What is an ad network?

What a network does

A network buys or contracts for publisher inventory, packages it into something advertisers want to buy — a category, an audience, a format — and resells it. It carries risk, sets prices, and earns the margin between what it pays and what it charges.

That risk-taking is the defining difference from an exchange, which matches bids to impressions and takes no position at all.

Why networks persist

Programmatic was widely expected to eliminate networks, and did not. They persist because aggregation is genuinely valuable in three cases.

Long-tail supply. Thousands of small publishers individually not worth an integration, collectively worth buying.

Curation. A network that has vetted its supply is selling something an open exchange cannot.

Simplicity. One contract, one creative spec, one report, no bidding operation to staff.

What to check before buying one

• Where the inventory actually comes from. Ask for a supply path and a publisher list.
• [Sub-publisher](/resources/glossary/sub-publishers) transparency, which is where traffic quality problems usually hide.
• Fraud posture. What is detected, what is refunded, and on whose evidence.
• Whether reporting is independently verifiable, or whether the network is marking its own homework — see biased attribution.

AI ad networks

The same logic applies with more force on AI surfaces, because the supply base is more fragmented. Beneath a handful of large assistants sit thousands of AI-native applications, each with small but highly intentional inventory that no advertiser would integrate with individually.

An AI ad network aggregates that into one buy. The harder part than in display is normalisation — formats, context signalling and eligibility rules differ genuinely between surfaces, so the network is reconciling real differences rather than repackaging identical banner slots.