What Meta's ad system rewards

Meta's page About the learning phase describes the period when its delivery system is still learning how an ad set may perform. Meta says ad sets exit learning once they can deliver stably, which usually occurs after about 50 results in the week after the ad set's last significant edit.

During learning, Meta says, ad sets are less stable and usually have a higher CPA. Learning is necessary, so the goal is to get through it, not avoid it.

Meta advises waiting to edit an ad set until it leaves learning, because editing resets learning. It also warns that edits which meaningfully change how an ad set might perform can send it back into learning.

Our read: decide what you are testing before launch, then leave it alone long enough to learn something.

Meta says that when you create many ads and ad sets, the system learns less about each one, and that combining similar ad sets also combines learnings. Splitting a small budget across many ad sets slows everything down.

Meta warns against very small or inflated budgets, and recommends a budget large enough to get enough total results. It also notes frequent budget changes can send an ad set back into learning.

A quick check: multiply your expected cost per result by 50. That is roughly what one ad set needs per week to leave learning.

Meta's Privacy Violations and Personal Attributes policy says ads must not assert or imply personal attributes, including health conditions, financial status, religion, age and more. Meta says ads should focus on the benefits of the product or service instead.

Meta's own examples: "Depression counseling" is allowed, while "Depression getting you down? Get help now." is not. "Do you have diabetes?" is not allowed, while "New diabetes treatment available" is.

Our read: this rule hits health, finance and personal services hardest. Write about what you offer, not about what you assume the reader has.

Meta's system rewards clarity: one goal, enough budget to reach about 50 results a week, few edits, combined audiences, and creative that describes the offer. Most "the algorithm hates us" weeks trace back to one of those.

What does "Learning limited" mean? Meta says the status appears when an ad set is not getting enough results to exit learning.

Do edits always reset learning? Meta says significant edits can, and lists edits that meaningfully change performance as the ones to avoid.

Can I target by health interests and mention the condition in the ad? Meta's policy bars ads that assert or imply a personal health condition.

Give Meta's system a clear goal, enough results to learn, and stable settings, and write ads about your offer instead of your audience's personal attributes.