How Meta decides who sees your ad
Every time someone scrolls Facebook or Instagram and there is space for an ad, Meta has to choose one ad out of millions. It does this in stages. A retrieval system called Andromeda narrows the candidates to a few thousand, ranking models predict how each would perform for that person, and an auction picks the ad with the highest total value.
Those stages explain things advertisers see every day. A lower bid can win. Two ads in the same ad set get very different spend. An ad with a low quality ranking costs more to run. The articles in this topic explain each stage from Meta's own documentation and mark where outside interpretation begins.
Start here
How does the Meta ad auction work?
Every ad impression on Facebook and Instagram goes to the ad with the highest total value, and the bid is only one part of that number.
7 min read
What is Meta Andromeda?
Andromeda decides which few thousand ads are worth scoring for each person, before ranking models and the auction pick a winner.
7 min read
Key terms
Common questions
- Why does Meta spend most of my budget on one ad?
- Meta shows each person the ad in your ad set that it predicts will get the lowest cost per optimisation event. It decides from predictions of future performance, not from each ad's past results. Meta says this means your ads won't necessarily be delivered the same number of times.
- Does Meta favour big advertisers in the ad auction?
- Meta says the highest bid does not always win, because estimated action rates and ad quality also count toward an ad's total value. An ad that is more relevant to a person can beat ads with higher bids. Meta's 2020 explainer said this lets businesses of all sizes compete.
- Where can I see how Meta rates my ads?
- Ads Manager reports three ad relevance diagnostics: quality ranking, engagement rate ranking and conversion rate ranking. Each compares your ad with ads competing for the same audience. They need at least 500 impressions, and Meta says they are not inputs into the auction.
- Does Meta use AI to decide which ads people see?
- Yes. Meta's delivery system uses machine learning to predict how likely each person is to take an advertiser's chosen action and to judge ad quality, and those predictions feed the auction. Meta has published details of several of these systems, including Andromeda for retrieval and GEM, a large model that improves its other ads models.
Put it to work
Winning Meta ads: what they are, where to find them and how to make your own
What a winning ad is, where to find winning Meta ads, and how to remake one.
How to test Facebook ad creative: a simple framework that works
What to test first, how many variations, how to judge, and how to keep the pipeline full.
CPM, CPC and CTR calculator
CPM, CPC, CTR and conversion rate from your Ads Manager numbers.
CPA calculator
Cost per acquisition against what you can afford to pay.
Other topics
Campaign structure
How campaigns, ad sets and Advantage+ automation fit together, and how the learning phase works.
Ad creative
What a static ad is, and why Meta favours ads that are genuinely different from each other.
Creative testing
How Meta's tests split people and budget, and how to tell a real winner from noise.
Metrics and attribution
What each Ads Manager number measures and how Meta decides which sales an ad gets credit for.
Creative fatigue
Why ads wear out, how Meta flags it and what frequency tells you about repetition.
Ad research
What the Meta Ad Library shows, what it hides, and what you may take from a competitor's ad.
AI ad creative
How AI ad generators work, where they fail, and what Meta requires of AI-made ads.