People will tell you three. People will tell you fifty. Both are wrong because the question assumes the number matters more than the budget behind it. Here's what actually decides it.
The number follows the budget, not the other way around
Meta's delivery is budget-driven now. It doesn't need a big menu of ads to find winners, it needs enough spend per ad to test it. If you put 20 ads in an ad set on a Rs.1,000/day budget, each ad gets Rs.50. None of them get enough spend to prove anything. You've just created 20 ads stuck in limbo.
Think about what proving an ad actually requires. Each ad has to spend maybe one to two times its expected cost per result before you know if it works. If your cost per purchase is Rs.600 and you've spent Rs.5,000 on the ad set, twenty ads means each one saw Rs.250. At a Rs.600 cost per conversion, not a single one has earned a fair verdict. They're all innocent until proven guilty, and the jury went home.
The real constraint is this: every ad needs enough budget and time to exit learning and produce a clean read. If your ad set can't give each ad meaningful spend, cut the count. Fewer ads, more spend each, faster answers.
Rough call to make it usable: assume each ad needs roughly the spend of one to two expected conversions in its first week. Cost per purchase Rs.600, seven ads in the ad set, that's Rs.4,200 to Rs.8,400 of weekly spend that ad set needs to run to actually test those seven properly. If your ad set only spends Rs.3,000 a week, you can honestly test four to five. Count the budget, then count the ads.
What works at different budget levels
On a small budget, under Rs.3,000/day on an ad set, keep it tight. Two to five ads. Enough variety to test an angle or two, small enough that each gets real spend.
On a bigger budget, Rs.10,000/day or more, you can run more. Ten to twenty. The spend is there to test them properly and the volume keeps the algorithm fed with fresh options before fatigue sets in.
Why the fresh options matter at scale: a big-budget ad set burns through its best creative fast, and delivery has nothing to rotate when the hero ad fatigues. Ten to twenty decent ads let Meta rotate, so the ad set keeps producing while you develop the next generation. That's a pipeline argument, not a testing argument, and it only applies once the budget exists to support both.
The old advice of 60+ ads per ad set came from a phase when Meta's machine learning was less capable and needed a wide menu to work with. It's mostly obsolete. What matters now is conceptual diversity, a few genuinely different ideas, not sixty variations of the same hook. People quoting the old numbers are passing on advice from an auction that no longer exists.
The part that actually matters
Don't count ads. Count concepts. Three ads that are slight tweaks of each other teach you almost nothing. Three ads built on three genuinely different angles, the problem it solves, the objection it answers, the persona it speaks to, tell you which idea has legs.
Here's why one idea dressed six ways is a trap. Six hook variations of "our protein oats taste better than shakes" share a topic, an audience, and an objection. Whichever performs best wins the spend, the other five sit at near-zero delivery, and your read is one data point: this idea works, maybe. You've spent the whole ad set's testing capacity to learn a single thing. Three real concepts in the same spend teach you three things.
A workable mix for most ad sets: one concept aimed at the core problem, one aimed at the biggest objection, one aimed at a different segment or use case. Whatever budget you have, split by concept and let the algorithm fight.
The honest answer
One more constraint nobody budgets for
Ads inside an ad set don't just share spend, they compete for it. Meta doesn't pay equal attention to 15 ads, and at the top of a big ad set the system concentrates delivery on whoever looks best, which means your slow-burn concept never accumulates the impressions it needs. The workable fix when you want more ads than the budget supports: more ad sets, not more ads per set, each carrying a few concepts at enough spend apiece to get a real verdict. Same total budget, more lanes, each readable. Just remember the trade-off from the article before this one: more ad sets means more structures learning, and your total conversion volume has to cover all of them.
So the honest answer to "how many ads per ad set" is: as many distinct concepts as your budget can properly test, and no more. If the budget can't give each one a real shot, you have too many. If you only have variations of one idea, you have too few, no matter what the count says.
Check the budget first, the concept diversity second, and never the raw number someone swears by. The guy running three ads is right for his budget. The guy running fifteen is right for his. They're both running the same rule without knowing it: enough spend per genuinely different idea, and nothing more.
I'm Hirdesh Matai, founder of HDM GRO-ads, the science of e-commerce advertising.