How the ad auction works
What actually decides whether your ad is shown, and why CPM is an outcome rather than a setting.
Nobody buys placements any more. You enter an auction, and the auction runs once per impression, for every user, on every platform. Understanding what it optimizes explains most of what looks arbitrary in an ad account.
What the auction ranks
Every eligible ad is scored, and the highest score wins the impression. The score has three parts:
| Component | What it means | Who controls it |
|---|---|---|
| Bid | What you are willing to pay for the outcome | You, through bid strategy and budget |
| Estimated action rate | How likely this user is to do what you asked for | The platform's model, fed by your creative and pixel |
| Ad quality | Signals of user experience: engagement, hides, post-click behaviour | Your creative and landing page |
The consequence is that you do not win by bidding the most. You win by being the ad the platform expects this specific person to act on. Better creative lowers the bid needed to win the same impression.
Why you pay less than you bid
The winner generally pays what was needed to beat the next ad, not their full bid. This is why raising a bid often does not raise costs proportionally, and why an ad with strong action rates can sit at a much lower effective cost than a competitor bidding higher.
CPM is a result, not a price
CPM is often read as "what this audience costs". It is closer to a scoreboard:
| CPM moves up when | CPM moves down when |
|---|---|
| More advertisers compete for the same people | Competition thins out |
| Your ad is being hidden or scrolled past | Engagement and action rates are strong |
| Your targeting narrows to a small pool | The pool is broad and the model has room |
| Seasonal demand peaks (Q4, sale periods) | Off-peak periods |
A rising CPM with stable conversion rates is competition. A rising CPM with a falling click rate is usually creative fatigue.
The signal loop
The auction cannot estimate an action rate without examples. That is what the learning phase is: the platform buying exploratory impressions until it has enough conversions to predict reliably. Everything that fragments conversions — too many ad sets, too small budgets, a broken pixel — extends that period and raises what you pay for the same result. See campaign architecture for how to concentrate signal.
Two ads in the same ad set are not competing with each other for your budget in a fair fight. Delivery concentrates on whichever one the model believes will perform, often within hours. If you need a genuine read on two concepts, give them comparable conditions rather than assuming the split was even.