dollars a month on ads at $5.7 million in revenue. Two months later MyFitnessPal bought the company. Nobody has published how many creatives they ran. It did not matter. The number that mattered was the one before it, which was zero.

Are you at zero? Then start. The gap between none and ten creatives is the biggest step on the chart, the ad-free group is the one that lost installs this year, and the MIT result says the organic floor goes with the paid one.

Are you copying the leader's number? Stop. ChatGPT's 580 is eight languages, three products and thirty launches a day. Your category's median app runs 51 to 114. Start there, and read the leader's library for what its ads say, not how many there are.

Are you at fifty and thinking about a hundred? At the median, doubling buys 11%. The hit rate does not change with volume; the number of tickets does. Spend the difference on the idea inside the creative, because that is the only part of the count that has ever moved the installs.