Glossary
What are cost models in advertising?
A cost model is the pricing structure that defines what an advertiser actually pays for: an impression, a click, an install, or a completed action. The choice determines who carries the risk between buyer and seller.
The models and what they price
- [CPM](/resources/glossary/cost-per-mille-cpm) — per thousand impressions. The advertiser pays for exposure and carries all the performance risk.
- [CPC](/resources/glossary/cost-per-click-cpc) — per click. Risk splits: the publisher carries the risk of the ad not being interesting, the advertiser carries everything after the click.
- [CPI](/resources/glossary/cost-per-install-cpi) — per app install. Most of the risk sits with the publisher.
- [CPA](/resources/glossary/cost-per-action-cpa) — per completed action. The advertiser pays only for outcomes; the publisher carries almost all the risk.
- Flat fee or sponsorship — a fixed price for a defined placement or period, with risk determined entirely by what was negotiated.
The model is a risk allocation, not a price
Every argument about cost models is really an argument about who absorbs uncertainty. Moving from CPM to CPA does not make media cheaper; it moves the risk of it not working from the buyer to the seller, and the seller prices that risk in.
Two consequences follow, and both get forgotten. Outcome-based pricing always carries a risk premium — a CPA that looks expensive against an equivalent CPM is usually charging you for the certainty. And a publisher will only accept outcome risk where they can predict the outcome, which means CPA is available on inventory that already converts well and unavailable exactly where you would most want it.
Which models fit AI inventory
CPM and CPC transfer directly and are what most AI inventory clears on today. CPM works because an answer insertion is a countable exposure; CPC works where the placement carries a link.
CPA is harder here than anywhere else, for a specific reason: a large share of AI-driven outcomes arrive with no click and no traceable path, so neither side can agree on which actions were caused. Publishers will not price against an outcome they cannot observe, and advertisers will not pay for one they cannot verify. It is coming, and it needs a measurement layer that does not exist yet.
Our pricing page sets out what is currently available, and what ChatGPT ads actually cost covers where benchmarks sit.
Choosing one
Buy on CPM when you want reach and have your own measurement — it is the cheapest per unit of exposure and you keep the upside. Buy on CPC when the click is a genuine step in your funnel rather than a vanity event. Buy on CPA when you value predictability more than efficiency and can accept the premium.
And compare like with like: an effective CPM computed from any model is the only figure that lets you rank inventory across sellers pricing differently.
Common questions
Which cost model is cheapest?
CPM, per unit of exposure, because you carry the risk. Whether it is cheapest per outcome depends entirely on whether the inventory works for you — which is what the other models charge a premium to guarantee.
Can I buy AI assistant inventory on CPA?
Rarely today. Too many outcomes arrive without an observable path for either side to agree what was caused. CPM and CPC are the practical options, with your own holdout supplying the outcome read.
More in pricing and return
What a campaign costs to run, and what it gives back.
- Ad revenue
- Ad spend
- Average revenue per daily active user (ARPDAU)
- Average revenue per user (ARPU)
- Cost aggregation
- Cost per action (CPA)
- Cost per click (CPC)
- Cost per install (CPI)
- Cost per mille (CPM)
- Customer acquisition cost (CAC)
- Installs per mille (IPM)
- Lifetime value (LTV)
- Monetization
- Offerwall
- Paywall
- Return on ad spend (ROAS)
- Return on investment (ROI)